Theory · learning theory papers, 2023–2026

Key findings

• Theory · learning theory went from 86 papers in 2023 to 482 in 2026, roughly 5.6x in three years.
• NeurIPS publishes the most Theory · learning theory work: 540 papers from 2023 to 2026, 40% of the four-venue total.
• The fastest-growing overlap is with Social aspects · accountability transparency and interpretability: papers tagged with both rose from 0 in 2023 to 33 in 2026.
• Among the 5 topics under Theory and Statistics, it ranks #2 by 2026 paper count; the fastest grower in the group, Theory · optimization, is up 4.7x versus 5.6x here.

Representative papers

Spread across years and venues so every edition shows up

• NeurIPS 2026A dynamical systems theory of reward-modulated learning in linear recurrent networks
• ICML 2026A Fine-Grained Understanding of Uniform Convergence for Halfspaces
• ICLR 2026A Law of Data Reconstruction for Random Features (And Beyond)
• NeurIPS 2025A Geometric Analysis of PCA
• ICML 2025A Generalization Theory for Zero-Shot Prediction
• ICLR 2025A Solvable Attention for Neural Scaling Laws
• CVPR 2025Dynamic Stereotype Theory Induced Micro-expression Recognition with Oriented Deformation
• NeurIPS 20244+3 Phases of Compute-Optimal Neural Scaling Laws
• ICML 2024A Statistical Theory of Regularization-Based Continual Learning
• ICLR 2024A Sublinear Adversarial Training Algorithm
• CVPR 2024Accurate Training Data for Occupancy Map Prediction in Automated Driving Using Evidence Theory
• NeurIPS 2023$\varepsilon$-fractional core stability in Hedonic Games.
• ICML 2023Robust Collaborative Learning with Linear Gradient Overhead
• ICLR 2023Bispectral Neural Networks
• NeurIPS 2026A Mechanistic Investigation of Theory of Mind in a Large Language Model
• ICML 2026A Random Matrix Theory of Masked Self-Supervised Learning
• ICLR 2026A New Approach to Controlling Linear Dynamical Systems
• NeurIPS 2025Approximation theory for 1-Lipschitz ResNets
• ICML 2025A Two-Stage Learning-to-Defer Approach for Multi-Task Learning
• ICLR 2025A Theoretical Analysis of Self-Supervised Learning for Vision Transformers

