Paper proposes a new method to measure model sensitivity using final model only.
problem Understanding model behavior using only the final trained model.
method Reframe TDA as measuring sensitivity, propose further training as gold standard, unify gradient-based methods.
result Gradient-based methods approximate further training but vary in quality.
Gradient-based training and pruning for radial basis function networks in materials physics.
problem Interpretable and robust machine learning for materials physics problems.
method Gradient-based training and pruning of radial basis function networks with closed-form optimization criteria.
result Pruned models provide compact and interpretable versions of larger models, offering insights into atom-level migration processes.
In this work we study generalization of neural networks in gradient-based meta-learning by analyzing various properties of the objective landscapes. We experimentally demonstrate that as meta-training progresses, the meta-test solutions, obtained after adapting the meta-train solution of the model, to new tasks via few…
Paper proposes HTAF for stable training of binary neural networks.
problem Challenges in training binary neural networks with gradient-based optimization.
method HTAF is a smooth approximation to the Heaviside function that enables stable training.
result HTAF enables stable training of various binary neural networks with gradient-based optimization.
Generative model initializes 2-layer network weights for small datasets.
problem Approximating functions with 2-layer networks using small datasets and gradient-based training.
method Initialize hidden weights with a learned proposal distribution parameterized as a deep generative model. Refine with gradient-based post-processing and regularization.
result Demonstrates effectiveness of the approach with numerical examples.
GFM models neural network training as a dynamical system to forecast final weights.
problem Computational intensity and inefficiency in training deep neural networks.
method Gradient Flow Matching (GFM) treats training as a dynamical system with learned vector fields.
result GFM achieves forecasting accuracy competitive with Transformer-based models and significantly outperforms classical baselines.
Bayesian Neural Networks are robust to gradient-based attacks in the large-data limit.
problem Vulnerability of deep learning models to adversarial attacks.
method Analysis of adversarial attacks in the large-data, overparametrized limit for Bayesian Neural Networks.
result BNN posteriors are robust to gradient-based adversarial attacks in the limit.
Background: It is still an open research area to theoretically understand why Deep Neural Networks (DNNs)---equipped with many more parameters than training data and trained by (stochastic) gradient-based methods---often achieve remarkably low generalization error. Contribution: We study DNN training by Fourier analysi…
Paper analyzes dataset distillation for efficient encoding of task-relevant information.
problem Efficiently encoding task-relevant information from gradient-based learning of non-linear tasks.
method Theoretical analysis of dataset distillation applied to two-layer neural networks with gradient-based training.
result Low-dimensional structure of the problem is efficiently encoded into distilled data, reproducing a model with high generalization ability.
New SMC method for pBNNs improves scalability and predictive performance.
problem Training pBNNs with high-dimensional stochastic parameters.
method Gradient-based proposals within SMC samplers.
result New method outperforms state-of-the-art in predictive performance and training time.
Paper presents a novel gradient-based method for training models and hyperparameters simultaneously.
problem Achieving generalization in machine learning models.
method A novel gradient-based framework that trains parameters and hyperparameters simultaneously.
result Significantly smaller runtime compared to benchmark methods for equivalent prediction scores.
A method to reduce Hessian matrix calculation cost in gradient-based meta-learning.
problem High memory footprint in calculating Hessian matrix for large-scale applications.
method Multi-step estimation of gradients to reuse the same gradient in a window of inner steps.
result Significant reduction in training time and memory usage with competitive or improved accuracies.
Gradient-based methods can be biased by distributional asymmetries in bivariate categorical data.
problem Gradient-based causal discovery methods can be biased by distributional asymmetries in bivariate categorical data.
method Identified and examined two distributional biases: Marginal Distribution Asymmetry and Marginal Distribution Shift Asymmetry. Employed two simple models to demonstrate and control these biases.
result Gradient-based methods can be biased by distributional asymmetries, and these biases can be controlled.
Neural networks perform differently when regression is treated as classification.
problem Understanding why neural networks perform better when regression is treated as classification.
method Analyzing two-layer ReLU networks and their feature spaces, focusing on the cross entropy loss vs. square loss.
result The support of the measure induced by the square loss differs from that of the cross entropy loss, indicating optimization difficulties.
New method reconstructs significant parts of training data from neural networks.
problem Understanding and reconstructing training data from neural networks.
method Proposes a novel reconstruction scheme based on recent theoretical results about neural network training.
result Shows that a significant fraction of training data can be reconstructed from neural network parameters.
