{"total":88,"rows":[{"id":"2110.05025","title":"Self-supervised Learning is More Robust to Dataset Imbalance","created":"2021-10-11","cited_by_count":123,"area":null},{"id":"2203.12221","title":"Modality Competition: What Makes Joint Training of Multi-modal Network Fail in Deep Learning? (Provably)","created":"2022-03-23","cited_by_count":122,"area":null},{"id":"2106.06530","title":"Label Noise SGD Provably Prefers Flat Global Minimizers","created":"2021-06-11","cited_by_count":121,"area":null},{"id":"2106.09524","title":"Implicit Bias of SGD for Diagonal Linear Networks: a Provable Benefit of Stochasticity","created":"2021-06-17","cited_by_count":110,"area":null},{"id":"2110.06914","title":"What Happens after SGD Reaches Zero Loss? --A Mathematical Framework","created":"2021-10-13","cited_by_count":107,"area":null},{"id":"2206.07085","title":"Understanding the Generalization Benefit of Normalization Layers: Sharpness Reduction","created":"2022-06-14","cited_by_count":86,"area":null},{"id":"2208.12591","title":"On the Implicit Bias in Deep-Learning Algorithms","created":"2022-08-26","cited_by_count":86,"area":null},{"id":"2210.14199","title":"Same Pre-training Loss, Better Downstream: Implicit Bias Matters for Language Models","created":"2022-10-25","cited_by_count":84,"area":null},{"id":"2311.18817","title":"Dichotomy of Early and Late Phase Implicit Biases Can Provably Induce Grokking","created":"2023-11-30","cited_by_count":80,"area":null},{"id":"2308.16898","title":"Transformers as Support Vector Machines","created":"2023-08-31","cited_by_count":74,"area":null},{"id":"2210.05337","title":"SGD with Large Step Sizes Learns Sparse Features","created":"2022-10-11","cited_by_count":68,"area":null},{"id":"2207.02628","title":"The alignment property of SGD noise and how it helps select flat minima: A stability analysis","created":"2022-07-06","cited_by_count":64,"area":null},{"id":"2109.14119","title":"Stochastic Training is Not Necessary for Generalization","created":"2021-09-29","cited_by_count":63,"area":null},{"id":"2306.13596","title":"Max-Margin Token Selection in Attention Mechanism","created":"2023-06-23","cited_by_count":58,"area":null},{"id":"2304.00488","title":"Saddle-to-Saddle Dynamics in Diagonal Linear Networks","created":"2023-04-02","cited_by_count":57,"area":null},{"id":"2202.02831","title":"Anticorrelated Noise Injection for Improved Generalization","created":"2022-02-06","cited_by_count":42,"area":null},{"id":"2507.07101","title":"Small Batch Size Training for Language Models: When Vanilla SGD Works, and Why Gradient Accumulation Is Wasteful","created":"2025-07-09","cited_by_count":41,"area":null},{"id":"2207.05931","title":"Towards understanding how momentum improves generalization in deep learning","created":"2022-07-13","cited_by_count":40,"area":null},{"id":"2102.05375","title":"Strength of Minibatch Noise in SGD","created":"2021-02-10","cited_by_count":39,"area":null},{"id":"2207.04036","title":"Implicit Bias of Gradient Descent on Reparametrized Models: On Equivalence to Mirror Descent","created":"2022-07-08","cited_by_count":39,"area":null},{"id":"2306.04251","title":"Stochastic Collapse: How Gradient Noise Attracts SGD Dynamics Towards Simpler Subnetworks","created":"2023-06-07","cited_by_count":39,"area":null},{"id":"2307.11007","title":"Sharpness Minimization Algorithms Do Not Only Minimize Sharpness To Achieve Better Generalization","created":"2023-07-20","cited_by_count":39,"area":null},{"id":"2301.12832","title":"Deep networks for system identification: a