All Theory · learning theory papers

389 papers, page 3 of 8

101. ICLR 2025Optimal Learning of Kernel Logistic Regression for Complex Classification Scenarios
102. ICLR 2025Optimality and Adaptivity of Deep Neural Features for Instrumental Variable Regression
103. ICLR 2025Oracle efficient truncated statistics
104. ICLR 2025Prediction Risk and Estimation Risk of the Ridgeless Least Squares Estimator under General Assumptions on Regression Errors
105. ICLR 2025Progress or Regress? Self-Improvement Reversal in Post-training
106. ICLR 2025Provable Uncertainty Decomposition via Higher-Order Calibration
107. ICLR 2025Provable weak-to-strong generalization via benign overfitting
108. ICLR 2025Quality over Quantity in Attention Layers: When Adding More Heads Hurts
109. ICLR 2025Quantitative Approximation for Neural Operators in Nonlinear Parabolic Equations
110. ICLR 2025Regret Bounds for Episodic Risk-Sensitive Linear Quadratic Regulator
111. ICLR 2025Rethinking Self-Distillation: Label Averaging and Enhanced Soft Label Refinement with Partial Labels
112. ICLR 2025RNNs are not Transformers (Yet): The Key Bottleneck on In-Context Retrieval
113. ICLR 2025Robust Feature Learning for Multi-Index Models in High Dimensions
114. ICLR 2025Robustness Auditing for Linear Regression: To Singularity and Beyond
115. ICLR 2025Satisficing Regret Minimization in Bandits
116. ICLR 2025Second Order Bounds for Contextual Bandits with Function Approximation
117. ICLR 2025Selective Task Group Updates for Multi-Task Optimization
118. ICLR 2025Self-Improvement in Language Models: The Sharpening Mechanism
119. ICLR 2025Semialgebraic Neural Networks: From roots to representations
120. ICLR 2025Separation Power of Equivariant Neural Networks
121. ICLR 2025Sharper Guarantees for Learning Neural Network Classifiers with Gradient Methods
122. ICLR 2025Sketching for Convex and Nonconvex Regularized Least Squares with Sharp Guarantees
123. ICLR 2025State Space Models are Provably Comparable to Transformers in Dynamic Token Selection
124. ICLR 2025Statistical Advantages of Perturbing Cosine Router in Mixture of Experts
125. ICLR 2025Statistical Tractability of Off-policy Evaluation of History-dependent Policies in POMDPs
126. ICLR 2025Stochastic Bandits Robust to Adversarial Attacks
127. ICLR 2025Strong Model Collapse
128. ICLR 2025Task Descriptors Help Transformers Learn Linear Models In-Context
129. ICLR 2025The Breakdown of Gaussian Universality in Classification of High-dimensional Linear Factor Mixtures
130. ICLR 2025The Computational Complexity of Positive Non-Clashing Teaching in Graphs
131. ICLR 2025The Foundations of Tokenization: Statistical and Computational Concerns
132. ICLR 2025Theory on Mixture-of-Experts in Continual Learning
133. ICLR 2025Theory on Score-Mismatched Diffusion Models and Zero-Shot Conditional Samplers
134. ICLR 2025Towards Auto-Regressive Next-Token Prediction: In-context Learning Emerges from Generalization
135. ICLR 2025Towards Generalization Bounds of GCNs for Adversarially Robust Node Classification
136. ICLR 2025Towards Understanding the Universality of Transformers for Next-Token Prediction
137. ICLR 2025Tractable Multi-Agent Reinforcement Learning through Behavioral Economics
138. ICLR 2025Trained Transformer Classifiers Generalize and Exhibit Benign Overfitting In-Context
139. ICLR 2025Training Nonlinear Transformers for Chain-of-Thought Inference: A Theoretical Generalization Analysis
140. ICLR 2025Transformer Learns Optimal Variable Selection in Group-Sparse Classification
141. ICLR 2025Transformers are Universal In-context Learners
142. ICLR 2025Transformers Handle Endogeneity in In-Context Linear Regression
143. ICLR 2025Transformers Learn to Implement Multi-step Gradient Descent with Chain of Thought
144. ICLR 2025Transformers Provably Learn Two-Mixture of Linear Classification via Gradient Flow
145. ICLR 2025Transformers Provably Solve Parity Efficiently with Chain of Thought
146. ICLR 2025U-Nets as Belief Propagation: Efficient Classification, Denoising, and Diffusion in Generative Hierarchical Models
147. ICLR 2025Understanding Factual Recall in Transformers via Associative Memories
148. ICLR 2025uniINF: Best-of-Both-Worlds Algorithm for Parameter-Free Heavy-Tailed MABs
149. ICLR 2025When does compositional structure yield compositional generalization? A kernel theory.
150. ICLR 2025When is Task Vector Provably Effective for Model Editing? A Generalization Analysis of Nonlinear Transformers

FAQ

How many Theory · learning theory papers were published at NeurIPS, ICLR, ICML and CVPR from 2023 to 2026?

1,350 in total: 86 in 2023, 393 in 2024, 389 in 2025, 482 in 2026.

How many Theory · learning theory papers are at NeurIPS 2026?

NeurIPS 2026 has 112 Theory · learning theory papers, versus 106 the year before.

How many Theory · learning theory papers are at ICLR 2026?

ICLR 2026 has 200 Theory · learning theory papers, versus 147 the year before.

How many Theory · learning theory papers are at ICML 2026?

ICML 2026 has 170 Theory · learning theory papers, versus 132 the year before.

How many Theory · learning theory papers are at CVPR 2025?

CVPR 2025 has 4 Theory · learning theory papers, versus 1 the year before.

Which conference publishes the most Theory · learning theory papers?

NeurIPS, with 540 papers from 2023 to 2026.