Gradient descent learns useful features even in the NTK regime.
problem The ability of neural networks to learn useful features.
method Local convergence analysis of gradient descent with regularization.
result Gradient descent can capture ground-truth directions for feature learning even after the loss threshold is reached.
Gradient descent amplifies random features in neural networks to useful ones.
problem Generalization in neural networks trained on corrupted data.
method Characterization of feature-learning process in two-layer ReLU networks trained by gradient descent.
result Gradient descent amplifies random features to useful ones, achieving near optimal generalization error.
Gradient Starvation causes neural networks to focus on a subset of features, leading to imbalance.
problem Neural networks focusing on only a subset of features despite having access to more.
method Using Dynamical Systems theory, identified properties of learning dynamics leading to feature imbalance.
result Gradient descent can lead to feature imbalance in over-parameterized networks.
Proposes using gradients as features for efficient deep learning adaptation.
problem Efficient deep representation learning for different tasks.
method Designs a linear model incorporating gradients and activations of a pre-trained network.
result Shows strong results across various tasks and datasets.
In this paper we introduce Feature Gradients, a gradient-based search algorithm for feature selection. Our approach extends a recent result on the estimation of learnability in the sublinear data regime by showing that the calculation can be performed iteratively (i.e., in mini-batches) and in linear time and space wit…
Gradient descent biases neural networks to use an average of features, leading to non-robustness.
problem Non-robustness in neural networks due to feature averaging.
method Theoretical analysis and experiments on binary classification tasks.
result Gradient descent trains networks to rely on an average of features, making them vulnerable to adversarial attacks.
A feature selection algorithm should ideally satisfy four conditions: reliably extract relevant features; be able to identify non-linear feature interactions; scale linearly with the number of features and dimensions; allow the incorporation of known sparsity structure. In this work we propose a novel feature selection…
Gradient descent with growing learning rate enables learning non-linear features in neural networks.
problem Learning non-linear features in two-layer neural networks.
method Using gradient descent with a learning rate that grows with the sample size.
result Multiple rank-one components emerge, each corresponding to a specific polynomial feature.
Kernel ridgeless regression with random features shows good generalization without explicit regularization.
problem Generalization of kernel ridgeless regression without explicit regularization.
method Investigation of ridgeless regression with random features and stochastic gradient descent, exploring the effect of random features error and spectral density optimization.
result Random features error exhibits the double-descent curve, leading to improved generalization.
Greedy PIG adapts integrated gradients for better feature attribution.
problem Interpreting deep learning model predictions.
method Unified discrete optimization framework for feature attribution and selection.
result Greedy PIG improves feature attribution on various tasks.
Study on test risk dynamics in learning theory with stochastic gradient flow.
problem Understanding test risk in stochastic gradient flow dynamics.
method Path integral formulation for small learning rates, explicit computation for weak features.
result Explicit corrections due to stochastic term in dynamics, good agreement with simulations.
GENIE balances domain-invariant feature learning and gradient alignment for improved DG performance.
problem Domain Generalization (DG) overfitting to domain-specific features
method GENIE (Generalization-ENhancing Iterative Equalizer) optimizer
result Prevents a small subset of parameters from dominating optimization, promoting domain-invariant feature learning
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.
NM-PPG optimizes adaptive feature acquisition in POMDPs for better predictions.
problem Optimizing adaptive feature acquisition in prediction problems with costly features.
method Non-myopic pathwise policy gradients (NM-PPG) with continuous relaxation and straight-through rollout.
result NM-PPG outperforms state-of-the-art AFA methods on synthetic and real-world datasets.
Wide neural networks with asymmetrical node scaling converge globally and learn features.
problem Global convergence and feature learning in over-parameterised shallow networks.
method Gradient-based optimisation of wide, shallow neural networks with asymmetrical node scaling.
result Gradient flow and gradient descent converge to a global minimum and learn features, unlike in the NTK parameterisation.
RIG extends IG to Riemannian manifolds for explainable AI.
problem Lack of explainability in AI models.
method Extension of Integrated Gradients to Riemannian manifolds.
result RIG restricts to IG in Euclidean space.
