Showing 1–50 of 54 results for author: Agarwala, A

Tunable Topological Phases in Quantum Kirigamis

Authors: Rahul Singh, Adhip Agarwala

Abstract: Advances in engineering mesoscopic quantum devices have led to new material platforms where electronic transport can be achieved on foldable structures. In this respect, we study quantum phases and their transitions on a Kirigami structure, a Japanese craft form, where its individual building blocks are topologically non-trivial. In particular, we find that by mechanically deforming the Kirigami s… ▽ More Advances in engineering mesoscopic quantum devices have led to new material platforms where electronic transport can be achieved on foldable structures. In this respect, we study quantum phases and their transitions on a Kirigami structure, a Japanese craft form, where its individual building blocks are topologically non-trivial. In particular, we find that by mechanically deforming the Kirigami structure one can engineer topological phase transitions in the system. Using a multipronged approach, we show that the physics of the system can be captured within a transfer matrix formalism akin to Chalker-Coddington networks, where the junctions describe scattering between effective non-Hermitian one-dimensional edge channels. We further show that the nature of the Kirigami structure can affect the critical folding angles where the topological phase transitions occur. Our study shows the rich interplay between topological quantum phenomena and structural configuration of an underlying Kirigami network reflecting its potential as an intriguing platform. △ Less

Submitted 20 June, 2024; originally announced June 2024.

Comments: 6+5 pages, 4+8 figures

A Clipped Trip: the Dynamics of SGD with Gradient Clip** in High-Dimensions

Authors: Noah Marshall, Ke Liang Xiao, Atish Agarwala, Elliot Paquette

Abstract: The success of modern machine learning is due in part to the adaptive optimization methods that have been developed to deal with the difficulties of training large models over complex datasets. One such method is gradient clip**: a practical procedure with limited theoretical underpinnings. In this work, we study clip** in a least squares problem under streaming SGD. We develop a theoretical a… ▽ More The success of modern machine learning is due in part to the adaptive optimization methods that have been developed to deal with the difficulties of training large models over complex datasets. One such method is gradient clip**: a practical procedure with limited theoretical underpinnings. In this work, we study clip** in a least squares problem under streaming SGD. We develop a theoretical analysis of the learning dynamics in the limit of large intrinsic dimension-a model and dataset dependent notion of dimensionality. In this limit we find a deterministic equation that describes the evolution of the loss. We show that with Gaussian noise clip** cannot improve SGD performance. Yet, in other noisy settings, clip** can provide benefits with tuning of the clip** threshold. In these cases, clip** biases updates in a way beneficial to training which cannot be recovered by SGD under any schedule. We conclude with a discussion about the links between high-dimensional clip** and neural network training. △ Less

Submitted 17 June, 2024; originally announced June 2024.

Pinning Disfavors Nucleation in Colloidal Vapor Deposition

Authors: Noman Hanif Barbhuiya, Pritam K. Mohanty, Saikat Mondal, Aminul Hussian, Adhip Agarwala, Chandan K. Mishra

Abstract: Vapor deposition, known for precise structural control, is inevitably influenced by impurities. These impurities, often distinct from the depositing material, can significantly impact material properties, including local structure. Interestingly, the effect of immobilizing some of the depositing particles themselves, which would still preserve local structural symmetry, remains largely unexplored.… ▽ More Vapor deposition, known for precise structural control, is inevitably influenced by impurities. These impurities, often distinct from the depositing material, can significantly impact material properties, including local structure. Interestingly, the effect of immobilizing some of the depositing particles themselves, which would still preserve local structural symmetry, remains largely unexplored. By introducing a small fraction of pinned sites during colloidal vapor deposition and employing a combination of thermodynamic and kinematic measurements, we demonstrate that pinned sites, termed as "mobility impurities," are disfavored as nucleation centers. Moreover, tuning the mobility of mobile colloids can adjust nucleation likelihood. In later stages of growth, pinning induces mode localization, altering the thin film's vibrational spectrum. Our research thus underscores the potential of strategically incorporating mobility impurities to engineer material properties. △ Less

Submitted 7 May, 2024; v1 submitted 30 April, 2024; originally announced April 2024.

Comments: 16 pages, 4 figures

High dimensional analysis reveals conservative sharpening and a stochastic edge of stability

Authors: Atish Agarwala, Jeffrey Pennington

Abstract: Recent empirical and theoretical work has shown that the dynamics of the large eigenvalues of the training loss Hessian have some remarkably robust features across models and datasets in the full batch regime. There is often an early period of progressive sharpening where the large eigenvalues increase, followed by stabilization at a predictable value known as the edge of stability. Previous work… ▽ More Recent empirical and theoretical work has shown that the dynamics of the large eigenvalues of the training loss Hessian have some remarkably robust features across models and datasets in the full batch regime. There is often an early period of progressive sharpening where the large eigenvalues increase, followed by stabilization at a predictable value known as the edge of stability. Previous work showed that in the stochastic setting, the eigenvalues increase more slowly - a phenomenon we call conservative sharpening. We provide a theoretical analysis of a simple high-dimensional model which shows the origin of this slowdown. We also show that there is an alternative stochastic edge of stability which arises at small batch size that is sensitive to the trace of the Neural Tangent Kernel rather than the large Hessian eigenvalues. We conduct an experimental study which highlights the qualitative differences from the full batch phenomenology, and suggests that controlling the stochastic edge of stability can help optimization. △ Less

Submitted 30 April, 2024; originally announced April 2024.

Entanglement signatures of a percolating quantum system

Authors: Subrata Pachhal, Adhip Agarwala

Abstract: Entanglement measures have emerged as one of the versatile probes to diagnose quantum phases and their transitions. Universal features in them expand their applicability to a range of systems, including those with quenched disorder. In this work, we show that when the underlying lattice has percolation disorder, free fermions at a finite density show interesting entanglement properties due to mass… ▽ More Entanglement measures have emerged as one of the versatile probes to diagnose quantum phases and their transitions. Universal features in them expand their applicability to a range of systems, including those with quenched disorder. In this work, we show that when the underlying lattice has percolation disorder, free fermions at a finite density show interesting entanglement properties due to massively degenerate ground states. We define and calculate appropriate entanglement measures such as typical, annealed, and quenched entanglement entropy in both one and two dimensions, showing they can capture both geometrical aspects and electronic correlations of the percolated quantum system. In particular, while typical and annealed entanglement show volume law character directly dependent on the number of zero modes in the system, quenched entanglement is generally area law albeit showing characteristic signatures of the classical percolation transition. Our work presents an exotic interplay between the geometrical properties of a lattice and quantum entanglement in a many-body quantum system. △ Less

Submitted 28 June, 2024; v1 submitted 22 March, 2024; originally announced March 2024.

Comments: 7+3 pages, 4+3 Figures

Evidence of enhanced thermopower from emergent local moments in flatbands of magic-angle twisted bilayer graphene

Authors: Ayan Ghosh, Souvik Chakraborty, Ranit Dutta, Adhip Agarwala, K. Watanabe, T. Taniguchi, Sumilan Banerjee, Nandini Trivedi, Subroto Mukerjee, Anindya Das

Abstract: Recent experiments on magic-angle twisted bilayer graphene (MATBLG) indicate an unusual coexistence of heavy and light electrons, characteristic of Heavy-Fermion physics. Yet, these experiments lack direct evidence of the existence of local moments associated with the heavy electrons, a key component of Heavy-Fermion physics. Thermopower serves as a sensitive probe for measuring entropy and can, t… ▽ More Recent experiments on magic-angle twisted bilayer graphene (MATBLG) indicate an unusual coexistence of heavy and light electrons, characteristic of Heavy-Fermion physics. Yet, these experiments lack direct evidence of the existence of local moments associated with the heavy electrons, a key component of Heavy-Fermion physics. Thermopower serves as a sensitive probe for measuring entropy and can, therefore, unveil the presence of local moments by assessing their impact on entropy. In this work, we have carried out comprehensive thermopower studies on MATBLG by varying temperatures ($T$) and magnetic fields ($B$). While the resistance exhibits prominent resistance peaks at integer fillings apart from the Dirac point (DP, $ν= 0$) and full band filling ($ν= \pm 4$), the thermopower remains featureless and symmetric to the DP with opposite signs, in complete violation of the Mott formula, even up to $\sim 100K$. This discrepancy indicates that resistance and thermopower are carried by different kinds of carriers. With the expected sign changes at the DP and full band filling, the thermopower demonstrates additional sign changes within the conduction and valence bands around $ν\sim \pm 1$, persisting from $5K$ to $50K$. These observations cannot be explained solely by thermopower arising from an emergent Heavy-Fermion band picture, which is highly sensitive to temperature. Instead, our data is consistent with the dominant contribution coming from the entropy of the localized moments of heavy electrons. To validate the claim, we studied the thermopower with $B_{\parallel}$ and $B_{\perp}$, revealing a $30\%$ and $50\%$ reduction, respectively, and attributed to the partial polarization of the local moments (spin/valley), resulting in decreased entropy. Our results highlight the thermopower contribution from local moments and establish the Heavy-Fermion physics in MATBLG. △ Less

Submitted 13 March, 2024; originally announced March 2024.