Survey","created":"2023-01-30","cited_by_count":37,"area":null},{"id":"2302.08982","title":"(S)GD over Diagonal Linear Networks: Implicit Regularisation, Large Stepsizes and Edge of Stability","created":"2023-02-17","cited_by_count":36,"area":null},{"id":"2303.02749","title":"Revisiting the Noise Model of Stochastic Gradient Descent","created":"2023-03-05","cited_by_count":36,"area":null},{"id":"2407.12332","title":"Why Do You Grok? A Theoretical Analysis of Grokking Modular Addition","created":"2024-07-17","cited_by_count":36,"area":null},{"id":"2309.07418","title":"A Fast Optimization View: Reformulating Single Layer Attention in LLM Based on Tensor and SVM Trick, and Solving It in Matrix Multiplication Time","created":"2023-09-14","cited_by_count":30,"area":null},{"id":"2206.09841","title":"Label noise (stochastic) gradient descent implicitly solves the Lasso for quadratic parametrisation","created":"2022-06-20","cited_by_count":29,"area":null},{"id":"2208.14673","title":"Incremental Learning in Diagonal Linear Networks","created":"2022-08-31","cited_by_count":28,"area":null},{"id":"2201.11729","title":"Implicit Regularization in Hierarchical Tensor Factorization and Deep Convolutional Neural Networks","created":"2022-01-27","cited_by_count":27,"area":null},{"id":"2103.04902","title":"Stochasticity helps to navigate rough landscapes: comparing gradient-descent-based algorithms in the phase retrieval problem","created":"2021-03-08","cited_by_count":23,"area":null},{"id":"2403.08699","title":"Implicit Regularization of Gradient Flow on One-Layer Softmax Attention","created":"2024-03-13","cited_by_count":19,"area":null},{"id":"2102.12430","title":"Noisy Gradient Descent Converges to Flat Minima for Nonconvex Matrix Factorization","created":"2021-02-24","cited_by_count":15,"area":null},{"id":"2107.04518","title":"Optimal Gradient-based Algorithms for Non-concave Bandit Optimization","created":"2021-07-09","cited_by_count":15,"area":null},{"id":"2310.08513","title":"How connectivity structure shapes rich and lazy learning in neural circuits","created":"2023-10-12","cited_by_count":15,"area":null},{"id":"2403.06925","title":"Transformers Learn Low Sensitivity Functions: Investigations and Implications","created":"2024-03-11","cited_by_count":14,"area":null},{"id":"2212.02083","title":"On the Overlooked Structure of Stochastic Gradients","created":"2022-12-05","cited_by_count":13,"area":null},{"id":"2302.09693","title":"mSAM: Micro-Batch-Averaged Sharpness-Aware Minimization","created":"2023-02-19","cited_by_count":10,"area":null},{"id":"2306.08590","title":"Beyond Implicit Bias: The Insignificance of SGD Noise in Online Learning","created":"2023-06-14","cited_by_count":10,"area":null},{"id":"2107.06466","title":"Going Beyond Linear RL: Sample Efficient Neural Function Approximation","created":"2021-07-14","cited_by_count":9,"area":null},{"id":"2301.13703","title":"Dissecting the Effects of SGD Noise in Distinct Regimes of Deep Learning","created":"2023-01-31","cited_by_count":9,"area":null},{"id":"2310.00692","title":"A Theoretical Analysis of Noise Geometry in Stochastic Gradient Descent","created":"2023-10-01","cited_by_count":9,"area":null},{"id":"2310.02396","title":"Inductive biases of multi-task learning and finetuning: multiple regimes of feature reuse","created":"2023-10-03","cited_by_count":9,"area":null},{"id":"2403.05293","title":"Leveraging Continuous Time to Understand Momentum When Training Diagonal Linear Networks","created":"2024-03-08","cited_by_count":9,"area":null},{"id":"2103.13462","title":"Why Do Local Methods Solve