Causal structure learning has been a challenging task in the past decades and several mainstream approaches such as constraint- and score-based methods have been studied with theoretical guarantees. Recently, a new approach has transformed the combinatorial structure learning problem into a continuous one and then solv…
A gradient-based method learns the structure of TAN for Bayesian network classifiers.
problem Learning the structure of Bayesian networks is difficult.
method A distribution over graph structures learned via gradient-based optimization.
result Consistently outperforms random and Chow-Liu TAN structures.
ASTRA improves TDA by more accurately approximating iHVP.
problem Improving insights into training data attribution.
method ASTRA uses EKFAC-preconditioner on Neumann series iterations to accurately approximate iHVP.
result Improving iHVP approximation significantly improves TDA performance.
This work bridges competitive learning with gradient-based learning for faster feature extraction.
problem Lack of powerful feature extractors in competitive learning methods.
method Introduces gradient-based competitive layers for feature extraction.
result Demonstrates theoretical equivalence and faster convergence of gradient-based competitive layers.
Gradient-based method prunes large models to create transferable architectures.
problem Creating transferable architectures from large models with limited fine-tuning data.
method Gradient-based algorithm for architecture pruning and subset selection.
result Successfully retrain architectures on new tasks with few fine-tuning data.
We examine the role of memorization in deep learning, drawing connections to capacity, generalization, and adversarial robustness. While deep networks are capable of memorizing noise data, our results suggest that they tend to prioritize learning simple patterns first. In our experiments, we expose qualitative differen…
We propose Power Slow Feature Analysis, a gradient-based method to extract temporally slow features from a high-dimensional input stream that varies on a faster time-scale, as a variant of Slow Feature Analysis (SFA) that allows end-to-end training of arbitrary differentiable architectures and thereby significantly ext…
New algorithm estimates task affinities without repeated training, improving model performance and efficiency.
problem Efficiently estimating task affinities among multiple tasks for model training.
method Grad-TAG algorithm: trains a base model for all tasks and uses gradient-based linearization to estimate task affinities.
result Estimates task affinities with high accuracy and low computational cost.
We formulate a general framework for competitive gradient-based learning that encompasses a wide breadth of multi-agent learning algorithms, and analyze the limiting behavior of competitive gradient-based learning algorithms using dynamical systems theory. For both general-sum and potential games, we characterize a non…
This work investigates how gradient-based learning performs with structured data, revealing issues and improvements.
problem Gradient-based learning under structured data, particularly with a spiked covariance structure.
method Investigates the effect of a spiked covariance structure on gradient-based feature learning and proposes weight normalization.
result Gradient-based dynamics may fail to recover the true direction in anisotropic settings, but weight normalization can improve performance.
A novel gradient-based method optimizes decision trees for complex tasks.
problem Training decision trees with arbitrary differentiable loss functions.
method Gradient-based optimization using first and second derivatives of loss functions.
result Improves accuracy and flexibility in decision tree optimization.
This paper analyzes deep and wide transformer training dynamics.
problem Understanding the training dynamics of infinitely deep and wide transformers.
method Develops a mean-field framework for gradient-based training of transformers, controlling a neural PDE.
result Establishes a rigorous foundation for gradient-based transformer training, proving convergence to global minima.
Adaptor 'E' extends gradient-based optimizers to explore loss landscapes, improving generalization.
problem Finding lower and better-generalizing minima in deep learning.
method Proposes an adaptor 'E' to extend gradient-based optimizers, encouraging exploration along landscape valleys.
result Adapted optimizers increase test accuracy by an average of 2.5% in large-batch training tasks.
Graph neural networks (GNNs) which apply the deep neural networks to graph data have achieved significant performance for the task of semi-supervised node classification. However, only few work has addressed the adversarial robustness of GNNs. In this paper, we first present a novel gradient-based attack method that fa…
Differentiable pipeline replaces non-differentiable CAE components for shape optimization.
problem Gradient-based optimization is limited by non-differentiable components in CAE workflows.
method Surrogate models replace non-differentiable pipeline components, enabling gradient-based optimization.
result Gradient-based shape optimization possible without differentiable solvers.