Quasicrystalline Spin Liquid

Authors: Sunghoon Kim, Mohammad Saad, Dan Mao, Adhip Agarwala, Debanjan Chowdhury

Abstract: The interplay of electronic interactions and frustration in crystalline systems leads to a panoply of correlated phases, including exotic Mott insulators with non-trivial patterns of entanglement. Disorder introduces additional quantum interference effects that can drive localization phenomena. Quasicrystals, which are neither disordered nor perfectly crystalline, are interesting playgrounds for s… ▽ More The interplay of electronic interactions and frustration in crystalline systems leads to a panoply of correlated phases, including exotic Mott insulators with non-trivial patterns of entanglement. Disorder introduces additional quantum interference effects that can drive localization phenomena. Quasicrystals, which are neither disordered nor perfectly crystalline, are interesting playgrounds for studying the effects of interaction, frustration, and quantum interference. Here we consider a solvable example of a quantum spin liquid on a tri-coordinated quasicrystal. We extend Kitaev's original construction for the spin model to our quasicrystalline setting and perform a large scale flux-sampling to find the ground-state configuration in terms of the emergent majorana fermions and flux excitations. This reveals a fully gapped quantum spin liquid, regardless of the exchange anisotropies, accompanied by a tendency towards non-trivial (de-)localization at the edge and the bulk. The advent of moiré materials and a variety of quantum simulators provide a new platform to bring phases of quasicrystalline quantum matter to life in a controlled fashion. △ Less

Submitted 12 February, 2024; originally announced February 2024.

Comments: 5 pages, 3 figures. Supplementary material: 5 pages, 7 figures

Feature learning as alignment: a structural property of gradient descent in non-linear neural networks

Authors: Daniel Beaglehole, Ioannis Mitliagkas, Atish Agarwala

Abstract: Understanding the mechanisms through which neural networks extract statistics from input-label pairs through feature learning is one of the most important unsolved problems in supervised learning. Prior works demonstrated that the gram matrices of the weights (the neural feature matrices, NFM) and the average gradient outer products (AGOP) become correlated during training, in a statement known as… ▽ More Understanding the mechanisms through which neural networks extract statistics from input-label pairs through feature learning is one of the most important unsolved problems in supervised learning. Prior works demonstrated that the gram matrices of the weights (the neural feature matrices, NFM) and the average gradient outer products (AGOP) become correlated during training, in a statement known as the neural feature ansatz (NFA). Through the NFA, the authors introduce map** with the AGOP as a general mechanism for neural feature learning. However, these works do not provide a theoretical explanation for this correlation or its origins. In this work, we further clarify the nature of this correlation, and explain its emergence. We show that this correlation is equivalent to alignment between the left singular structure of the weight matrices and the newly defined pre-activation tangent features at each layer. We further establish that the alignment is driven by the interaction of weight changes induced by SGD with the pre-activation features, and analyze the resulting dynamics analytically at early times in terms of simple statistics of the inputs and labels. Finally, motivated by the observation that the NFA is driven by this centered correlation, we introduce a simple optimization rule that dramatically increases the NFA correlations at any given layer and improves the quality of features learned. △ Less

Submitted 24 June, 2024; v1 submitted 7 February, 2024; originally announced February 2024.

Neglected Hessian component explains mysteries in Sharpness regularization

Authors: Yann N. Dauphin, Atish Agarwala, Hossein Mobahi

Abstract: Recent work has shown that methods like SAM which either explicitly or implicitly penalize second order information can improve generalization in deep learning. Seemingly similar methods like weight noise and gradient penalties often fail to provide such benefits. We show that these differences can be explained by the structure of the Hessian of the loss. First, we show that a common decomposition… ▽ More Recent work has shown that methods like SAM which either explicitly or implicitly penalize second order information can improve generalization in deep learning. Seemingly similar methods like weight noise and gradient penalties often fail to provide such benefits. We show that these differences can be explained by the structure of the Hessian of the loss. First, we show that a common decomposition of the Hessian can be quantitatively interpreted as separating the feature exploitation from feature exploration. The feature exploration, which can be described by the Nonlinear Modeling Error matrix (NME), is commonly neglected in the literature since it vanishes at interpolation. Our work shows that the NME is in fact important as it can explain why gradient penalties are sensitive to the choice of activation function. Using this insight we design interventions to improve performance. We also provide evidence that challenges the long held equivalence of weight noise and gradient penalties. This equivalence relies on the assumption that the NME can be ignored, which we find does not hold for modern networks since they involve significant feature learning. We find that regularizing feature exploitation but not feature exploration yields performance similar to gradient penalties. △ Less

Submitted 24 January, 2024; v1 submitted 19 January, 2024; originally announced January 2024.

Signatures of quantum phases in a dissipative system

Authors: Rohan Joshi, Saikat Mondal, Souvik Bandyopadhyay, Sourav Bhattacharjee, Adhip Agarwala

Abstract: Lindbladian formalism, as tuned to dissipative and open systems, has been all-pervasive to interpret non-equilibrium steady states of quantum many-body systems. We study the fate of free fermionic and superconducting phases in a dissipative one-dimensional Kitaev model - where the bath acts both as a source and a sink of fermionic particles with different coupling rates. As a function of these two… ▽ More Lindbladian formalism, as tuned to dissipative and open systems, has been all-pervasive to interpret non-equilibrium steady states of quantum many-body systems. We study the fate of free fermionic and superconducting phases in a dissipative one-dimensional Kitaev model - where the bath acts both as a source and a sink of fermionic particles with different coupling rates. As a function of these two couplings, we investigate the steady state, its entanglement content, and its approach from varying initial states. Interestingly, we find that the steady state phase diagram retains decipherable signatures of ground state critical physics. We also show that early-time fidelity is a useful marker to find a subclass of phase transitions in such situations. Moreover, we show that the survival of critical signatures at late-times, strongly depend on the thermal nature of the steady state. This connection hints at a correspondence between quantum observables and classical magnetism in the steady state of such systems. Our work uncovers interesting connections between dissipative quantum many-body systems, thermalization of a classical spin and many-body quantum critical phenomena. △ Less

Submitted 16 April, 2024; v1 submitted 28 December, 2023; originally announced December 2023.

Comments: 14 pages, 10 figures

Journal ref: Journal of Physics: Condensed Matter 36 (2024) 275601

On the Interplay Between Stepsize Tuning and Progressive Sharpening

Authors: Vincent Roulet, Atish Agarwala, Fabian Pedregosa

Abstract: Recent empirical work has revealed an intriguing property of deep learning models by which the sharpness (largest eigenvalue of the Hessian) increases throughout optimization until it stabilizes around a critical value at which the optimizer operates at the edge of stability, given a fixed stepsize (Cohen et al, 2022). We investigate empirically how the sharpness evolves when using stepsize-tuners… ▽ More Recent empirical work has revealed an intriguing property of deep learning models by which the sharpness (largest eigenvalue of the Hessian) increases throughout optimization until it stabilizes around a critical value at which the optimizer operates at the edge of stability, given a fixed stepsize (Cohen et al, 2022). We investigate empirically how the sharpness evolves when using stepsize-tuners, the Armijo linesearch and Polyak stepsizes, that adapt the stepsize along the iterations to local quantities such as, implicitly, the sharpness itself. We find that the surprisingly poor performance of a classical Armijo linesearch in the deterministic setting may be well explained by its tendency to ever-increase the sharpness of the objective. On the other hand, we observe that Polyak stepsizes operate generally at the edge of stability or even slightly beyond, outperforming its Armijo and constant stepsizes counterparts in the deterministic setting. We conclude with an analysis that suggests unlocking stepsize tuners requires an understanding of the joint dynamics of the step size and the sharpness. △ Less

Submitted 29 December, 2023; v1 submitted 30 November, 2023; originally announced December 2023.

Comments: Presented at the NeurIPS 2023 OPT Wokshop

Percolation Transition in a Topological Phase

Authors: Saikat Mondal, Subrata Pachhal, Adhip Agarwala

Abstract: Transition out of a topological phase is typically characterized by discontinuous changes in topological invariants along with bulk gap closings. However, as a clean system is geometrically punctured, it is natural to ask the fate of an underlying topological phase. To understand this physics we introduce and study both short and long-ranged toy models where a one dimensional topological phase is… ▽ More Transition out of a topological phase is typically characterized by discontinuous changes in topological invariants along with bulk gap closings. However, as a clean system is geometrically punctured, it is natural to ask the fate of an underlying topological phase. To understand this physics we introduce and study both short and long-ranged toy models where a one dimensional topological phase is subjected to bond percolation protocols. We find that non-trivial boundary phenomena follow competing energy scales even while global topological response is governed via geometrical properties of the percolated lattice. Using numerical, analytical and appropriate mean-field studies we uncover the rich phenomenology and the various cross-over regimes of these systems. In particular, we discuss emergence of "fractured topological region" where an overall trivial system contains macroscopic number of topological clusters. Our study shows the interesting physics that can arise from an interplay of geometrical disorder within a topological phase. △ Less

Submitted 14 December, 2023; v1 submitted 12 September, 2023; originally announced September 2023.

Comments: 6+4 pages, 7 figures

Journal ref: Phys. Rev. B 108, L220201 (2023)

Symmetry-Enriched Criticality in a Coupled Spin-Ladder

Authors: Suman Mondal, Adhip Agarwala, Tapan Mishra, Abhishodh Prakash

Abstract: We study a one-dimensional ladder of two coupled XXZ spin chains and identify several distinct gapless symmetry-enriched critical phases. These have the same unbroken symmetries and long-wavelength description, but cannot be connected without encountering either a phase transition or other intermediate phases. Using bosonizaion, we analyze the nature of their distinction by determining how microsc… ▽ More We study a one-dimensional ladder of two coupled XXZ spin chains and identify several distinct gapless symmetry-enriched critical phases. These have the same unbroken symmetries and long-wavelength description, but cannot be connected without encountering either a phase transition or other intermediate phases. Using bosonizaion, we analyze the nature of their distinction by determining how microscopic symmetries are manifested in the long-wavelength fields, the behavior of charged local and nonlocal operators, and identify the universality class of all direct continuous phase transitions between them. One of these phases is a gapless topological phase with protected edge modes. We characterize its precise nature and place it within the broader classification. We also find the occurrence of `multiversality' in the phase diagram, wherein two fixed phases are separated by continuous transitions with different universality classes in different parameter regimes. We determine the phase diagram and all its aspects, as well as verify our predictions numerically using density matrix renormalization group and a map** onto an effective spin-1 model. △ Less

Submitted 29 April, 2024; v1 submitted 8 September, 2023; originally announced September 2023.

