Showing 1–14 of 14 results for author: Maxfield, T

Utilizing Low-Dimensional Molecular Embeddings for Rapid Chemical Similarity Search

Authors: Kathryn E. Kirchoff, James Wellnitz, Joshua E. Hochuli, Travis Maxfield, Konstantin I. Popov, Shawn Gomez, Alexander Tropsha

Abstract: Nearest neighbor-based similarity searching is a common task in chemistry, with notable use cases in drug discovery. Yet, some of the most commonly used approaches for this task still leverage a brute-force approach. In practice this can be computationally costly and overly time-consuming, due in part to the sheer size of modern chemical databases. Previous computational advancements for this task… ▽ More Nearest neighbor-based similarity searching is a common task in chemistry, with notable use cases in drug discovery. Yet, some of the most commonly used approaches for this task still leverage a brute-force approach. In practice this can be computationally costly and overly time-consuming, due in part to the sheer size of modern chemical databases. Previous computational advancements for this task have generally relied on improvements to hardware or dataset-specific tricks that lack generalizability. Approaches that leverage lower-complexity searching algorithms remain relatively underexplored. However, many of these algorithms are approximate solutions and/or struggle with typical high-dimensional chemical embeddings. Here we evaluate whether a combination of low-dimensional chemical embeddings and a k-d tree data structure can achieve fast nearest neighbor queries while maintaining performance on standard chemical similarity search benchmarks. We examine different dimensionality reductions of standard chemical embeddings as well as a learned, structurally-aware embedding -- SmallSA -- for this task. With this framework, searches on over one billion chemicals execute in less than a second on a single CPU core, five orders of magnitude faster than the brute-force approach. We also demonstrate that SmallSA achieves competitive performance on chemical similarity benchmarks. △ Less

Submitted 12 February, 2024; originally announced February 2024.

SALSA: Semantically-Aware Latent Space Autoencoder

Authors: Kathryn E. Kirchoff, Travis Maxfield, Alexander Tropsha, Shawn M. Gomez

Abstract: In deep learning for drug discovery, chemical data are often represented as simplified molecular-input line-entry system (SMILES) sequences which allow for straightforward implementation of natural language processing methodologies, one being the sequence-to-sequence autoencoder. However, we observe that training an autoencoder solely on SMILES is insufficient to learn molecular representations th… ▽ More In deep learning for drug discovery, chemical data are often represented as simplified molecular-input line-entry system (SMILES) sequences which allow for straightforward implementation of natural language processing methodologies, one being the sequence-to-sequence autoencoder. However, we observe that training an autoencoder solely on SMILES is insufficient to learn molecular representations that are semantically meaningful, where semantics are defined by the structural (graph-to-graph) similarities between molecules. We demonstrate by example that autoencoders may map structurally similar molecules to distant codes, resulting in an incoherent latent space that does not respect the structural similarities between molecules. To address this shortcoming we propose Semantically-Aware Latent Space Autoencoder (SALSA), a transformer-autoencoder modified with a contrastive task, tailored specifically to learn graph-to-graph similarity between molecules. Formally, the contrastive objective is to map structurally similar molecules (separated by a single graph edit) to nearby codes in the latent space. To accomplish this, we generate a novel dataset comprised of sets of structurally similar molecules and opt for a supervised contrastive loss that is able to incorporate full sets of positive samples. We compare SALSA to its ablated counterparts, and show empirically that the composed training objective (reconstruction and contrastive task) leads to a higher quality latent space that is more 1) structurally-aware, 2) semantically continuous, and 3) property-aware. △ Less

Submitted 4 October, 2023; originally announced October 2023.

The N-ary in the Coal Mine: Avoiding Mixture Model Failure with Proper Validation

Authors: Travis Maxfield, Joshua Hochuli, James Wellnitz, Cleber Melo-Filho, Konstantin I. Popov, Eugene Muratov, Alex Tropsha

