Technical report on f-divergences and f-GAN training properties.
problem Understanding and optimizing f-divergences for GAN training.
method Elementary derivation and detailed expressions of f-divergences and their variational lower bounds.
result Informative properties of f-divergences and f-GAN training, including gradient matching and stability improvements.
This paper studies how key tensor properties are inherited in subtensors of tensor train decompositions.
problem Theoretical development of property inheritance for subtensors in tensor train decompositions.
method Theoretical analysis of incoherence and condition number preservation, and tensor train rank preservation through fiber-wise sampling.
result Key tensor properties (incoherence and condition number) can be well preserved to subtensors formed via fiber-wise sampling.
Machine learning predicts liquid water properties from cluster data.
problem Accuracy of bulk properties from machine-learned potentials is limited by training data.
method Local, atom-centred descriptors enable prediction of bulk properties from cluster data.
result Excellent agreement with experimental and theoretical counterparts of liquid water properties.
Paper introduces attacks to infer GAN training dataset properties.
problem Security and privacy risks of generative models like GANs.
method Proposes a general attack pipeline for two attack scenarios.
result Demonstrates strong performance in inferring GAN training dataset properties.
The paper proposes a method to align AI models using conformal risk control.
problem Aligning AI models to meet end-user requirements in non-generative settings.
method Post-processing a pre-trained model to better align with a subset of functions using conformal risk control.
result A probabilistic guarantee that the resulting conformal interval around a model contains a function approximately satisfying a desired property.
We introduce a new parameterization method for deep learning layers using spectral tensor train decomposition.
problem Efficiency and stability in deep learning models with weight matrix compression.
method Spectral Tensor Train Parameterization (STTP) of weight matrices.
result Improved compression and training stability in neural networks.
This paper proposes a new algorithmic framework, predictor-verifier training, to train neural networks that are verifiable, i.e., networks that provably satisfy some desired input-output properties. The key idea is to simultaneously train two networks: a predictor network that performs the task at hand,e.g., predicting…
Neural networks encode inputs deterministically and categorically, behaving like hash encoders.
problem Understanding the encoding properties of neural networks.
method Analyzed the input space partitioned by ReLU-like activations in neural networks.
result Neural networks can be represented by unique activation patterns, similar to hash encoders.
With access to large datasets, deep neural networks (DNN) have achieved human-level accuracy in image and speech recognition tasks. However, in chemistry, data is inherently small and fragmented. In this work, we develop an approach of using rule-based knowledge for training ChemNet, a transferable and generalizable de…
Monotonic Linear Interpolation property in neural networks persists despite non-convexity.
problem Understanding the geometric properties of neural network loss landscapes.
method Tools from differential geometry to analyze the monotonicity of neural network weights.
result Sufficient conditions for the Monotonic Linear Interpolation property under mean squared error.
Study compares memorization of SimCLR to supervised and random labels training.
problem Understanding memorization in contrastive learning.
method Investigated SimCLR's memorization properties compared to supervised and random labels training.
result SimCLR's memorization is similar to random labels training in terms of training object complexity distribution.
Paper improves molecular property prediction using denoising autoencoders.
problem Limited data for molecular property prediction from 3D structures.
method Pre-training via denoising for learning molecular force fields.
result Achieves new state-of-the-art performance on QM9 dataset.
The paper investigates how activation functions impact the training of Neural ODEs, leading to global convergence.
problem Challenges in training Neural ODEs, particularly gradient computation accuracy and convergence analysis.
method Investigates the impact of activation functions on the training dynamics of Neural ODEs.
result Establishes global convergence of Neural ODEs under gradient descent in overparameterized regimes.
Investigates spectral properties of neural networks, showing invariance under certain conditions.
problem Understanding the spectral evolution and invariance in linear-width neural networks.
method Empirical and theoretical analysis of spectra of weight matrices in high-dimensional settings.
result Spectra of weight matrices are invariant under certain training conditions, with implications for feature learning.