Comments: v2: minor improvements, added references; 26 pages, 16 figures, 7 tables

Journal ref: Phys. Rev. B 108, 245135 (2023)

Dimensional reduction of Kitaev spin liquid at quantum criticality

Authors: Shi Feng, Adhip Agarwala, Nandini Trivedi

Abstract: We investigate the fate of the Kitaev spin liquid (KSL) under the influence of an external magnetic field $h$ in the [001] direction and upon tuning bond anisotropy of the Kitaev coupling $K_z$ kee** $K_x = K_y = K$. Guided by density matrix renormalization group, exact diagonalization, and with insights from parton mean field theory, we uncover a field-induced gapless-to-gapless Lifshitz transi… ▽ More We investigate the fate of the Kitaev spin liquid (KSL) under the influence of an external magnetic field $h$ in the [001] direction and upon tuning bond anisotropy of the Kitaev coupling $K_z$ kee** $K_x = K_y = K$. Guided by density matrix renormalization group, exact diagonalization, and with insights from parton mean field theory, we uncover a field-induced gapless-to-gapless Lifshitz transition from the nodal KSL to an intermediate gapless phase. The intermediate phase sandwiched between $h_{c1}$ and $h_{c2}$, which persists for a wide range of anisotropy $K_z/K > 0$, is composed of weakly coupled one-dimensional quantum critical chains. This intermediate phase is a dimensional crossover which asymptotically leads to the one-dimensional quantum Ising criticality characterized by the (1+1)D conformal field theory as the field reaches the phase transition at $h_{c2}$. Beyond $h_{c2}$ the system enters a partially polarized phase describable as effectively decoupled bosonic chains in which spin waves propagate along the one-dimensional zigzag direction. Our findings provide a comprehensive phase diagram and offer insights into the unusual physics of dimensional reduction generated by a uniform magnetic field in an otherwise two-dimensional quantum spin liquid. △ Less

Submitted 18 March, 2024; v1 submitted 15 August, 2023; originally announced August 2023.

Comments: 14 pages, 9 figures

Journal ref: Phys. Rev. Research 6, 013298 (2024)

Domain specificity and data efficiency in typo tolerant spell checkers: the case of search in online marketplaces

Authors: Dayananda Ubrangala, Juhi Sharma, Ravi Prasad Kondapalli, Kiran R, Amit Agarwala, Laurent Boué

Abstract: Typographical errors are a major source of frustration for visitors of online marketplaces. Because of the domain-specific nature of these marketplaces and the very short queries users tend to search for, traditional spell cheking solutions do not perform well in correcting typos. We present a data augmentation method to address the lack of annotated typo data and train a recurrent neural network… ▽ More Typographical errors are a major source of frustration for visitors of online marketplaces. Because of the domain-specific nature of these marketplaces and the very short queries users tend to search for, traditional spell cheking solutions do not perform well in correcting typos. We present a data augmentation method to address the lack of annotated typo data and train a recurrent neural network to learn context-limited domain-specific embeddings. Those embeddings are deployed in a real-time inferencing API for the Microsoft AppSource marketplace to find the closest match between a misspelled user query and the available product names. Our data efficient solution shows that controlled high quality synthetic data may be a powerful tool especially considering the current climate of large language models which rely on prohibitively huge and often uncontrolled datasets. △ Less

Submitted 3 August, 2023; originally announced August 2023.

Journal ref: Microsoft Journal of Applied Research, Volume 19, 2023

Correlation-driven non-trivial phases in single bi-layer Kagome intermetallics

Authors: Aabhaas Vineet Mallik, Adhip Agarwala, Tanusri Saha-Dasgupta

Abstract: Bi-layer Kagome compounds provide an exciting playground where the interplay of topology and strong correlations can give rise to exotic phases of matter. Motivated by recent first principles calculation on such systems (Phys. Rev. Lett 125, 026401), reporting stabilization of a Chern metal with topological nearly-flat band close to Fermi level, we build minimal models to study the effect of stron… ▽ More Bi-layer Kagome compounds provide an exciting playground where the interplay of topology and strong correlations can give rise to exotic phases of matter. Motivated by recent first principles calculation on such systems (Phys. Rev. Lett 125, 026401), reporting stabilization of a Chern metal with topological nearly-flat band close to Fermi level, we build minimal models to study the effect of strong electron-electron interactions on such a Chern metal. Using approriate numerical and analytical techniques, we show that the topologically non-trivial bands present in this system at the Fermi energy can realize fractional Chern insulator states. We further show that if the time-reversal symmetry is restored due to destruction of magnetism by low dimensionality and fluctuation, the system can realize a superconducting phase in the presence of strong local repulsive interactions. Furthermore, we identify an interesting phase transition from the superconducting phase to a correlated metal by tuning nearest-neighbor repulsion. Our study uncovers a rich set of non-trivial phases realizable in this system, and contextualizes the physically meaningful regimes where such phases can be further explored. △ Less

Submitted 30 June, 2023; originally announced June 2023.

Comments: 16 pages, 14 figures

Journal ref: Phys. Rev. B 108, 205125, 2023

Spectral Form Factors of Topological Phases

Authors: Anurag Sarkar, Subrata Pachhal, Adhip Agarwala, Diptarka Das

Abstract: Signatures of dynamical quantum phase transitions and chaos can be found in the time evolution of generalized partition functions such as spectral form factors (SFF) and Loschmidt echoes. While a lot of work has focused on the nature of such systems in a variety of strongly interacting quantum theories, in this work, we study their behavior in short-range entangled topological phases, particularly… ▽ More Signatures of dynamical quantum phase transitions and chaos can be found in the time evolution of generalized partition functions such as spectral form factors (SFF) and Loschmidt echoes. While a lot of work has focused on the nature of such systems in a variety of strongly interacting quantum theories, in this work, we study their behavior in short-range entangled topological phases, particularly focusing on the role of symmetry-protected topological zero modes. We show, using both analytical and numerical methods, how the existence of such zero modes in any representative system can mask the SFF with large period (akin to generalized Rabi) oscillations, hiding any behavior arising from the bulk of the spectrum. Moreover, in a quenched disordered system, these zero modes fundamentally change the late-time universal behavior reflecting the chaotic signatures of the zero-energy manifold. Our study uncovers the rich physics underlying the interplay of chaotic signatures and topological characteristics in a quantum system. △ Less

Submitted 11 April, 2024; v1 submitted 22 June, 2023; originally announced June 2023.

Comments: Expanded and published version : 8+4 pages, 10 figures

Journal ref: Phys. Rev. B 109, 155126 (2024)

Classical and quantum facilitated exclusion processes

Authors: Amit Kumar Chatterjee, Adhip Agarwala

Abstract: We demonstrate exciting similarities between classical and quantum many body systems whose microscopic dynamics are composed of non-reciprocal three-site facilitated exclusion processes. We show that the quantum analogue of the classical facilitated process engineers an interesting $quantum$ $absorbing$ $transition$ where the quantum particles transit from an unentangled direct-product absorbing p… ▽ More We demonstrate exciting similarities between classical and quantum many body systems whose microscopic dynamics are composed of non-reciprocal three-site facilitated exclusion processes. We show that the quantum analogue of the classical facilitated process engineers an interesting $quantum$ $absorbing$ $transition$ where the quantum particles transit from an unentangled direct-product absorbing phase to an entangled steady state with a finite current at density $ρ=1/2$. In the generalised classical facilitated exclusion process, which includes independent hop** of particles with rate $p$, our analytical and Monte-Carlo results establish emergence of a special density $ρ^*=1/3$ that demarcates two regimes in the steady state, based on the competition between two current carrying modes (facilitated and independent). The corresponding quantum system also displays similar qualitative behaviours with striking non-monotonic features in the bipartite entanglement. Our work ties the two sub-fields of classically interacting exclusion processes, and interacting non-Hermitian quantum Hamiltonians to show common themes in the non-equilibrium phases they realise. △ Less

Submitted 27 February, 2023; v1 submitted 17 February, 2023; originally announced February 2023.

Comments: 10 pages, 5 figures, typos corrected

MSC Class: 60G10

SAM operates far from home: eigenvalue regularization as a dynamical phenomenon

Authors: Atish Agarwala, Yann N. Dauphin

Abstract: The Sharpness Aware Minimization (SAM) optimization algorithm has been shown to control large eigenvalues of the loss Hessian and provide generalization benefits in a variety of settings. The original motivation for SAM was a modified loss function which penalized sharp minima; subsequent analyses have also focused on the behavior near minima. However, our work reveals that SAM provides a strong r… ▽ More The Sharpness Aware Minimization (SAM) optimization algorithm has been shown to control large eigenvalues of the loss Hessian and provide generalization benefits in a variety of settings. The original motivation for SAM was a modified loss function which penalized sharp minima; subsequent analyses have also focused on the behavior near minima. However, our work reveals that SAM provides a strong regularization of the eigenvalues throughout the learning trajectory. We show that in a simplified setting, SAM dynamically induces a stabilization related to the edge of stability (EOS) phenomenon observed in large learning rate gradient descent. Our theory predicts the largest eigenvalue as a function of the learning rate and SAM radius parameters. Finally, we show that practical models can also exhibit this EOS stabilization, and that understanding SAM must account for these dynamics far away from any minima. △ Less

Submitted 16 February, 2023; originally announced February 2023.