In this paper we present CatBoost, a new open-sourced gradient boosting library that successfully handles categorical features and outperforms existing publicly available implementations of gradient boosting in terms of quality on a set of popular publicly available datasets. The library has a GPU implementation of lea…
Proposes new attribution methods for trees with regularization.
problem Feature attribution for trees trained with regularization.
method Prediction Decomposition Attribution (PreDecomp) and TreeInner.
result TreeInner shows state-of-the-art feature selection performance.
Distributed learning with random features and gradient descent improves performance and reduces memory usage.
problem Improving generalization in decentralized learning with limited memory.
method Distributed Gradient Descent with Random Features and Implicit Regularization.
result High probability bounds on predictive performance with optimal statistical rates.
Study tests if input gradients highlight discriminative features, finds they often fail.
problem Validity of assumption that input gradients highlight discriminative features in model predictions.
method Developed DiffROAR framework and BlockMNIST dataset to test assumption on four benchmarks.
result Input gradients of standard models often fail to highlight discriminative features, while robust models do.
Time series data constitutes a distinct and growing problem in machine learning. As the corpus of time series data grows larger, deep models that simultaneously learn features and classify with these features can be intractable or suboptimal. In this paper, we present feature learning via long short term memory (LSTM) …
Paper explores how feature interactions improve XGBoost models.
problem Improving XGBoost model performance and interpretability.
method Identifies feature interactions as constraints in XGBoost models.
result Accurate identification of feature interactions leads to significant model performance improvements.
Study examines how two-layer networks learn features after one gradient step.
problem Understanding feature learning in neural networks after a single gradient descent step.
method Modeling the trained network as a spiked Random Features (sRF) model and leveraging Gaussian universality.
result Exact asymptotic description of the generalization error of the sRF in high-dimensional limit.
Adapts IG for better feature attributions and robustness.
problem Reliability concerns in feature attributions for deep learning models.
method Adaptation of path-based feature attribution to Riemannian geometry of data manifolds.
result IG along geodesics generates more intuitive and robust explanations.
Study reveals conditions for neural networks to forget learned features.
problem Understanding feature unlearning in neural networks.
method Infinite-width limit analysis with stochastic gradient descent, fast-slow dynamics.
result Conditions for feature unlearning are determined by the strength of nonlinear terms and initial weights.
One gradient step improves neural network feature learning by aligning weights with the teacher model.
problem Improving feature learning in neural networks through gradient descent.
method First gradient descent step on the first-layer parameters of a two-layer neural network.
result The first gradient update contains a rank-1 spike, leading to alignment with the teacher model.
Scalable feature selection improves GBDT model training speed.
problem Efficiently selecting features for GBDT models in high-dimensional settings.
method Developed a scalable forward feature selection method using group testing.
result Significant speedups in training time with competitive model performance.
A neural network approach for feature selection using mutual information.
problem Feature ranking and selection leading to sub-optimal solutions for class separability.
method Stochastic mutual information gradient estimation for dimensionality reduction.
result The network projects features onto an output space maximizing mutual information with class labels.
Gradient descent aligns neural feature matrices with pre-activation tangent features.
problem Understanding neural feature learning mechanisms.
method Analytical proof of alignment between weight matrices and pre-activation tangent features.
result Derivative alignment occurs almost surely in high-dimensional settings.
AGF explains feature learning in neural networks through alternating steps.
problem Understanding what features neural networks learn and how they learn them.
method AGF is an algorithmic framework that approximates the dynamics of feature learning in two-layer networks.
result AGF provides a unified framework to understand feature learning in neural networks, matching experimental results across various architectures.
Improved convergence in deep learning networks via gradient transformation.
problem Reduced convergence speed in deep learning networks.
method Gradient transformation to weight gradients and recursive condition number reduction.
result Improved convergence speed and accuracy in image classification tasks.
PS-IG improves feature attribution by reducing noise and variance.
problem Improving feature attribution in machine learning models.
method Path-sampled integrated gradients (PS-IG) computes expected value over sampled baselines.
result PS-IG reduces attribution variance by a factor of 1/3 under uniform sampling.