Nonconvex Problems?","created":"2021-03-24","cited_by_count":8,"area":null},{"id":"2406.11110","title":"How Neural Networks Learn the Support is an Implicit Regularization Effect of SGD","created":"2024-06-17","cited_by_count":8,"area":null},{"id":"2409.09745","title":"The Optimality of (Accelerated) SGD for High-Dimensional Quadratic Optimization","created":"2024-09-15","cited_by_count":8,"area":null},{"id":"2202.13328","title":"Thinking Outside the Ball: Optimal Learning with Gradient Descent for Generalized Linear Stochastic Convex Optimization","created":"2022-02-27","cited_by_count":7,"area":null},{"id":"2206.07892","title":"Max-Margin Works while Large Margin Fails: Generalization without Uniform Convergence","created":"2022-06-16","cited_by_count":6,"area":null},{"id":"2401.11600","title":"Understanding the Generalization Benefits of Late Learning Rate Decay","created":"2024-01-21","cited_by_count":6,"area":null},{"id":"2106.06153","title":"Towards Understanding Generalization via Decomposing Excess Risk Dynamics","created":"2021-06-11","cited_by_count":5,"area":null},{"id":"2209.09162","title":"On the Theoretical Properties of Noise Correlation in Stochastic Optimization","created":"2022-09-19","cited_by_count":5,"area":null},{"id":"2210.16400","title":"Flatter, faster: scaling momentum for optimal speedup of SGD","created":"2022-10-28","cited_by_count":5,"area":null},{"id":"2212.04343","title":"Improved Deep Neural Network Generalization Using m-Sharpness-Aware Minimization","created":"2022-12-07","cited_by_count":5,"area":null},{"id":"2512.23422","title":"Entropy-Guided Token Dropout: Training Autoregressive Language Models with Limited Domain Data","created":"2025-12-29","cited_by_count":5,"area":null},{"id":"2202.09885","title":"On Optimal Early Stopping: Over-informative versus Under-informative Parametrization","created":"2022-02-20","cited_by_count":4,"area":null},{"id":"2410.16401","title":"Simplicity Bias via Global Convergence of Sharpness Minimization","created":"2024-10-21","cited_by_count":4,"area":null},{"id":"2509.18766","title":"Diagonal Linear Networks and the Lasso Regularization Path","created":"2025-09-23","cited_by_count":4,"area":null},{"id":"2510.17526","title":"How Does Label Noise Gradient Descent Improve Generalization in the Low SNR Regime?","created":"2025-10-20","cited_by_count":4,"area":null},{"id":"2511.03548","title":"Flat Minima and Generalization: Insights from Stochastic Convex Optimization","created":"2025-11-05","cited_by_count":4,"area":null},{"id":"2604.09258","title":"Nexus: Same Pretraining Loss, Better Downstream Generalization via Common Minima","created":"2026-04-10","cited_by_count":4,"area":null},{"id":"2605.26895","title":"Negligible in Size, Significant in Effect: On Scale Vectors in Large Language Models","created":"2026-05-26","cited_by_count":4,"area":null},{"id":"2112.09741","title":"Envisioning Future Deep Learning Theories: Some Basic Concepts and Characteristics","created":"2021-12-17","cited_by_count":3,"area":null},{"id":"2207.07612","title":"Blessing of Nonconvexity in Deep Linear Models: Depth Flattens the Optimization Landscape Around the True Solution","created":"2022-07-15","cited_by_count":3,"area":null},{"id":"2311.00274","title":"Generalization Bounds for Label Noise Stochastic Gradient Descent","created":"2023-11-01","cited_by_count":3,"area":null},{"id":"2407.02322","title":"Stochastic Differential Equations models for Least-Squares Stochastic Gradient Descent","created":"2024-07-02","cited_by_count":3,"area":null},{"id":"2601.06597","title":"Understanding