Second-order regression models exhibit progressive sharpening to the edge of stability

Authors: Atish Agarwala, Fabian Pedregosa, Jeffrey Pennington

Abstract: Recent studies of gradient descent with large step sizes have shown that there is often a regime with an initial increase in the largest eigenvalue of the loss Hessian (progressive sharpening), followed by a stabilization of the eigenvalue near the maximum value which allows convergence (edge of stability). These phenomena are intrinsically non-linear and do not happen for models in the constant N… ▽ More Recent studies of gradient descent with large step sizes have shown that there is often a regime with an initial increase in the largest eigenvalue of the loss Hessian (progressive sharpening), followed by a stabilization of the eigenvalue near the maximum value which allows convergence (edge of stability). These phenomena are intrinsically non-linear and do not happen for models in the constant Neural Tangent Kernel (NTK) regime, for which the predictive function is approximately linear in the parameters. As such, we consider the next simplest class of predictive models, namely those that are quadratic in the parameters, which we call second-order regression models. For quadratic objectives in two dimensions, we prove that this second-order regression model exhibits progressive sharpening of the NTK eigenvalue towards a value that differs slightly from the edge of stability, which we explicitly compute. In higher dimensions, the model generically shows similar behavior, even without the specific structure of a neural network, suggesting that progressive sharpening and edge-of-stability behavior aren't unique features of neural networks, and could be a more general property of discrete learning algorithms in high-dimensional non-linear models. △ Less

Submitted 10 October, 2022; originally announced October 2022.

Deep equilibrium networks are sensitive to initialization statistics

Authors: Atish Agarwala, Samuel S. Schoenholz

Abstract: Deep equilibrium networks (DEQs) are a promising way to construct models which trade off memory for compute. However, theoretical understanding of these models is still lacking compared to traditional networks, in part because of the repeated application of a single set of weights. We show that DEQs are sensitive to the higher order statistics of the matrix families from which they are initialized… ▽ More Deep equilibrium networks (DEQs) are a promising way to construct models which trade off memory for compute. However, theoretical understanding of these models is still lacking compared to traditional networks, in part because of the repeated application of a single set of weights. We show that DEQs are sensitive to the higher order statistics of the matrix families from which they are initialized. In particular, initializing with orthogonal or symmetric matrices allows for greater stability in training. This gives us a practical prescription for initializations which allow for training with a broader range of initial weight scales. △ Less

Submitted 19 July, 2022; originally announced July 2022.

Anyon dynamics in field-driven phases of the anisotropic Kitaev model

Authors: Shi Feng, Adhip Agarwala, Subhro Bhattacharjee, Nandini Trivedi

Abstract: The Kitaev model on a honeycomb lattice with bond-dependent Ising interactions offers an exactly solvable model of a quantum spin liquid (QSL) with gapped $Z_2$ fluxes and gapless linearly dispersing Majorana fermions in the isotropic limit ($K_x=K_y=K_z$). We explore the phase diagram along two axes, an external magnetic field, $h$, applied out-of-plane of the honeycomb, and anisotropic interacti… ▽ More The Kitaev model on a honeycomb lattice with bond-dependent Ising interactions offers an exactly solvable model of a quantum spin liquid (QSL) with gapped $Z_2$ fluxes and gapless linearly dispersing Majorana fermions in the isotropic limit ($K_x=K_y=K_z$). We explore the phase diagram along two axes, an external magnetic field, $h$, applied out-of-plane of the honeycomb, and anisotropic interactions, $K_z$ larger than the other two. For $K_z/K\gg 2$ and $h=0$, the matter Majorana fermions have the largest gap, and the system is described by a gapped $Z_2$ Toric code in which the $Z_2$ fluxes form the low energy bosonic Ising electric (e) and magnetic (m) charges along with their fermionic bound state $ε=e\times m$. In this regime, we find that a small out-of-plane magnetic field creates $ε$ fermions that disperse in fixed one-dimensional directions before the transition to a valence bond solid phase, providing a direct dynamical signature of low energy $Z_2$ Abelian flux excitations separated from the Majorana sector. At lower $K_z$ in the center of the Abelian phase, in a regime we dub the primordial fractionalized (PF) regime, the field generates a hybridization between the $ε$ fermions and the Majorana matter fermions, resulting in a $ψ$ fermion. All the other phases in the field-anisotropy plane are naturally obtained from this primordial soup. We show that in the $Z_2$ Abelian phase, including the PF regime, the dynamical structure factors of local spin flip operators reveal distinct peaks that can be identified as arising from different anyonic excitations. We present in detail their signatures in energy and momentum and propose their identification by inelastic light scattering or inelastic polarized neutron scattering as ``smoking gun" signatures of fractionalization in the QSL phase. △ Less

Submitted 31 July, 2023; v1 submitted 26 June, 2022; originally announced June 2022.

Comments: 19 pages, 8 figures, 1 table

Journal ref: Phys. Rev. B 108, 035149 (2023)

Neural Volumetric Object Selection

Authors: Zhongzheng Ren, Aseem Agarwala, Bryan Russell, Alexander G. Schwing, Oliver Wang

Abstract: We introduce an approach for selecting objects in neural volumetric 3D representations, such as multi-plane images (MPI) and neural radiance fields (NeRF). Our approach takes a set of foreground and background 2D user scribbles in one view and automatically estimates a 3D segmentation of the desired object, which can be rendered into novel views. To achieve this result, we propose a novel voxel fe… ▽ More We introduce an approach for selecting objects in neural volumetric 3D representations, such as multi-plane images (MPI) and neural radiance fields (NeRF). Our approach takes a set of foreground and background 2D user scribbles in one view and automatically estimates a 3D segmentation of the desired object, which can be rendered into novel views. To achieve this result, we propose a novel voxel feature embedding that incorporates the neural volumetric 3D representation and multi-view image features from all input views. To evaluate our approach, we introduce a new dataset of human-provided segmentation masks for depicted objects in real-world multi-view scene captures. We show that our approach out-performs strong baselines, including 2D segmentation and 3D segmentation approaches adapted to our task. △ Less

Submitted 30 May, 2022; originally announced May 2022.

Comments: CVPR 2022 camera ready

Fractionalization and topology in amorphous electronic solids

Authors: Sunghoon Kim, Adhip Agarwala, Debanjan Chowdhury

Abstract: Band-topology is traditionally analyzed in terms of gauge-invariant observables associated with crystalline Bloch wavefunctions. Recent work has demonstrated that many of the free fermion topological characteristics survive even in an amorphous setting. In this work, we extend these studies to incorporate the effect of strong repulsive interactions on the fate of topology and other correlation ind… ▽ More Band-topology is traditionally analyzed in terms of gauge-invariant observables associated with crystalline Bloch wavefunctions. Recent work has demonstrated that many of the free fermion topological characteristics survive even in an amorphous setting. In this work, we extend these studies to incorporate the effect of strong repulsive interactions on the fate of topology and other correlation induced phenomena. Using a parton-based approach, we obtain the interacting phase diagram for an electronic two-orbital model with tunable topology in a two dimensional amorphous network. In addition to the (non-)topological phases that are adiabatically connected to the free fermion limit, we find a number of strongly interacting amorphous analogs of crystalline Mott insulating phases with non-trivial chiral neutral edge modes, and a fractionalized Anderson insulating phase. The amorphous networks thus provide a new playground for studying a plethora of exotic states of matter, and their glassy dynamics, due to the combined effects of non-trivial topology, disorder, and strong interactions. △ Less

Submitted 19 January, 2023; v1 submitted 23 May, 2022; originally announced May 2022.

Comments: 5 pages, 4 figures. Supplementary material: 3 pages, 5 figures

Journal ref: Phys. Rev. Lett. 130, 026202 (2023)

Chiral metals and entrapped insulators in a one-dimensional topological non-Hermitian system

Authors: Ayan Banerjee, Suraj S. Hegde, Adhip Agarwala, Awadhesh Narayan

Abstract: In this work we study many-body 'steady states' that arise in the non-Hermitian generalisation of the non-interacting Su-Schrieffer-Heeger model at a finite density of fermions. We find that the hitherto known phase diagrams for this system, derived from the single-particle gap closings, in fact correspond to distinct non-equilibrium phases, which either carry finite currents or are dynamical insu… ▽ More In this work we study many-body 'steady states' that arise in the non-Hermitian generalisation of the non-interacting Su-Schrieffer-Heeger model at a finite density of fermions. We find that the hitherto known phase diagrams for this system, derived from the single-particle gap closings, in fact correspond to distinct non-equilibrium phases, which either carry finite currents or are dynamical insulators where particles are entrapped. Each of these have distinct quasi-particle excitations and steady state correlations and entanglement properties. Looking at finite-sized systems, we further modulate the boundary to uncover the topological features in such steady states -- in particular the emergence of leaky boundary modes. Using a variety of analytical and numerical methods we develop a theoretical understanding of the various phases and their transitions, and uncover the rich interplay of non-equilibrium many-body physics, quantum entanglement and topology in a simple looking, yet a rich model system. △ Less

Submitted 14 May, 2022; v1 submitted 3 November, 2021; originally announced November 2021.