Abstract: Modeling the properties of chemical mixtures is a difficult but important part of any modeling process intended to be applicable to the often messy and impure phenomena of everyday life, including food and environmental safety, healthcare, etc. Part of this difficulty stems from the increased complexity of designing suitable model validation schemes for mixture data, a fact which has been elucidat… ▽ More Modeling the properties of chemical mixtures is a difficult but important part of any modeling process intended to be applicable to the often messy and impure phenomena of everyday life, including food and environmental safety, healthcare, etc. Part of this difficulty stems from the increased complexity of designing suitable model validation schemes for mixture data, a fact which has been elucidated in previous work only in the case of binary mixture models. We extend these previously defined validation strategies for QSAR modeling of binary mixtures to the more complex case of general, $N$-ary mixtures and argue that these strategies are applicable to many modeling tasks beyond simple chemical mixtures. Additionally, we propose a method of establishing a baseline model performance for each mixture dataset to be in used in model selection comparisons. This baseline is intended to account for the statistical dependence generically present between the properties of mixtures that share constituents. We contend that without such a baseline, estimates of model performance can be dramatically overestimated, and we demonstrate this with multiple case studies using real and simulated data. △ Less

Submitted 11 August, 2023; originally announced August 2023.

Comments: 22 pages, 1 figure

Software-based Automatic Differentiation is Flawed

Authors: Daniel Johnson, Trevor Maxfield, Yongxu **, Ronald Fedkiw

Abstract: Various software efforts embrace the idea that object oriented programming enables a convenient implementation of the chain rule, facilitating so-called automatic differentiation via backpropagation. Such frameworks have no mechanism for simplifying the expressions (obtained via the chain rule) before evaluating them. As we illustrate below, the resulting errors tend to be unbounded. Various software efforts embrace the idea that object oriented programming enables a convenient implementation of the chain rule, facilitating so-called automatic differentiation via backpropagation. Such frameworks have no mechanism for simplifying the expressions (obtained via the chain rule) before evaluating them. As we illustrate below, the resulting errors tend to be unbounded. △ Less

Submitted 5 May, 2023; originally announced May 2023.

Achieving and Understanding Out-of-Distribution Generalization in Systematic Reasoning in Small-Scale Transformers

Authors: Andrew J. Nam, Mustafa Abdool, Trevor Maxfield, James L. McClelland

Abstract: Out-of-distribution generalization (OODG) is a longstanding challenge for neural networks. This challenge is quite apparent in tasks with well-defined variables and rules, where explicit use of the rules could solve problems independently of the particular values of the variables, but networks tend to be tied to the range of values sampled in their training data. Large transformer-based language m… ▽ More Out-of-distribution generalization (OODG) is a longstanding challenge for neural networks. This challenge is quite apparent in tasks with well-defined variables and rules, where explicit use of the rules could solve problems independently of the particular values of the variables, but networks tend to be tied to the range of values sampled in their training data. Large transformer-based language models have pushed the boundaries on how well neural networks can solve previously unseen problems, but their complexity and lack of clarity about the relevant content in their training data obfuscates how they achieve such robustness. As a step toward understanding how transformer-based systems generalize, we explore the question of OODG in small scale transformers trained with examples from a known distribution. Using a reasoning task based on the puzzle Sudoku, we show that OODG can occur on a complex problem if the training set includes examples sampled from the whole distribution of simpler component tasks. Successful generalization depends on carefully managing positional alignment when absolute position encoding is used, but we find that suppressing sensitivity to absolute positions overcomes this limitation. Taken together our results represent a small step toward understanding and promoting systematic generalization in transformers. △ Less

Submitted 13 December, 2022; v1 submitted 6 October, 2022; originally announced October 2022.

Mirror Symmetry and Partition Functions

Authors: Travis Maxfield, David R. Morrison, M. Ronen Plesser

Abstract: Localization methods have produced explicit expressions for the sphere partition functions of (2,2) superconformal field theories. The mirror symmetry conjecture predicts an IR duality between pairs of Abelian gauged linear sigma models, a class of which describe families of Calabi-Yau manifolds realizable as complete intersections in toric varieties. We investigate this prediction for the sphere… ▽ More Localization methods have produced explicit expressions for the sphere partition functions of (2,2) superconformal field theories. The mirror symmetry conjecture predicts an IR duality between pairs of Abelian gauged linear sigma models, a class of which describe families of Calabi-Yau manifolds realizable as complete intersections in toric varieties. We investigate this prediction for the sphere partition functions and find agreement between that of a model and its mirror up to the scheme-dependent ambiguities inherent in the definitions of these quantities. △ Less

Submitted 26 October, 2021; v1 submitted 14 February, 2019; originally announced February 2019.