and inverse design of implicit bias in stochastic learning: a geometric perspective","created":"2026-01-10","cited_by_count":3,"area":null},{"id":"2210.13599","title":"Noise Injection as a Probe of Deep Learning Dynamics","created":"2022-10-24","cited_by_count":2,"area":null},{"id":"2308.03215","title":"The Effect of SGD Batch Size on Autoencoder Learning: Sparsity, Sharpness, and Feature Learning","created":"2023-08-06","cited_by_count":2,"area":null},{"id":"2402.08344","title":"Implicit Bias in Noisy-SGD: With Applications to Differentially Private Training","created":"2024-02-13","cited_by_count":2,"area":null},{"id":"2403.01420","title":"The Implicit Bias of Heterogeneity towards Invariance: A Study of Multi-Environment Matrix Sensing","created":"2024-03-03","cited_by_count":2,"area":null},{"id":"2502.11893","title":"Rethinking Benign Overfitting in Two-Layer Neural Networks","created":"2025-02-17","cited_by_count":2,"area":null},{"id":"2509.18001","title":"Unveiling m-Sharpness Through the Structure of Stochastic Gradient Noise","created":"2025-09-22","cited_by_count":2,"area":null},{"id":"2510.01930","title":"Precise Dynamics of Diagonal Linear Networks: A Unifying Analysis by Dynamical Mean-Field Theory","created":"2025-10-02","cited_by_count":2,"area":null},{"id":"2602.05600","title":"On the Superlinear Relationship between SGD Noise Covariance and Loss Landscape Curvature","created":"2026-02-05","cited_by_count":2,"area":null},{"id":"2510.05482","title":"ATOM: A Pretrained Neural Operator for Multitask Molecular Dynamics","created":"2025-10-07","cited_by_count":1,"area":null},{"id":"2601.10962","title":"Noise-Driven Exploration and Transient Freezing Select Flat Minima in Stochastic Gradient Descent","created":"2026-01-16","cited_by_count":1,"area":null},{"id":"2602.05065","title":"Does SGD Seek Flatness or Sharpness? An Exactly Solvable Model","created":"2026-02-04","cited_by_count":1,"area":null},{"id":"2602.05393","title":"Late-to-Early Training: LET LLMs Learn Earlier, So Faster and Better","created":"2026-02-05","cited_by_count":1,"area":null},{"id":"2604.21016","title":"SGD at the Edge of Stability: The Stochastic Sharpness Gap","created":"2026-04-22","cited_by_count":1,"area":null},{"id":"2202.03535","title":"Noise Regularizes Over-parameterized Rank One Matrix Recovery, Provably","created":"2022-02-07","cited_by_count":0,"area":null},{"id":"2406.16032","title":"Effect of Random Learning Rate: Theoretical Analysis of SGD Dynamics in Non-Convex Optimization via Stationary Distribution","created":"2024-06-23","cited_by_count":0,"area":null},{"id":"2509.26015","title":"Indirect Attention: Turning Context Misalignment into a Feature","created":"2025-09-30","cited_by_count":0,"area":null},{"id":"2603.10397","title":"On the Learning Dynamics of Two-layer Linear Networks with Label Noise SGD","created":"2026-03-11","cited_by_count":0,"area":null},{"id":"2605.02658","title":"Deciphering Shortcut Learning from an Evolutionary Game Theory Perspective","created":"2026-05-04","cited_by_count":0,"area":null},{"id":"2605.20642","title":"Same Target, Different Basins: Hard vs. Soft Labels for Annotator Distributions","created":"2026-05-20","cited_by_count":0,"area":null},{"id":"2608.09840","title":"Langevin dynamics along the zero set of real-analytic potentials","created":"2026-08-10","cited_by_count":0,"area":null},{"id":"2609.02373","title":"Percolation Dynamics in Optimization: Variance Cascades and Nested Symmetry","created":"2026-09-02","cited_by_count":0,"area":null}],"next_cursor":null}