-
Cyclotomic Structures in Symplectic Topology
Authors:
Semon Rezchikov
Abstract:
We extend the Cohen-Jones-Segal construction of stable homotopy types associated to flow categories of Morse-Smale functions $f$ to the setting where $f$ is equivariant under a finite group action and is Morse but no longer Morse-Smale. This setting occurs universally, as equivariant Morse functions can rarely be perturbed to nearby equivariant Morse-Smale functions. The method is very general, an…
▽ More
We extend the Cohen-Jones-Segal construction of stable homotopy types associated to flow categories of Morse-Smale functions $f$ to the setting where $f$ is equivariant under a finite group action and is Morse but no longer Morse-Smale. This setting occurs universally, as equivariant Morse functions can rarely be perturbed to nearby equivariant Morse-Smale functions. The method is very general, and allows one to do equivariant Floer theory while avoiding all the complications typically caused by issues of equivariant transversality. The construction assigns a (genuine) equivariant orthogonal spectrum to an equivariant framed virtually smooth flow category. Using this method, we construct, for a compact symplectic manifold $M$, which is symplectically atoroidal with contact boundary, and is equipped with an equivariant trivialization of its polarization class, a cyclotomic structure on the spectral lift of the symplectic cohomology $SH^*(M)$. This generalizes a variant of the map which sends loops to their $p$-fold covers on free loop spaces to the setting of general Liouville domains, and suggests a systematic connection between Floer homology and $p$-adic Hodge theory.
△ Less
Submitted 28 May, 2024;
originally announced May 2024.
-
How Flawed Is ECE? An Analysis via Logit Smoothing
Authors:
Muthu Chidambaram,
Holden Lee,
Colin McSwiggen,
Semon Rezchikov
Abstract:
Informally, a model is calibrated if its predictions are correct with a probability that matches the confidence of the prediction. By far the most common method in the literature for measuring calibration is the expected calibration error (ECE). Recent work, however, has pointed out drawbacks of ECE, such as the fact that it is discontinuous in the space of predictors. In this work, we ask: how fu…
▽ More
Informally, a model is calibrated if its predictions are correct with a probability that matches the confidence of the prediction. By far the most common method in the literature for measuring calibration is the expected calibration error (ECE). Recent work, however, has pointed out drawbacks of ECE, such as the fact that it is discontinuous in the space of predictors. In this work, we ask: how fundamental are these issues, and what are their impacts on existing results? Towards this end, we completely characterize the discontinuities of ECE with respect to general probability measures on Polish spaces. We then use the nature of these discontinuities to motivate a novel continuous, easily estimated miscalibration metric, which we term Logit-Smoothed ECE (LS-ECE). By comparing the ECE and LS-ECE of pre-trained image classification models, we show in initial experiments that binned ECE closely tracks LS-ECE, indicating that the theoretical pathologies of ECE may be avoidable in practice.
△ Less
Submitted 3 June, 2024; v1 submitted 15 February, 2024;
originally announced February 2024.
-
Renormalizing Diffusion Models
Authors:
Jordan Cotler,
Semon Rezchikov
Abstract:
We explain how to use diffusion models to learn inverse renormalization group flows of statistical and quantum field theories. Diffusion models are a class of machine learning models which have been used to generate samples from complex distributions, such as the distribution of natural images. These models achieve sample generation by learning the inverse process to a diffusion process which adds…
▽ More
We explain how to use diffusion models to learn inverse renormalization group flows of statistical and quantum field theories. Diffusion models are a class of machine learning models which have been used to generate samples from complex distributions, such as the distribution of natural images. These models achieve sample generation by learning the inverse process to a diffusion process which adds noise to the data until the distribution of the data is pure noise. Nonperturbative renormalization group schemes in physics can naturally be written as diffusion processes in the space of fields. We combine these observations in a concrete framework for building ML-based models for studying field theories, in which the models learn the inverse process to an explicitly-specified renormalization group scheme. We detail how these models define a class of adaptive bridge (or parallel tempering) samplers for lattice field theory. Because renormalization group schemes have a physical meaning, we provide explicit prescriptions for how to compare results derived from models associated to several different renormalization group schemes of interest. We also explain how to use diffusion models in a variational method to find ground states of quantum systems. We apply some of our methods to numerically find RG flows of interacting statistical field theories. From the perspective of machine learning, our work provides an interpretation of multiscale diffusion models, and gives physically-inspired suggestions for diffusion models which should have novel properties.