We propose a novel computational strategy for de novo design of molecules with desired properties termed ReLeaSE (Reinforcement Learning for Structural Evolution). Based on deep and reinforcement learning approaches, ReLeaSE integrates two deep neural networks - generative and predictive - that are trained separately b…
We present a deep learning framework for wide-field, content-aware estimation of absorption and scattering coefficients of tissues, called Generative Adversarial Network Prediction of Optical Properties (GANPOP). Spatial frequency domain imaging is used to obtain ground-truth optical properties from in vivo human hands…
The practical success of widely used machine learning (ML) and deep learning (DL) algorithms in Artificial Intelligence (AI) community owes to availability of large datasets for training and huge computational resources. Despite the enormous practical success of AI, these algorithms are only loosely inspired from the b…
Deep models generate geometric objects with global properties.
problem Comparing neural models' global properties from generated samples.
method Training on datasets of reflexive polytopes, comparing different representations.
result Models learn non-trivial global properties of geometric objects.
The paper discusses regularization properties of artificial data for deep learning. Artificial datasets allow to train neural networks in the case of a real data shortage. It is demonstrated that the artificial data generation process, described as injecting noise to high-level features, bears several similarities to e…
Training feedforward neural networks with standard logistic activations is considered difficult because of the intrinsic properties of these sigmoidal functions. This work aims at showing that these networks can be trained to achieve generalization performance comparable to those based on hyperbolic tangent activations…
The study addresses biases in evaluating molecular optimization methods and proposes methods to reduce these biases.
problem Biases in in silico evaluation of molecular optimization methods.
method Discussion and empirical investigation of bias reduction methods for predictor misspecification and sample reuse.
result Empirical investigation of bias reduction methods for predictor misspecification and sample reuse.
New DP training ensures models behave similarly at training and test time.
problem Standard SGD training leads to inconsistent model behavior at training and test time.
method Differentially-Private (DP) training ensures WYSIWYG property through distributional generalization.
result DP training guarantees high-level WYSIWYG property, improving model robustness and privacy.
This paper introduces the MCML approach for empirically studying the learnability of relational properties that can be expressed in the well-known software design language Alloy. A key novelty of MCML is quantification of the performance of and semantic differences among trained machine learning (ML) models, specifical…
Due to the surprisingly good representation power of complex distributions, neural network (NN) classifiers are widely used in many tasks which include natural language processing, computer vision and cyber security. In recent works, people noticed the existence of adversarial examples. These adversarial examples break…
Multitask Gaussian process regression reduces data generation costs for molecular property prediction.
problem Data bottleneck in training surrogate models for molecular properties.
method Multitask Gaussian process regression over heterogeneous data sources (CC and DFT).
result Predicts at CC-level accuracy with over an order of magnitude reduction in data generation cost.
Shallow supervised 1-hidden layer neural networks have a number of favorable properties that make them easier to interpret, analyze, and optimize than their deep counterparts, but lack their representational power. Here we use 1-hidden layer learning problems to sequentially build deep networks layer by layer, which ca…
The paper analyzes the training dynamics of a transformer for next-token prediction.
problem Understanding the non-asymptotic performance of transformers in next-token prediction.
method Characterizes training dataset properties, designs a two-stage training algorithm, and analyzes attention gradient properties.
result Trained transformers converge sub-linearly to max-margin solutions and exhibit linear convergence in cross-entropy loss.
This work presents a novel approach to train invertible linear layers by adding rank-one perturbations.
problem Training invertible linear layers during optimization with gradient-based methods is challenging.
method Train rank-one perturbations and add them to weight matrices infrequently, keeping track of inverses and determinants.
result Invertible linear layers improve mixing and mode separation in normalizing flows.
Three training regimes found for scale-invariant neural networks on the sphere.
problem Training scale-invariant neural networks on the sphere with varying effective learning rate.
method Investigated three regimes of training: convergence, chaotic equilibrium, and divergence.
result Discovered three distinct training regimes with unique characteristics.
Wavelet scattering predicts material properties beyond training data.
problem Predict material properties beyond training data.
method Atomic orbital wavelet scattering transform.
result Extrapolation of material properties achieved.
A standard Variational Autoencoder, with a Euclidean latent space, is structurally incapable of capturing topological properties of certain datasets. To remove topological obstructions, we introduce Diffusion Variational Autoencoders with arbitrary manifolds as a latent space. A Diffusion Variational Autoencoder uses t…
We investigate the problem of machine learning with mislabeled training data. We try to make the effects of mislabeled training better understood through analysis of the basic model and equations that characterize the problem. This includes results about the ability of the noisy model to make the same decisions as the …
We propose a framework for the completely unsupervised learning of latent object properties from their interactions: the perception-prediction network (PPN). Consisting of a perception module that extracts representations of latent object properties and a prediction module that uses those extracted properties to simula…
Greedy training of recursive partitioning estimators faces a computational barrier when the true function doesn't satisfy a specific property.
problem Computational inefficiency of greedy training for recursive partitioning estimators.
method Analysis of greedy training for sparse regression functions over binary features.
result Greedy training requires exponential samples when the true function doesn't satisfy a specific property (MSP), but only logarithmic samples when it does.