Recursive Feature Machines show grokking in modular arithmetic without neural networks.
problem Grokking in modular arithmetic tasks.
method Recursive Feature Machines (RFM) with Average Gradient Outer Product (AGOP).
result RFM and neural networks learn block-circulant features to solve modular arithmetic.
Gradient descent reshapes the function space of neural networks.
problem Understanding how feature learning affects the function space of neural networks.
method Characterized the evolution of the feature space during training using a two-layer neural network.
result Gradient descent induces a data-adaptive deformation that selectively enhances signal-aligned directions.
Contrastive learning struggles with class collapse and feature suppression, revealing bias towards simpler solutions.
problem Contrastive learning struggles with class collapse and feature suppression, especially in supervised and unsupervised settings.
method Unified theoretical framework to determine which features are learnt by CL, revealing bias towards simpler solutions.
result Bias towards simpler solutions is a key factor in class collapse and feature suppression.
Pruning method removes less important features in linear models.
problem Removing less important features in linear models trained by gradient flow.
method Iterative Magnitude Pruning (IMP) applied to linear models trained by gradient flow.
result IMP prunes features with smallest projection onto the data.
PathBoost boosts graph-level predictions using path-based features.
problem Graph-level classification and regression challenges.
method Gradient tree boosting method for graph-level prediction.
result PathBoost outperforms graph neural networks and graph kernel approaches in many cases.
Generative Adversarial Networks (GAN) have shown promising results on a wide variety of complex tasks. Recent experiments show adversarial training provides useful gradients to the generator that helps attain better performance. In this paper, we intend to theoretically analyze whether supervised learning with adversar…
Unified derivation of high-dimensional linear models using stochastic gradient descent.
problem Performance analysis of high-dimensional linear models trained with stochastic gradient descent.
method Derivation of a deterministic equivalence for the two-point function of a random matrix resolvent.
result Unified understanding of model performance including previously known and novel results.
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…
Proposes Geodesic Integrated Gradients (GIG) for more accurate feature attributions in deep networks.
problem Flawed attributions using straight paths from Integrated Gradients (IG).
method Introduces a model-induced Riemannian metric and computes attributions along geodesics.
result GIG produces more faithful attributions than IG on benchmarks.
agtboost speeds up gradient tree boosting with automatic complexity adjustment.
problem Speeding up and simplifying gradient tree boosting computations.
method Adaptive gradient tree boosting with automatic complexity adjustment and feature importance.
result Significant decrease in computation time and simplification of model complexity.
Study predicts SGD test loss for structured features.
problem Understanding test loss dynamics in SGD for structured data.
method Solveable model of SGD on mean square loss for arbitrary covariance structure.
result Simpler Gaussian model accurately predicts test loss of nonlinear models.
This research reverses feature visualization in neural networks to optimize for specific feature objectives.
problem The invertibility of feature visualization in neural networks is not well understood.
method The approach involves optimizing for the feature objective that generates the input used in feature visualization, using the gradient of a specific objective function.
result A closed-form solution is found to minimize the gradient, providing an alternative view on network sensitivity.
SGD in DLNs reveals feature learning dynamics.
problem Understanding SGD dynamics in DLNs during saddle-to-saddle training.
method Stochastic Langevin dynamics with anisotropic, state-dependent noise; one-dimensional per-mode SDEs; Boltzmann distribution approximation.
result SGD noise encodes feature learning progression but does not alter saddle-to-saddle dynamics.
TrIM improves gradient-based dimension reduction and regression.
problem Efficiently identifying relevant feature subspace for high-dimensional regression.
method Introduced TrIM forest, an iterative approach using Mondrian forest and EGOP estimate.
result Consistency guarantees and convergence rates for EGOP matrix and random forest estimator.
We develop a theory of higher-order feature attribution for complex models.
problem Interpreting feature contributions in models with interactions is challenging.
method We extend Integrated Gradients (IG) to higher-order feature attributions.
result We establish natural connections to statistics and topological signal processing.
Develops methods to analyze feature-outcome associations in subpopulations.
problem Challenges in understanding feature-outcome associations in high-dimensional data.
method Geometric decomposition framework using gradient flow and co-monotonicity decomposition.
result Identifies context-dependent patterns and improves statistical power and interpretability.