△ Less
Submitted 5 September, 2023; v1 submitted 23 August, 2023;
originally announced August 2023.
-
Diffusion with Forward Models: Solving Stochastic Inverse Problems Without Direct Supervision
Authors:
Ayush Tewari,
Tianwei Yin,
George Cazenavette,
Semon Rezchikov,
Joshua B. Tenenbaum,
Frédo Durand,
William T. Freeman,
Vincent Sitzmann
Abstract:
Denoising diffusion models are a powerful type of generative models used to capture complex distributions of real-world signals. However, their applicability is limited to scenarios where training samples are readily available, which is not always the case in real-world applications. For example, in inverse graphics, the goal is to generate samples from a distribution of 3D scenes that align with…
▽ More
Denoising diffusion models are a powerful type of generative models used to capture complex distributions of real-world signals. However, their applicability is limited to scenarios where training samples are readily available, which is not always the case in real-world applications. For example, in inverse graphics, the goal is to generate samples from a distribution of 3D scenes that align with a given image, but ground-truth 3D scenes are unavailable and only 2D images are accessible. To address this limitation, we propose a novel class of denoising diffusion probabilistic models that learn to sample from distributions of signals that are never directly observed. Instead, these signals are measured indirectly through a known differentiable forward model, which produces partial observations of the unknown signal. Our approach involves integrating the forward model directly into the denoising process. This integration effectively connects the generative modeling of observations with the generative modeling of the underlying signals, allowing for end-to-end training of a conditional generative model over signals. During inference, our approach enables sampling from the distribution of underlying signals that are consistent with a given partial observation. We demonstrate the effectiveness of our method on three challenging computer vision tasks. For instance, in the context of inverse graphics, our model enables direct sampling from the distribution of 3D scenes that align with a single 2D input image.
△ Less
Submitted 16 November, 2023; v1 submitted 20 June, 2023;
originally announced June 2023.
-
Holomorphic Floer Theory and the Fueter Equation
Authors:
Aleksander Doan,
Semon Rezchikov
Abstract:
We outline a proposal for a $2$-category $\mathrm{Fuet}_M$ associated to a hyperkähler manifold $M$, which categorifies the subcategory of the Fukaya category of $M$ generated by complex Lagrangians. Morphisms in this $2$-category are formally the Fukaya--Seidel categories of holomorphic symplectic action functionals. As such, $\mathrm{Fuet}_M$ is based on counting maps to $M$ satisfying the Fuete…
▽ More
We outline a proposal for a $2$-category $\mathrm{Fuet}_M$ associated to a hyperkähler manifold $M$, which categorifies the subcategory of the Fukaya category of $M$ generated by complex Lagrangians. Morphisms in this $2$-category are formally the Fukaya--Seidel categories of holomorphic symplectic action functionals. As such, $\mathrm{Fuet}_M$ is based on counting maps to $M$ satisfying the Fueter equation with boundary values on holomorphic Lagrangians. We make the first step towards constructing this category by establishing some basic analytic results about Fueter maps, such as the energy bound and maximum principle. When $M=T^*X$ is the cotangent bundle of a Kähler manifold $X$ and $(L_0, L_1)$ are the zero section and the graph of the differential of a holomorphic function $F: X \to \mathbb{C}$, we prove that all Fueter maps correspond to the complex gradient trajectories of $F$ in $X$, which relates our proposal to the Fukaya--Seidel category of $F$. This is a complexification of Floer's theorem on pseudo-holomorphic strips in cotangent bundles. Throughout the paper, we suggest problems and research directions for analysts and geometers that may be interested in the subject.