Paper proposes f-EBM for training deep EBMs using various f-divergences.
problem Training deep EBMs with intractable partition functions.
method Introduces f-EBM framework and optimization algorithm for any f-divergence.
result f-EBM outperforms contrastive divergence and other f-divergences.
Predict training time without training using a linearized model.
problem Predicting the training time of a deep network without actual training.
method Approximate training dynamics with a linearized model and solve a low-dimensional SDE in function space.
result Predict training time within 20% error margin and 30-45x reduction in cost.
Lectures on deep learning properties in infinite and large-width networks.
problem Understanding deep neural networks in extreme width conditions.
method Analysis of random deep neural networks, connections to linear models, kernels, and Gaussian processes, perturbative and non-perturbative treatments.
result Properties and behaviors of deep neural networks in the infinite-width limit and large-width regime.
PG-EVIKAL refines molecular property predictions using neighbor fusion and evidential neural networks.
problem Improving molecular property predictions using test-time neighbor fusion.
method Adapting evidential neural networks to refine predictions by re-ranking structurally similar neighbors.
result PG-EVIKAL reduces RMSE on 14 out of 16 molecular datasets, improving calibration and sequential refinement.
Adversarial training enhances model transferability without sacrificing accuracy.
problem The principle of minimal information in classification models is challenged by adversarial training.
method Investigation of the dual relationship between adversarial training and information theory.
result Adversarial training improves linear transferability and introduces a trade-off between transferability and source task accuracy.
Study on local elasticity in neural network training, improving detection of class-specific changes.
problem Improving the detection of class-specific changes in neural network training.
method Comprehensive study of local elasticity, proposing a new definition to address limitations.
result New definition of local elasticity more sharply detects class-specific changes in neural network training.
This paper improves adversarial robustness of deep learning models.
problem Vulnerability of machine learning models to adversarial perturbations.
method Analyzes adversarial training for linear regression and neural networks, incorporating L1 penalty.
result Incorporating L1 penalty leads to consistent adversarially robust estimation in high-dimensional settings.
Deep neural networks are often trained in the over-parametrized regime (i.e. with far more parameters than training examples), and understanding why the training converges to solutions that generalize remains an open problem. Several studies have highlighted the fact that the training procedure, i.e. mini-batch Stochas…
Three important properties of a classification machinery are: (i) the system preserves the core information of the input data; (ii) the training examples convey information about unseen data; and (iii) the system is able to treat differently points from different classes. In this work we show that these fundamental pro…
This paper seeks to answer the question: as the (near-) orthogonality of weights is found to be a favorable property for training deep convolutional neural networks, how can we enforce it in more effective and easy-to-use ways? We develop novel orthogonality regularizations on training deep CNNs, utilizing various adva…
The finding that very large networks can be trained efficiently and reliably has led to a paradigm shift in computer vision from engineered solutions to learning formulations. As a result, the research challenge shifts from devising algorithms to creating suitable and abundant training data for supervised learning. How…
Neural networks and linear systems linked, revealing training loss and kernel limitations.
problem Exploring the training loss and limitations of neural networks and their kernels.
method Drawing connections between neural networks and under-determined linear systems, providing lower bounds, and analyzing gradient descent.
result Zero training loss achievable for neural networks under certain conditions, but not for ReLU kernels.
Study stabilizes adversarial training in neural networks over infinite-dimensional spaces.
problem Stability issues in adversarial training of neural networks.
method Functional analysis of minimax optimization over infinite-dimensional spaces of continuous functions and probability measures.
result Convergence property of minimax problems under certain conditions, interpreted as stabilization techniques.
Symbolic regression improved by incorporating prior knowledge.
problem Insufficient guidance from training data alone for model accuracy.
method Multi-objective symbolic regression combining training data and prior constraints.
result Models that fit training data well and comply with prior knowledge.