Comments: v3: corrected an equation

(2,2) Geometry from Gauge Theory

Authors: João Caldeira, Travis Maxfield, Savdeep Sethi

Abstract: Using gauge theory, we describe how to construct generalized Kahler geometries with (2,2) two-dimensional supersymmetry, which are analogues of familiar examples like projective spaces and Calabi-Yau manifolds. For special cases, T-dual descriptions can be found which are squashed Kahler spaces. We explore the vacuum structure of these gauge theories by studying the Coulomb branch, which usually e… ▽ More Using gauge theory, we describe how to construct generalized Kahler geometries with (2,2) two-dimensional supersymmetry, which are analogues of familiar examples like projective spaces and Calabi-Yau manifolds. For special cases, T-dual descriptions can be found which are squashed Kahler spaces. We explore the vacuum structure of these gauge theories by studying the Coulomb branch, which usually encodes the quantum cohomology ring. Some models without Kahler dual descriptions possess unusual Coulomb branches. Specifically, there appear to be an infinite number of supersymmetric vacua. △ Less

Submitted 16 October, 2018; v1 submitted 2 October, 2018; originally announced October 2018.

Comments: 50 pages; LaTeX; 5 figures; minor changes

Report number: EFI-18-14

Journal ref: JHEP 1811 (2018) 201

DBI from Gravity

Authors: Travis Maxfield, Savdeep Sethi

Abstract: We study the dynamics of gravitational lumps. By a lump, we mean a metric configuration that asymptotes to a flat space-time. Such lumps emerge in string theory as strong coupling descriptions of D-branes. We provide a physical argument that the broken global symmetries of such a background, generated by certain large diffeomorphisms, constrain the dynamics of localized modes. These modes include… ▽ More We study the dynamics of gravitational lumps. By a lump, we mean a metric configuration that asymptotes to a flat space-time. Such lumps emerge in string theory as strong coupling descriptions of D-branes. We provide a physical argument that the broken global symmetries of such a background, generated by certain large diffeomorphisms, constrain the dynamics of localized modes. These modes include the translation zero modes and any localized tensor modes. The constraints we find are gravitational analogues of those found in brane physics. For the example of a Taub-NUT metric in eleven-dimensional supergravity, we argue that a critical value for the electric field arises from standard gravity without higher derivative interactions. △ Less

Submitted 6 January, 2017; v1 submitted 1 December, 2016; originally announced December 2016.

Comments: 31 pages, LaTeX; references added

Supergravity Backgrounds for Four-Dimensional Maximally Supersymmetric Yang-Mills

Authors: Travis Maxfield

Abstract: In this note, we describe supersymmetric backgrounds for the four-dimensional maximally supersymmetric Yang-Mills theory. As an extension of the method of Festuccia and Seiberg to sixteen supercharges in four dimensions, we utilize the coupling of the gauge theory to maximally extended conformal supergravity. Included among the fields of the conformal supergravity multiplet is the complexified cou… ▽ More In this note, we describe supersymmetric backgrounds for the four-dimensional maximally supersymmetric Yang-Mills theory. As an extension of the method of Festuccia and Seiberg to sixteen supercharges in four dimensions, we utilize the coupling of the gauge theory to maximally extended conformal supergravity. Included among the fields of the conformal supergravity multiplet is the complexified coupling parameter of the gauge theory; therefore, backgrounds with spacetime varying coupling--such as appear in F-theory and Janus configurations--are naturally included in this formalism. We demonstrate this with a few examples from past literature. △ Less

Submitted 6 March, 2017; v1 submitted 19 September, 2016; originally announced September 2016.

Comments: 22 pages, LaTeX; v2: typos corrected; v3: reference added, published version

A Landscape of Field Theories

Authors: Travis Maxfield, Daniel Robbins, Savdeep Sethi

Abstract: Studying a quantum field theory involves a choice of space-time manifold and a choice of background for any global symmetries of the theory. We argue that many more choices are possible when specifying the background. In the context of branes in string theory, the additional data corresponds to a choice of supergravity tensor fluxes. We propose the existence of a landscape of field theory backgrou… ▽ More Studying a quantum field theory involves a choice of space-time manifold and a choice of background for any global symmetries of the theory. We argue that many more choices are possible when specifying the background. In the context of branes in string theory, the additional data corresponds to a choice of supergravity tensor fluxes. We propose the existence of a landscape of field theory backgrounds, characterized by the space-time metric, global symmetry background and a choice of tensor fluxes. As evidence for this landscape, we study the supersymmetric six-dimensional (2,0) theory compactified to two dimensions. Different choices of metric and flux give rise to distinct two-dimensional theories, which can preserve differing amounts of supersymmetry. △ Less

Submitted 12 December, 2015; originally announced December 2015.