△ Less
Submitted 21 August, 2023; v1 submitted 21 October, 2022;
originally announced October 2022.
-
Integral Arnol'd Conjecture
Authors:
Semon Rezchikov
Abstract:
We explain how to adapt the methods of Abouzaid-McLean-Smith to the setting of Hamiltonian Floer theory. We develop a language around equivariant ``$\langle k \rangle$-manifolds'', which are a type of manifold-with-corners that suffices to capture the combinatorics of Floer-theoretic constructions. We describe some geometry which allows us to straightforwardly adapt Lashofs's stable equivariant sm…
▽ More
We explain how to adapt the methods of Abouzaid-McLean-Smith to the setting of Hamiltonian Floer theory. We develop a language around equivariant ``$\langle k \rangle$-manifolds'', which are a type of manifold-with-corners that suffices to capture the combinatorics of Floer-theoretic constructions. We describe some geometry which allows us to straightforwardly adapt Lashofs's stable equivariant smoothing theory and Bau-Xu's theory of FOP-perturbations to $\langle k \rangle$-manifolds. This allows us to compatibly smooth global Kuranishi charts on all Hamiltonian Floer trajectories at once, in order to extract a Floer complex and prove the Arnol'd conjecture over the integers. We also make first steps towards a further development of the theory, outlining the analog of bifurcation analysis in this setting, which can give short independence proofs of the independence of Floer-theoretic invariants of all choices involved in their construction.
△ Less
Submitted 22 September, 2022;
originally announced September 2022.
-
Renormalization Group Flow as Optimal Transport
Authors:
Jordan Cotler,
Semon Rezchikov
Abstract:
We establish that Polchinski's equation for exact renormalization group flow is equivalent to the optimal transport gradient flow of a field-theoretic relative entropy. This provides a compelling information-theoretic formulation of the exact renormalization group, expressed in the language of optimal transport. A striking consequence is that a regularization of the relative entropy is in fact an…
▽ More
We establish that Polchinski's equation for exact renormalization group flow is equivalent to the optimal transport gradient flow of a field-theoretic relative entropy. This provides a compelling information-theoretic formulation of the exact renormalization group, expressed in the language of optimal transport. A striking consequence is that a regularization of the relative entropy is in fact an RG monotone. We compute this monotone in several examples. Our results apply more broadly to other exact renormalization group flow equations, including widely used specializations of Wegner-Morris flow. Moreover, our optimal transport framework for RG allows us to reformulate RG flow as a variational problem. This enables new numerical techniques and establishes a systematic connection between neural network methods and RG flows of conventional field theories.
△ Less
Submitted 12 March, 2023; v1 submitted 23 February, 2022;
originally announced February 2022.
-
Rational Quantum Cohomology of Steenrod Uniruled Manifolds
Authors:
Semon Rezchikov
Abstract:
We show that if a semipositive symplectic manifold $M^{2n}$ is Steenrod uniruled, in the sense that the quantum Steenrod power of the point class does not agree with its classical Steenrod power for any prime, then the (rational) quantum product on $M$ is deformed. This bridges the gap between the recent advances towards the Chance-McDuff conjecture utilizing quantum Steenrod operations, and the n…
▽ More
We show that if a semipositive symplectic manifold $M^{2n}$ is Steenrod uniruled, in the sense that the quantum Steenrod power of the point class does not agree with its classical Steenrod power for any prime, then the (rational) quantum product on $M$ is deformed. This bridges the gap between the recent advances towards the Chance-McDuff conjecture utilizing quantum Steenrod operations, and the natural formulation of the Chance-McDuff conjecture in terms of rational Gromov-Witten theory.
△ Less
Submitted 16 November, 2021;
originally announced November 2021.