Comments: Published version

Journal ref: Phys. Rev. B 105, 205403 (2022)

Interaction driven giant thermopower in magic-angle twisted bilayer graphene

Authors: Arup Kumar Paul, Ayan Ghosh, Souvik Chakraborty, Ujjal Roy, Ranit Dutta, K. Watanabe, T. Taniguchi, Animesh Panda, Adhip Agarwala, Subroto Mukerjee, Sumilan Banerjee, Anindya Das

Abstract: Magic-angle twisted bilayer graphene (MtBLG) has proven to be an extremely promising new platform to realize and study a host of emergent quantum phases arising from the strong correlations in its narrow bandwidth flat band. In this regard, thermal transport phenomena like thermopower, in addition to being coveted technologically, is also sensitive to the particle-hole (PH) asymmetry, making it a… ▽ More Magic-angle twisted bilayer graphene (MtBLG) has proven to be an extremely promising new platform to realize and study a host of emergent quantum phases arising from the strong correlations in its narrow bandwidth flat band. In this regard, thermal transport phenomena like thermopower, in addition to being coveted technologically, is also sensitive to the particle-hole (PH) asymmetry, making it a crucial tool to probe the underlying electronic structure of this material. We have carried out thermopower measurements of MtBLG as a function of carrier density, temperature and magnetic field, and report the observation of an unusually large thermopower reaching up to a value as high as $\sim \bf{100μV/K}$ at a low temperature of 1K. Surprisingly, our observed thermopower exhibiting peak-like features in close correspondence to the resistance peaks around the integer Moire fillings, including the Dirac Point, violating the Mott formula. %Surprisingly, our observed thermopower exhibits peak-like features in close correspondence to the resistance peaks around the integer Moire fillings, including the Dirac Point, which completely violates the Mott formula. We show that the large thermopower peaks and their %non-monotonic dependence with temperature and magnetic field associated behaviour arise from the emergent highly PH asymmetric electronic structure due to the cascade of Dirac revivals. Furthermore, the thermopower shows an anomalous peak around the superconducting transition on the hole side and points towards the possible role of enhanced superconducting fluctuations in MtBLG. △ Less

Submitted 10 September, 2021; v1 submitted 1 September, 2021; originally announced September 2021.

Exploring ideas in topological quantum phenomena: A journey through the SSH model

Authors: Anantha Hegde, Adarsh Kumar, Adhip Agarwala, Bhaskaran Muralidharan

Abstract: Geared as an invitation for undergraduates, beginning graduate students, we present a pedagogical introduction to one-dimensional topological phases -- in particular the Su-Schrieffer-Heeger model. In the process, we delve upon ideas of entanglement using the correlator method and the von-Neumann density-matrix method, geometric phase, polarization, transport signatures and the role of electron-el… ▽ More Geared as an invitation for undergraduates, beginning graduate students, we present a pedagogical introduction to one-dimensional topological phases -- in particular the Su-Schrieffer-Heeger model. In the process, we delve upon ideas of entanglement using the correlator method and the von-Neumann density-matrix method, geometric phase, polarization, transport signatures and the role of electron-electron interactions. Through hands-on numerical experiments, whose $\href{https://github.com/hnoend/SSH_codes}{codes}$ are shared, we try to drive home the message why a program of simulating quantum electronics with topological toy models is the store house for discovering fantastic physics ideas. △ Less

Submitted 3 August, 2021; originally announced August 2021.

Comments: Pedagogical Article, 68 pages, 24 figures

Phases and Quantum Phase Transitions in Anisotropic Antiferromagnetic Kitaev-Heisenberg-$Γ$ magnet

Authors: Animesh Nanda, Adhip Agarwala, Subhro Bhattacharjee

Abstract: We study the Kitaev-Heisenberg-$Γ$ model with antiferromagnetic Kitaev exchanges in the strong anisotropic (toric code) limit to understand the phases and the intervening phase transitions between the gapped $Z_2$ quantum spin liquid and the spin-ordered (in the Heisenberg limit) as well as paramagnetic phases (in the pseudo-dipolar, $Γ$, limit). We find that the paramagnetic phase obtained in the… ▽ More We study the Kitaev-Heisenberg-$Γ$ model with antiferromagnetic Kitaev exchanges in the strong anisotropic (toric code) limit to understand the phases and the intervening phase transitions between the gapped $Z_2$ quantum spin liquid and the spin-ordered (in the Heisenberg limit) as well as paramagnetic phases (in the pseudo-dipolar, $Γ$, limit). We find that the paramagnetic phase obtained in the large $Γ$ limit has no topological entanglement entropy and is proximate to a gapless critical point of a system described by an equal superposition of differently oriented stacked one-dimensional $Z_2\times Z_2$ symmetry protected topological phases. Using a combination of exact diagonalization calculations and field-theoretic analysis we map out the phases and phase transitions to reveal the complete phase diagram as a function of the Heisenberg, the Kitaev, and the pseudo-dipolar interactions. Our work shows a rich plethora of unconventional phases and phase transitions and provides a comprehensive understanding of the physics of anisotropic Kitaev-Heisenberg-$Γ$ systems along with our recent paper [Phys. Rev. B 102, 235124 (2020)] where the ferromagnetic Kitaev exchange was studied. △ Less

Submitted 20 July, 2021; originally announced July 2021.

Comments: 33 pages, 29 figures, Companion paper to arXiv:2006.10081

One Network Fits All? Modular versus Monolithic Task Formulations in Neural Networks

Authors: Atish Agarwala, Abhimanyu Das, Brendan Juba, Rina Panigrahy, Vatsal Sharan, Xin Wang, Qiuyi Zhang

Abstract: Can deep learning solve multiple tasks simultaneously, even when they are unrelated and very different? We investigate how the representations of the underlying tasks affect the ability of a single neural network to learn them jointly. We present theoretical and empirical findings that a single neural network is capable of simultaneously learning multiple tasks from a combined data set, for a vari… ▽ More Can deep learning solve multiple tasks simultaneously, even when they are unrelated and very different? We investigate how the representations of the underlying tasks affect the ability of a single neural network to learn them jointly. We present theoretical and empirical findings that a single neural network is capable of simultaneously learning multiple tasks from a combined data set, for a variety of methods for representing tasks -- for example, when the distinct tasks are encoded by well-separated clusters or decision trees over certain task-code attributes. More concretely, we present a novel analysis that shows that families of simple programming-like constructs for the codes encoding the tasks are learnable by two-layer neural networks with standard training. We study more generally how the complexity of learning such combined tasks grows with the complexity of the task codes; we find that combining many tasks may incur a sample complexity penalty, even though the individual tasks are easy to learn. We provide empirical support for the usefulness of the learning bounds by training networks on clusters, decision trees, and SQL-style aggregation. △ Less

Submitted 28 March, 2021; originally announced March 2021.

Comments: 30 pages, 6 figures

Temperature check: theory and practice for training models with softmax-cross-entropy losses

Authors: Atish Agarwala, Jeffrey Pennington, Yann Dauphin, Sam Schoenholz

Abstract: The softmax function combined with a cross-entropy loss is a principled approach to modeling probability distributions that has become ubiquitous in deep learning. The softmax function is defined by a lone hyperparameter, the temperature, that is commonly set to one or regarded as a way to tune model confidence after training; however, less is known about how the temperature impacts training dynam… ▽ More The softmax function combined with a cross-entropy loss is a principled approach to modeling probability distributions that has become ubiquitous in deep learning. The softmax function is defined by a lone hyperparameter, the temperature, that is commonly set to one or regarded as a way to tune model confidence after training; however, less is known about how the temperature impacts training dynamics or generalization performance. In this work we develop a theory of early learning for models trained with softmax-cross-entropy loss and show that the learning dynamics depend crucially on the inverse-temperature $β$ as well as the magnitude of the logits at initialization, $||β{\bf z}||_{2}$. We follow up these analytic results with a large-scale empirical study of a variety of model architectures trained on CIFAR10, ImageNet, and IMDB sentiment analysis. We find that generalization performance depends strongly on the temperature, but only weakly on the initial logit magnitude. We provide evidence that the dependence of generalization on $β$ is not due to changes in model confidence, but is a dynamical phenomenon. It follows that the addition of $β$ as a tunable hyperparameter is key to maximizing model performance. Although we find the optimal $β$ to be sensitive to the architecture, our results suggest that tuning $β$ over the range $10^{-2}$ to $10^1$ improves performance over all architectures studied. We find that smaller $β$ may lead to better peak performance at the cost of learning stability. △ Less

Submitted 14 October, 2020; originally announced October 2020.

Statistics tuned entanglement of the boundary modes in coupled Su-Schrieffer-Heeger chains

Authors: Saikat Santra, Adhip Agarwala, Subhro Bhattacharjee

Abstract: We show that mutual statistics between quantum particles can be tuned to generate emergent novel few particle quantum mechanics for the boundary modes of symmetry-protected topological phases of matter. As a concrete setting, we study a system of pseudofermions, defined as quantum particles with tunable algebra, which lie on two distinct Su-Schrieffer-Heeger (SSH) chains. We find that as the mutua… ▽ More We show that mutual statistics between quantum particles can be tuned to generate emergent novel few particle quantum mechanics for the boundary modes of symmetry-protected topological phases of matter. As a concrete setting, we study a system of pseudofermions, defined as quantum particles with tunable algebra, which lie on two distinct Su-Schrieffer-Heeger (SSH) chains. We find that as the mutual statistics of the particles are tuned -- the boundary modes present in the two chains gets non-trivially entangled showing a sudden jump in their mutual entanglement entropy. We further show that, such tuning of statistics engenders a first-order transition between two topologically non-trivial phases which differ in the behavior of inter-chain entanglement. Using a combination of analytical and numerical techniques and effective modeling, we uncover the rich physics that this system hosts. The results are of particular relevance in context of the study of the effective low energy quantum mechanics of topological edge modes in one hand and their recent realization in ultracold atoms on the other. This then provides for controlled manipulation of such low energy modes. △ Less

Submitted 14 October, 2020; originally announced October 2020.