Comments: 33 pages, LaTeX

Constraining de Sitter Space in String Theory

Authors: David Kutasov, Travis Maxfield, Ilarion Melnikov, Savdeep Sethi

Abstract: We argue that the heterotic string does not have classical vacua corresponding to de Sitter space-times of dimension four or higher. The same conclusion applies to type II vacua in the absence of RR fluxes. Our argument extends prior supergravity no-go results to regimes of high curvature. We discuss the interpretation of the heterotic result from the perspective of dual type II orientifold constr… ▽ More We argue that the heterotic string does not have classical vacua corresponding to de Sitter space-times of dimension four or higher. The same conclusion applies to type II vacua in the absence of RR fluxes. Our argument extends prior supergravity no-go results to regimes of high curvature. We discuss the interpretation of the heterotic result from the perspective of dual type II orientifold constructions. Our result suggests that the genericity arguments used in string landscape discussions should be viewed with caution. △ Less

Submitted 10 May, 2015; v1 submitted 31 March, 2015; originally announced April 2015.

Comments: 12 pages; LaTeX; references added

Journal ref: Phys. Rev. Lett. 115, 071305 (2015)

Domain Walls, Triples and Acceleration

Authors: Travis Maxfield, Savdeep Sethi

Abstract: We present a construction of domain walls in string theory. The domain walls can bridge both Minkowski and AdS string vacua. A key ingredient in the construction are novel classical Yang-Mills configurations, including instantons, which interpolate between toroidal Yang-Mills vacua. Our construction provides a concrete framework for the study of inflating metrics in string theory. In some cases, t… ▽ More We present a construction of domain walls in string theory. The domain walls can bridge both Minkowski and AdS string vacua. A key ingredient in the construction are novel classical Yang-Mills configurations, including instantons, which interpolate between toroidal Yang-Mills vacua. Our construction provides a concrete framework for the study of inflating metrics in string theory. In some cases, the accelerating space-time comes with a holographic description. The general form of the holographic dual is a field theory with parameters that vary over space-time. △ Less

Submitted 9 April, 2014; originally announced April 2014.

Comments: 63 pages, 1 figure; LaTeX

Report number: EFI-13-33

New Examples of Flux Vacua

Authors: Travis Maxfield, Jock McOrist, Daniel Robbins, Savdeep Sethi

Abstract: Type IIB toroidal orientifolds are among the earliest examples of flux vacua. By applying T-duality, we construct the first examples of massive IIA flux vacua with Minkowski space-times, along with new examples of type IIA flux vacua. The backgrounds are surprisingly simple with no four-form flux at all. They serve as illustrations of the ingredients needed to build type IIA and massive IIA soluti… ▽ More Type IIB toroidal orientifolds are among the earliest examples of flux vacua. By applying T-duality, we construct the first examples of massive IIA flux vacua with Minkowski space-times, along with new examples of type IIA flux vacua. The backgrounds are surprisingly simple with no four-form flux at all. They serve as illustrations of the ingredients needed to build type IIA and massive IIA solutions with scale separation. To check that these backgrounds are actually solutions, we formulate the complete set of type II supergravity equations of motion in a very useful form that treats the R-R fields democratically. △ Less

Submitted 20 November, 2013; v1 submitted 10 September, 2013; originally announced September 2013.

Comments: 38 pages, LaTeX; references updated; additional minor comments added; published version

Report number: EFI-13-20, DMUS-MP-13/18

The Conformal Anomaly of M5-Branes

Authors: Travis Maxfield, Savdeep Sethi

Abstract: We show that the conformal anomaly for N M5-branes grows like $N^3$. The method we employ relates Coulomb branch interactions in six dimensions to interactions in four dimensions using supersymmetry. This leads to a relation between the six-dimensional conformal anomaly and the conformal anomaly of N=4 Yang-Mills. Along the way, we determine the structure of the four derivative interactions for th… ▽ More We show that the conformal anomaly for N M5-branes grows like $N^3$. The method we employ relates Coulomb branch interactions in six dimensions to interactions in four dimensions using supersymmetry. This leads to a relation between the six-dimensional conformal anomaly and the conformal anomaly of N=4 Yang-Mills. Along the way, we determine the structure of the four derivative interactions for the toroidally compactified (2,0) theory, while encountering interesting novelties in the structure of the six derivative interactions. △ Less

Submitted 29 May, 2012; v1 submitted 9 April, 2012; originally announced April 2012.

Comments: 38 pages, LaTeX; references added