-
Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering
Authors:
Vincent Sitzmann,
Semon Rezchikov,
William T. Freeman,
Joshua B. Tenenbaum,
Fredo Durand
Abstract:
Inferring representations of 3D scenes from 2D observations is a fundamental problem of computer graphics, computer vision, and artificial intelligence. Emerging 3D-structured neural scene representations are a promising approach to 3D scene understanding. In this work, we propose a novel neural scene representation, Light Field Networks or LFNs, which represent both geometry and appearance of the…
▽ More
Inferring representations of 3D scenes from 2D observations is a fundamental problem of computer graphics, computer vision, and artificial intelligence. Emerging 3D-structured neural scene representations are a promising approach to 3D scene understanding. In this work, we propose a novel neural scene representation, Light Field Networks or LFNs, which represent both geometry and appearance of the underlying 3D scene in a 360-degree, four-dimensional light field parameterized via a neural implicit representation. Rendering a ray from an LFN requires only a single network evaluation, as opposed to hundreds of evaluations per ray for ray-marching or volumetric based renderers in 3D-structured neural scene representations. In the setting of simple scenes, we leverage meta-learning to learn a prior over LFNs that enables multi-view consistent light field reconstruction from as little as a single image observation. This results in dramatic reductions in time and memory complexity, and enables real-time rendering. The cost of storing a 360-degree light field via an LFN is two orders of magnitude lower than conventional methods such as the Lumigraph. Utilizing the analytical differentiability of neural implicit representations and a novel parameterization of light space, we further demonstrate the extraction of sparse depth maps from LFNs.
△ Less
Submitted 18 January, 2022; v1 submitted 4 June, 2021;
originally announced June 2021.
-
Floer homology via Twisted Loop Spaces
Authors:
Semon Rezchikov
Abstract:
Answering a question of Witten, we introduce a novel method for defining an integral version of Lagrangian Floer homology, removing the standard restriction that the Lagrangians in question must be relatively Pin. Using this technique, we derive stronger bounds on the self-intersection of certain exact Lagrangians $\mathbb{RP}^2 \times L'$ than those that follow from traditional methods. We define…
▽ More
Answering a question of Witten, we introduce a novel method for defining an integral version of Lagrangian Floer homology, removing the standard restriction that the Lagrangians in question must be relatively Pin. Using this technique, we derive stronger bounds on the self-intersection of certain exact Lagrangians $\mathbb{RP}^2 \times L'$ than those that follow from traditional methods. We define a integral version of Lagrangian Floer homology all oriented closed exact Lagrangians $L$ in a Liouville domain and prove a general self-intersection bound coming from the algebraic properties of the diagonal bimodule of a twist of the dg-algebra of chains on the based loop space of $L$.
△ Less
Submitted 4 December, 2019; v1 submitted 3 September, 2019;
originally announced September 2019.
-
Stochastic natural gradient descent draws posterior samples in function space
Authors:
Samuel L. Smith,
Daniel Duckworth,
Semon Rezchikov,
Quoc V. Le,
Jascha Sohl-Dickstein
Abstract:
Recent work has argued that stochastic gradient descent can approximate the Bayesian uncertainty in model parameters near local minima. In this work we develop a similar correspondence for minibatch natural gradient descent (NGD). We prove that for sufficiently small learning rates, if the model predictions on the training set approach the true conditional distribution of labels given inputs, the…
▽ More
Recent work has argued that stochastic gradient descent can approximate the Bayesian uncertainty in model parameters near local minima. In this work we develop a similar correspondence for minibatch natural gradient descent (NGD). We prove that for sufficiently small learning rates, if the model predictions on the training set approach the true conditional distribution of labels given inputs, the stationary distribution of minibatch NGD approaches a Bayesian posterior near local minima. The temperature $T = εN / (2B)$ is controlled by the learning rate $ε$, training set size $N$ and batch size $B$. However minibatch NGD is not parameterisation invariant and it does not sample a valid posterior away from local minima. We therefore propose a novel optimiser, "stochastic NGD", which introduces the additional correction terms required to preserve both properties.
△ Less
Submitted 28 November, 2018; v1 submitted 25 June, 2018;
originally announced June 2018.