Comments: 17 pages, 15 figures

Journal ref: Phys. Rev. B 103, 195134 (2021)

Gapless state of interacting Majorana fermions in a strain-induced Landau level

Authors: Adhip Agarwala, Subhro Bhattacharjee, Johannes Knolle, Roderich Moessner

Abstract: Mechanical strain can generate a pseudo-magnetic field, and hence Landau levels (LL), for low energy excitations of quantum matter in two dimensions. We study the collective state of the fractionalised Majorana fermions arising from residual generic spin interactions in the central LL, where the projected Hamiltonian reflects the spin symmetries in intricate ways: emergent U(1) and particle-hole s… ▽ More Mechanical strain can generate a pseudo-magnetic field, and hence Landau levels (LL), for low energy excitations of quantum matter in two dimensions. We study the collective state of the fractionalised Majorana fermions arising from residual generic spin interactions in the central LL, where the projected Hamiltonian reflects the spin symmetries in intricate ways: emergent U(1) and particle-hole symmetries forbid any bilinear couplings, leading to an intrinsically strongly interacting system; also, they allow the definition of a filling fraction, which is fixed at 1/2. We argue that the resulting many-body state is gapless within our numerical accuracy, implying ultra-short-ranged spin correlations, while chirality correlators decay algebraically. This amounts to a Kitaev `non-Fermi' spin liquid, and shows that interacting Majorana Fermions can exhibit intricate behaviour akin to fractional quantum Hall physics in an insulating magnet. △ Less

Submitted 27 July, 2020; originally announced July 2020.

Comments: 4 pages, 4 figures, Supplemental Material: 4 pages

Journal ref: Phys. Rev. B 103, 134427 (2021)

Learning the gravitational force law and other analytic functions

Authors: Atish Agarwala, Abhimanyu Das, Rina Panigrahy, Qiuyi Zhang

Abstract: Large neural network models have been successful in learning functions of importance in many branches of science, including physics, chemistry and biology. Recent theoretical work has shown explicit learning bounds for wide networks and kernel methods on some simple classes of functions, but not on more complex functions which arise in practice. We extend these techniques to provide learning bound… ▽ More Large neural network models have been successful in learning functions of importance in many branches of science, including physics, chemistry and biology. Recent theoretical work has shown explicit learning bounds for wide networks and kernel methods on some simple classes of functions, but not on more complex functions which arise in practice. We extend these techniques to provide learning bounds for analytic functions on the sphere for any kernel method or equivalent infinitely-wide network with the corresponding activation function trained with SGD. We show that a wide, one-hidden layer ReLU network can learn analytic functions with a number of samples proportional to the derivative of a related function. Many functions important in the sciences are therefore efficiently learnable. As an example, we prove explicit bounds on learning the many-body gravitational force function given by Newton's law of gravitation. Our theoretical bounds suggest that very wide ReLU networks (and the corresponding NTK kernel) are better at learning analytic functions as compared to kernel learning with Gaussian kernels. We present experimental evidence that the many-body gravitational force function is easier to learn with ReLU networks as compared to networks with exponential activations. △ Less

Submitted 15 May, 2020; originally announced May 2020.

Deep Homography Estimation for Dynamic Scenes

Authors: Hoang Le, Feng Liu, Shu Zhang, Aseem Agarwala

Abstract: Homography estimation is an important step in many computer vision problems. Recently, deep neural network methods have shown to be favorable for this problem when compared to traditional methods. However, these new methods do not consider dynamic content in input images. They train neural networks with only image pairs that can be perfectly aligned using homographies. This paper investigates and… ▽ More Homography estimation is an important step in many computer vision problems. Recently, deep neural network methods have shown to be favorable for this problem when compared to traditional methods. However, these new methods do not consider dynamic content in input images. They train neural networks with only image pairs that can be perfectly aligned using homographies. This paper investigates and discusses how to design and train a deep neural network that handles dynamic scenes. We first collect a large video dataset with dynamic content. We then develop a multi-scale neural network and show that when properly trained using our new dataset, this neural network can already handle dynamic scenes to some extent. To estimate a homography of a dynamic scene in a more principled way, we need to identify the dynamic content. Since dynamic content detection and homography estimation are two tightly coupled tasks, we follow the multi-task learning principles and augment our multi-scale network such that it jointly estimates the dynamics masks and homographies. Our experiments show that our method can robustly estimate homography for challenging scenarios with dynamic scenes, blur artifacts, or lack of textures. △ Less

Submitted 5 April, 2020; originally announced April 2020.

Comments: CVPR 2020, https://github.com/lcmhoang/hmg-dynamics

Topological and conventional phases of a three dimensional electron glass

Authors: Prateek Mukati, Adhip Agarwala, Subhro Bhattacharjee

Abstract: We investigate a symmetry protected $Z_2$ topological electron glass -- a glassy equivalent of the $Z_2$ topological band insulator in crystalline systems -- and uncover associated quantum phase transitions in this three dimensional amorphous network of atoms. Through explicit numerical calculations of the Witten effect, we show that the $Z_2$ glass is characterized by an anomalous electromagnetic… ▽ More We investigate a symmetry protected $Z_2$ topological electron glass -- a glassy equivalent of the $Z_2$ topological band insulator in crystalline systems -- and uncover associated quantum phase transitions in this three dimensional amorphous network of atoms. Through explicit numerical calculations of the Witten effect, we show that the $Z_2$ glass is characterized by an anomalous electromagnetic response -- dyons with 1/2 electronic charge. We further study, using a variety of numerical diagnostics including such electromagnetic responses, the phase transitions of the $Z_2$ glass into a metallic and/or a trivial insulating phase. We find that the phase transitions here are governed by subtle features of mobility edges and "spectral inversion" which are possibly unique to structurally amorphous systems. Our results provide a concrete setting to understand the general underpinnings of such phases -- where strong structural disorder interplays with symmetry-protected topological order. △ Less

Submitted 23 July, 2019; originally announced July 2019.

Comments: 8 pages, 9 figures

Journal ref: Phys. Rev. B 101, 035142 (2020)

Higher-order topological insulators in amorphous solids

Authors: Adhip Agarwala, Vladimir Juricic, Bitan Roy

Abstract: We identify the possibility of realizing higher order topological (HOT) phases in noncrystalline or amorphous materials. Starting from two and three dimensional crystalline HOT insulators, accommodating topological corner states, we gradually enhance structural randomness in the system. Within a parameter regime, as long as amorphousness is confined by outer crystalline boundary, the system contin… ▽ More We identify the possibility of realizing higher order topological (HOT) phases in noncrystalline or amorphous materials. Starting from two and three dimensional crystalline HOT insulators, accommodating topological corner states, we gradually enhance structural randomness in the system. Within a parameter regime, as long as amorphousness is confined by outer crystalline boundary, the system continues to host corner states, yielding amorphous HOT insulators. However, as structural disorder percolates to the edges, corner states start to dissolve into amorphous bulk, and ultimately the system becomes a trivial insulator when amorphousness plagues the entire system. These outcomes are further substantiated by computing the quadrupolar (octupolar) moment in two (three) dimensions. Therefore, HOT phases can be realized in amorphous solids, when wrapped by a thin (lithographically grown, for example) crystalline layer. Our findings suggest that crystalline topological phases can be realized even in the absence of local crystalline symmetry. △ Less

Submitted 17 March, 2020; v1 submitted 1 February, 2019; originally announced February 2019.

Comments: 6 Pages, 5 Figures: Published version (Supplemental Materials as ancillary file)

Journal ref: Phys. Rev. Research 2, 012067 (2020)

Statistics-tuned phases of pseudofermions in one dimension

Authors: Adhip Agarwala, Gaurav Kumar Gupta, Vijay B. Shenoy, Subhro Bhattacharjee

Abstract: We show that a quadratic system of pseudofermions, with tunable fractionalised statistics, can host a rich phase diagram on a one dimensional chain with nearest and next nearest neighbor hop**. Using a combination of numerical and analytical techniques, we show that that by varying the statistical angle and the ratio of the hop**, the system stabilizes two Tomonaga-Luttinger liquids (TLL) with… ▽ More We show that a quadratic system of pseudofermions, with tunable fractionalised statistics, can host a rich phase diagram on a one dimensional chain with nearest and next nearest neighbor hop**. Using a combination of numerical and analytical techniques, we show that that by varying the statistical angle and the ratio of the hop**, the system stabilizes two Tomonaga-Luttinger liquids (TLL) with central charges c = 1 and 2 respectively along with the inversion symmetry broken bond ordered (BO) insulating phase. Interestingly, the two quantum phase transitions in the system - (1) between the two TLLs, and, (2) the c = 1 TLL and BO phase can be engendered by solely tuning the statistics of the pseudofermions. Our analysis shows that both these transition are continuous and novel with the former lacking a local order-parameter based description and the latter of Berezinskii-Kosterlitz-Thouless type. These phases and phase transitions can be of direct experimental relevance in context of recent studies of fermionic cold atoms. △ Less

Submitted 28 November, 2018; originally announced November 2018.