We propose a simple change to existing neural network structures for better defending against gradient-based adversarial attacks. Instead of using popular activation functions (such as ReLU), we advocate the use of k-Winners-Take-All (k-WTA) activation, a C0 discontinuous function that purposely invalidates the neural …
New framework improves model robustness by focusing on stable relations across environments.
problem Standard supervised learning fails under data distribution shift.
method Gradient-based learning framework derived from the principle of independent causal mechanisms (ICM).
result Models generalize well to unseen scenarios, ignoring unstable relations.
Generalization in deep neural networks can be analyzed using minimax rates for gradient methods.
problem Generalization performance of over-parameterized neural networks
method Establishing a connection between gradient-based methods and kernel methods
result Deriving minimax-optimal rates for GD and SGD under polynomial network width scaling
Catastrophic forgetting is a problem faced by many machine learning models and algorithms. When trained on one task, then trained on a second task, many machine learning models "forget" how to perform the first task. This is widely believed to be a serious problem for neural networks. Here, we investigate the extent to…
Paper presents a new method for Bayesian deep learning that scales to Atari games.
problem Training neural networks on complex environments like Atari games is challenging.
method Adapted temporal difference Q-learning to work with Bayesian inference.
result TAGI allows for analytical inference of neural network parameters, achieving performance comparable to gradient-based methods.
Tricks adversarial attacks to target specific classes, improving classifier accuracy.
problem Recent adversarial defense approaches have failed to protect classifiers from untargeted attacks.
method Target Training defense tricks untargeted attacks into targeted attacks on designated classes, then derives the real class.
result 86.2% accuracy for CW-L2 (confidence=0) in CIFAR10, outperforming unsecured classifiers.
Combines gradient-based and competitive learning for unsupervised feature extraction.
problem Handling input data without supervision and replicating input manifold topology.
method Integrates gradient-based and competitive learning approaches to learn topological structures.
result The dual competitive layer outperforms the vanilla layer in high-dimensional datasets.
Linear algebra approach for parallel deep learning models.
problem Training large DNNs in distributed environments.
method Linear algebraic approach to model parallelism.
result Manual development of backward operators for gradient-based training.
Empirical studies show that gradient-based methods can learn deep neural networks (DNNs) with very good generalization performance in the over-parameterization regime, where DNNs can easily fit a random labeling of the training data. Very recently, a line of work explains in theory that with over-parameterization and p…
Recent breakthroughs in defenses against adversarial examples, like adversarial training, make the neural networks robust against various classes of attackers (e.g., first-order gradient-based attacks). However, it is an open question whether the adversarially trained networks are truly robust under unknown attacks. In…
New scaling laws explain deep learning performance growth.
problem Understanding neural network performance growth.
method Analyzed entire training dynamics of various architectures.
result Identified two dynamical scaling laws.
The continuous dynamical system approach to deep learning is explored in order to devise alternative frameworks for training algorithms. Training is recast as a control problem and this allows us to formulate necessary optimality conditions in continuous time using the Pontryagin's maximum principle (PMP). A modificati…
Introduces a differentiable approximation to the zero-one loss.
problem Incompatibility of zero-one loss with gradient-based optimization.
method Smooth projection onto hypersimplex through constrained optimization.
result Achieves significant improvements in generalization under large-batch training.
Transformers can handle endogeneity in linear regression using IV methods.
problem Endogeneity in in-context linear regression models.
method Transformer architecture with gradient-based bi-level optimization and in-context pretraining.
result Transformers provide more robust predictions and estimates than 2SLS in endogenous scenarios.
K-Means and RBF networks are shown to be equivalent under certain conditions.
problem Discrete clustering vs. continuous optimization in machine learning.
method Established variational and gradient-based equivalence between K-Means and RBF networks.
result Gradient-based updates of RBF centers recover K-Means centroid update rule.
We consider the task of low-multilinear-rank functional regression, i.e., learning a low-rank parametric representation of functions from scattered real-valued data. Our first contribution is the development and analysis of an efficient gradient computation that enables gradient-based optimization procedures, including…
Method generates time-series attribution maps with identifiability guarantees.
problem Lack of identifiability guarantees in gradient-based attribution methods.
method Regularized contrastive learning algorithm trained on time-series data.
result Empirically shows robust approximation of zero vs. non-zero entries in the ground-truth attribution map.
Proves deep networks can learn hierarchical structures efficiently.
problem Understanding how deep networks learn hierarchical structures in data.
method Random Hierarchy Models, gradient-based methods, layerwise training.
result Proves deep networks can efficiently learn hierarchical structures.