Comments: 6+12 pages, 4+14 figures

Journal ref: Phys. Rev. B 99, 165125 (2019)

A Compact Embedding for Facial Expression Similarity

Authors: Raviteja Vemulapalli, Aseem Agarwala

Abstract: Most of the existing work on automatic facial expression analysis focuses on discrete emotion recognition, or facial action unit detection. However, facial expressions do not always fall neatly into pre-defined semantic categories. Also, the similarity between expressions measured in the action unit space need not correspond to how humans perceive expression similarity. Different from previous wor… ▽ More Most of the existing work on automatic facial expression analysis focuses on discrete emotion recognition, or facial action unit detection. However, facial expressions do not always fall neatly into pre-defined semantic categories. Also, the similarity between expressions measured in the action unit space need not correspond to how humans perceive expression similarity. Different from previous work, our goal is to describe facial expressions in a continuous fashion using a compact embedding space that mimics human visual preferences. To achieve this goal, we collect a large-scale faces-in-the-wild dataset with human annotations in the form: Expressions A and B are visually more similar when compared to expression C, and use this dataset to train a neural network that produces a compact (16-dimensional) expression embedding. We experimentally demonstrate that the learned embedding can be successfully used for various applications such as expression retrieval, photo album summarization, and emotion recognition. We also show that the embedding learned using the proposed dataset performs better than several other embeddings learned using existing emotion or action unit datasets. △ Less

Submitted 9 January, 2019; v1 submitted 27 November, 2018; originally announced November 2018.

Fractalized Metals

Authors: Adhip Agarwala, Shriya Pai, Vijay B. Shenoy

Abstract: The classification of gapped phases of non-interacting fermions hinges on the tenfold symmetries and on the spatial dimension. The notion of dimension leads to a well defined demarcation between bulk and edge. Here we explore the nature of topological phases in systems where the distinction between bulk and edge is nebulous, of which fractal lattices are canonical examples. Our key finding is that… ▽ More The classification of gapped phases of non-interacting fermions hinges on the tenfold symmetries and on the spatial dimension. The notion of dimension leads to a well defined demarcation between bulk and edge. Here we explore the nature of topological phases in systems where the distinction between bulk and edge is nebulous, of which fractal lattices are canonical examples. Our key finding is that in homogeneous fractal lattices (where every site is equally coordinated), there are no gapped topological phases. What appears instead is a novel metallic state -- the fractalized metal -- whose low energy states arrange hierarchically on the structure of the fractal that hosts them. We study the properties (such as chiral transport) of this metal and demonstrate its robustness to disorder. Further, by studying a variety of fractal models we establish that the homogeneity of the fractal is a key condition for the realization of such fractalized metallic states. △ Less

Submitted 4 March, 2018; originally announced March 2018.

Comments: 5 pages, 5 figures and Supplemental Material (6 pages, 9 figures)

Effects of local periodic driving on transport and generation of bound states

Authors: Adhip Agarwala, Diptiman Sen

Abstract: We periodically kick a local region in a one-dimensional lattice and demonstrate, by studying wave packet dynamics, that the strength and the time period of the kicking can be used as tuning parameters to control the transmission probability across the region. Interestingly, we can tune the transmission to zero which is otherwise impossible to do in a time-independent system. We adapt the non-equi… ▽ More We periodically kick a local region in a one-dimensional lattice and demonstrate, by studying wave packet dynamics, that the strength and the time period of the kicking can be used as tuning parameters to control the transmission probability across the region. Interestingly, we can tune the transmission to zero which is otherwise impossible to do in a time-independent system. We adapt the non-equilibrium Green's function method to take into account the effects of periodic driving; the results obtained by this method agree with those found by wave packet dynamics if the time period is small. We discover that Floquet bound states can exist in certain ranges of parameters; when the driving frequency is decreased, these states get delocalized and turn into resonances by mixing with the Floquet bulk states. We extend these results to incorporate the effects of local interactions at the driven site, and we find some interesting features in the transmission and the bound states. △ Less

Submitted 7 September, 2017; v1 submitted 8 June, 2017; originally announced June 2017.

Comments: 14 pages, 12 figures; added several references and corrected some typos

Journal ref: Phys. Rev. B 96, 104309 (2017)

Video Frame Synthesis using Deep Voxel Flow

Authors: Ziwei Liu, Raymond A. Yeh, Xiaoou Tang, Yiming Liu, Aseem Agarwala

Abstract: We address the problem of synthesizing new video frames in an existing video, either in-between existing frames (interpolation), or subsequent to them (extrapolation). This problem is challenging because video appearance and motion can be highly complex. Traditional optical-flow-based solutions often fail where flow estimation is challenging, while newer neural-network-based methods that hallucina… ▽ More We address the problem of synthesizing new video frames in an existing video, either in-between existing frames (interpolation), or subsequent to them (extrapolation). This problem is challenging because video appearance and motion can be highly complex. Traditional optical-flow-based solutions often fail where flow estimation is challenging, while newer neural-network-based methods that hallucinate pixel values directly often produce blurry results. We combine the advantages of these two methods by training a deep network that learns to synthesize video frames by flowing pixel values from existing ones, which we call deep voxel flow. Our method requires no human supervision, and any video can be used as training data by drop**, and then learning to predict, existing frames. The technique is efficient, and can be applied at any video resolution. We demonstrate that our method produces results that both quantitatively and qualitatively improve upon the state-of-the-art. △ Less

Submitted 5 August, 2017; v1 submitted 8 February, 2017; originally announced February 2017.

Comments: To appear in ICCV 2017 as an oral paper. More details at the project page: https://liuziwei7.github.io/projects/VoxelFlow.html

Topological Insulators in Random Lattices

Authors: Adhip Agarwala, Vijay B. Shenoy

Abstract: Our understanding of topological insulators is based on an underlying crystalline lattice where the local electronic degrees of freedom at different sites hybridize with each other in ways that produce nontrivial band topology, and the search for material systems to realize such phases have been strongly influenced by this. Here we theoretically demonstrate topological insulators in systems with a… ▽ More Our understanding of topological insulators is based on an underlying crystalline lattice where the local electronic degrees of freedom at different sites hybridize with each other in ways that produce nontrivial band topology, and the search for material systems to realize such phases have been strongly influenced by this. Here we theoretically demonstrate topological insulators in systems with a random distribution of sites in space, i. e., a random lattice. This is achieved by constructing hop** models on random lattices whose ground states possess nontrivial topological nature (characterized e. g., by Bott indices) that manifests as quantized conductances in systems with a boundary. By tuning parameters such as the density of sites (for a given range of fermion hop**), we can achieve transitions from trivial to topological phases. We discuss interesting features of these transitions. In two spatial dimensions, we show this for all five symmetry classes (A, AII, D, DIII and C) that are known to host nontrivial topology in crystalline systems. We expect similar physics to be realizable in any dimension and provide an explicit example of a $Z_2$ topological insulator on a random lattice in three spatial dimensions. Our study not only provides a deeper understanding of the topological phases of non-interacting fermions, but also suggests new directions in the pursuit of the laboratory realization of topological quantum matter. △ Less

Submitted 2 January, 2017; originally announced January 2017.

Comments: 6 pages, 5 figures

Journal ref: Phys. Rev. Lett. 118, 236402 (2017)

Killing the Hofstadter butterfly, one bond at a time

Authors: Adhip Agarwala

Abstract: Electronic bands in a square lattice when subjected to a perpendicular magnetic field form the Hofstadter butterfly pattern. We study the evolution of this pattern as a function of bond percolation disorder (removal or dilution of lattice bonds). With increasing concentration of the bonds removed, the butterfly pattern gets smoothly decimated. However, in this process of decimation, bands develop… ▽ More Electronic bands in a square lattice when subjected to a perpendicular magnetic field form the Hofstadter butterfly pattern. We study the evolution of this pattern as a function of bond percolation disorder (removal or dilution of lattice bonds). With increasing concentration of the bonds removed, the butterfly pattern gets smoothly decimated. However, in this process of decimation, bands develop interesting characteristics and features. For example, in the high disorder limit, some butterfly-like pattern still persists even as most of the states are localized. We also analyze, in the low disorder limit, the effect of percolation on wavefunctions (using inverse participation ratios) and on band gaps in the spectrum. We explain and provide the reasons behind many of the key features in our results by analyzing small clusters and finite size rings. Furthermore, we study the effect of bond dilution on transverse conductivity($σ_{xy}$). We show that starting from the clean limit, increasing disorder reduces $σ_{xy}$ to zero, even though the strength of percolation is smaller than the classical percolation threshold. This shows that the system undergoes a direct transition from a integer quantum Hall state to a localized Anderson insulator beyond a critical value of bond dilution. We further find that the energy bands close to the band edge are more stable to disorder than at the band center. To arrive at these results we use the coupling matrix approach to calculate Chern numbers for disordered systems. We point out the relevance of these results to signatures in magneto-oscillations. △ Less

Submitted 21 January, 2017; v1 submitted 4 December, 2016; originally announced December 2016.

Comments: minor typos fixed

Journal ref: Eur. Phys. J. B, 90 1 (2017) 15

Semantic Facial Expression Editing using Autoencoded Flow

Authors: Raymond Yeh, Ziwei Liu, Dan B Goldman, Aseem Agarwala

Abstract: High-level manipulation of facial expressions in images --- such as changing a smile to a neutral expression --- is challenging because facial expression changes are highly non-linear, and vary depending on the appearance of the face. We present a fully automatic approach to editing faces that combines the advantages of flow-based face manipulation with the more recent generative capabilities of V… ▽ More High-level manipulation of facial expressions in images --- such as changing a smile to a neutral expression --- is challenging because facial expression changes are highly non-linear, and vary depending on the appearance of the face. We present a fully automatic approach to editing faces that combines the advantages of flow-based face manipulation with the more recent generative capabilities of Variational Autoencoders (VAEs). During training, our model learns to encode the flow from one expression to another over a low-dimensional latent space. At test time, expression editing can be done simply using latent vector arithmetic. We evaluate our methods on two applications: 1) single-image facial expression editing, and 2) facial expression interpolation between two images. We demonstrate that our method generates images of higher perceptual quality than previous VAE and flow-based methods. △ Less

Submitted 29 November, 2016; originally announced November 2016.

Effects of interactions on periodically driven dynamically localized systems

Authors: Adhip Agarwala, Diptiman Sen

Abstract: It is known that there are lattice models in which non-interacting particles get dynamically localized when periodic $δ$-function kicks are applied with a particular strength. We use both numerical and analytical methods to study the effects of interactions in three different models in one dimension. The systems we have considered include spinless fermions with interactions between nearest-neighbo… ▽ More It is known that there are lattice models in which non-interacting particles get dynamically localized when periodic $δ$-function kicks are applied with a particular strength. We use both numerical and analytical methods to study the effects of interactions in three different models in one dimension. The systems we have considered include spinless fermions with interactions between nearest-neighbor sites, the Hubbard model of spin-1/2 fermions, and the Bose Hubbard model with on-site interactions. We derive effective Floquet Hamiltonians up to second order in the time period of kicking. Using these we show that interactions can give rise to a variety of interesting results such as two-body bound states in all three models and dispersionless many-body bound states for spinless fermions and bosons. We substantiate these results by exact diagonalization and stroboscopic time evolution of systems with a finite number of particles. We derive a low-energy pseudo-spin-1/2 limit of the Bose Hubbard system in the thermodynamic limit and show that a special case of this has an exponentially large number of ground states. Finally we study the effect of changing the strength of the $δ$-function kicks slightly away from perfect dynamical localization; we find that a single particle remains dynamically localized for a long time after which it moves ballistically. △ Less

Submitted 16 January, 2017; v1 submitted 18 August, 2016; originally announced August 2016.

Comments: 52 pages, single-column, 9 figures; this is the published version

Journal ref: Phys. Rev. B 95, 014305 (2017)

The tenfold way redux: Fermionic systems with $N$-body interactions

Authors: Adhip Agarwala, Arijit Haldar, Vijay B. Shenoy

Abstract: We provide a systematic treatment of the tenfold way of classifying fermionic systems that naturally allows for the study of those with arbitrary $N$-body interactions. We identify four types of symmetries that such systems can possess, which consist of one ordinary type (usual unitary symmetries), and three non-ordinary symmetries (such as time reversal, charge conjugation and sublattice). Focusi… ▽ More We provide a systematic treatment of the tenfold way of classifying fermionic systems that naturally allows for the study of those with arbitrary $N$-body interactions. We identify four types of symmetries that such systems can possess, which consist of one ordinary type (usual unitary symmetries), and three non-ordinary symmetries (such as time reversal, charge conjugation and sublattice). Focusing on systems that possess no non-trivial ordinary symmetries, we demonstrate that the non-ordinary symmetries are strongly constrained. This approach not only leads very naturally to the tenfold classes, but also obtains the canonical representations of these symmetries in each of the ten classes. We also provide a group cohomological perspective of our results in terms of projective representations. We then use the canonical representations of the symmetries to obtain the structure of Hamiltonians with arbitrary $N$-body interactions in each of the ten classes. We show that the space of $N$-body Hamiltonians has an affine subspace (of a vector space) structure in classes which have either or both charge conjugation and sublattice symmetries. Our results can help address open questions on the topological classification of interacting fermionic systems. △ Less

Submitted 17 June, 2016; originally announced June 2016.

Comments: 28 pages, 4 figures, 4 tables

Effects of periodic kicking on dispersion and wave packet dynamics in graphene

Authors: Adhip Agarwala, Utso Bhattacharya, Amit Dutta, Diptiman Sen

Abstract: We study the effects of $δ$-function periodic kicks on the Floquet energy-momentum dispersion in graphene. We find that a rich variety of dispersions can appear depending on the parameters of the kicking: at certain points in the Brillouin zone, the dispersion can become linear but anisotropic, linear in one direction and quadratic in the perpendicular direction, gapped with a quadratic dispersion… ▽ More We study the effects of $δ$-function periodic kicks on the Floquet energy-momentum dispersion in graphene. We find that a rich variety of dispersions can appear depending on the parameters of the kicking: at certain points in the Brillouin zone, the dispersion can become linear but anisotropic, linear in one direction and quadratic in the perpendicular direction, gapped with a quadratic dispersion, or completely flat (called dynamical localization). We show all these results analytically and demonstrate them numerically through the dynamics of wave packets propagating in graphene. We propose experimental methods for producing these effects. △ Less

Submitted 8 March, 2016; v1 submitted 26 February, 2016; originally announced February 2016.

Comments: 8 pages, 5 figures; added some references and made minor corrections

Journal ref: Phys. Rev. B 93,174301 (2016)

Fractional Local Moment and High Temperature Kondo Effect in Rashba-Fermi Gases

Authors: Adhip Agarwala, Vijay B. Shenoy

Abstract: We investigate the new physics that arises when a correlated quantum impurity hybridizes with Fermi gas with a generalized Rashba spin-orbit coupling produced via a uniform synthetic non-Abelian gauge field. We show that the impurity develops a {\it fractional} local moment which couples anti-ferromagnetically to the Rashba-Fermi gas. This results in a concomitant {\it Kondo effect with a high tem… ▽ More We investigate the new physics that arises when a correlated quantum impurity hybridizes with Fermi gas with a generalized Rashba spin-orbit coupling produced via a uniform synthetic non-Abelian gauge field. We show that the impurity develops a {\it fractional} local moment which couples anti-ferromagnetically to the Rashba-Fermi gas. This results in a concomitant {\it Kondo effect with a high temperature scale} that can be tuned by the strength of the Rashba spin-orbit coupling. △ Less

Submitted 30 September, 2015; v1 submitted 24 September, 2015; originally announced September 2015.

Comments: 5 pages, 4 figures, and supplemental material(4 pages); typos fixed

Journal ref: Phys. Rev. B 93, 241111 (2016)

Performance metrics in a hybrid MPI-OpenMP based molecular dynamics simulation with short-range interactions

Authors: Anirban Pal, Abhishek Agarwala, Soumyendu Raha, Baidurya Bhattacharya

Abstract: We discuss the computational bottlenecks in molecular dynamics (MD) and describe the challenges in parallelizing the computation intensive tasks. We present a hybrid algorithm using MPI (Message Passing Interface) with OpenMP threads for parallelizing a generalized MD computation scheme for systems with short range interatomic interactions. The algorithm is discussed in the context of nanoindentat… ▽ More We discuss the computational bottlenecks in molecular dynamics (MD) and describe the challenges in parallelizing the computation intensive tasks. We present a hybrid algorithm using MPI (Message Passing Interface) with OpenMP threads for parallelizing a generalized MD computation scheme for systems with short range interatomic interactions. The algorithm is discussed in the context of nanoindentation of Chromium films with carbon indenters using the Embedded Atom Method potential for Cr Cr interaction and the Morse potential for Cr C interactions. We study the performance of our algorithm for a range of MPIthread combinations and find the performance to depend strongly on the computational task and load sharing in the multicore processor. The algorithm scaled poorly with MPI and our hybrid schemes were observed to outperform the pure message passing scheme, despite utilizing the same number of processors or cores in the cluster. Speed-up achieved by our algorithm compared favourably with that achieved by standard MD packages. △ Less

Submitted 25 July, 2015; originally announced July 2015.

Journal ref: Journal of Parallel and Distributed Computing, Elsevier, vol. 74, no. 3, pp. 2203-2214, 2014

DeepFont: Identify Your Font from An Image

Authors: Zhangyang Wang, Jianchao Yang, Hailin **, Eli Shechtman, Aseem Agarwala, Jonathan Brandt, Thomas S. Huang

Abstract: As font is one of the core design concepts, automatic font identification and similar font suggestion from an image or photo has been on the wish list of many designers. We study the Visual Font Recognition (VFR) problem, and advance the state-of-the-art remarkably by develo** the DeepFont system. First of all, we build up the first available large-scale VFR dataset, named AdobeVFR, consisting o… ▽ More As font is one of the core design concepts, automatic font identification and similar font suggestion from an image or photo has been on the wish list of many designers. We study the Visual Font Recognition (VFR) problem, and advance the state-of-the-art remarkably by develo** the DeepFont system. First of all, we build up the first available large-scale VFR dataset, named AdobeVFR, consisting of both labeled synthetic data and partially labeled real-world data. Next, to combat the domain mismatch between available training and testing data, we introduce a Convolutional Neural Network (CNN) decomposition approach, using a domain adaptation technique based on a Stacked Convolutional Auto-Encoder (SCAE) that exploits a large corpus of unlabeled real-world text images combined with synthetic data preprocessed in a specific way. Moreover, we study a novel learning-based model compression approach, in order to reduce the DeepFont model size without sacrificing its performance. The DeepFont system achieves an accuracy of higher than 80% (top-5) on our collected dataset, and also produces a good font similarity measure for font selection and suggestion. We also achieve around 6 times compression of the model without any visible loss of recognition accuracy. △ Less

Submitted 12 July, 2015; originally announced July 2015.

Comments: To Appear in ACM Multimedia as a full paper