The mathematics under the hood: optimization methods, convergence guarantees, and the theory that explains when and why deep learning works. Explainers that take proofs seriously without losing the reader.
Statistical Inference via Sketched StoSQP: Online Second-Order Methods for Constrained Optimization | AI Trend Blend Optimization Theory · Journal of Machine Learning Research 26 (2025) 1–75 · 20 min read The Online Inference Problem That Second-Order Methods Finally Solved — Without Projections Sen Na at Georgia Tech and Michael Mahoney at UC Berkeley prove that […]
The ODE Method for Stochastic Approximation with Markovian Noise: Breaking the Deadly Triad in Reinforcement Learning | AI Trend Blend AITrendBlend Machine Learning Computer Vision About Reinforcement Learning Theory · Journal of Machine Learning Research 26 (2025) 1–76 · 20 min read The ODE Method Gets Its Markovian Upgrade — and Reinforcement Learning’s Most Stubborn
Orthogonal Bases for Equivariant Graph Learning with Provable k-WL Expressive Power | AI Trend Blend Graph Neural Networks · Journal of Machine Learning Research 26 (2025) 1–35 · 18 min read High-Order GNNs Were Too Expensive — Until These Compact Orthogonal Bases Changed the Game Jia He and Maggie X. Cheng from Illinois Institute of
Orthogonal Bases for Equivariant Graph Learning with Provable k-WL Expressive Power Read More »
Riemannian Bilevel Optimization — When Machine Learning Leaves Flat Space Behind | AI Trend Blend AITrendBlend Machine Learning Mathematics About Machine Learning Theory · Journal of Machine Learning Research 26 (2025) · University of Minnesota & Rice University · 20 min read Why Machine Learning on Curved Surfaces Is the Next Big Leap — And
Riemannian Bilevel Optimization — When Machine Learning Leaves Flat Space Behind Read More »
Random ReLU Neural Networks as Non-Gaussian Processes | AI Trend Blend AITrendBlend Machine Learning Computer Vision About Neural Network Theory · Journal of Machine Learning Research 26 (2025) 1–31 · 16 min read Wide Neural Networks Are Not Always Gaussian — Here’s the Proof A team from UC San Diego and EPFL’s Biomedical Imaging Group
Random ReLU Neural Networks as Non-Gaussian Processes Read More »
From Sparse to Dense Functional Data in High Dimensions: Phase Transitions Revisited | AI Trend Blend Functional Data Analysis · Journal of Machine Learning Research 26 (2025) 1–40 · 18 min read When Does Sampling Density Actually Matter? Phase Transitions in High-Dimensional Functional Data, Revisited A team from Renmin University of China, Tsinghua University, and
From Sparse to Dense Functional Data in High Dimensions: Phase Transitions Revisited Read More »
Why Hard Training Examples Hurt Neural Networks — And How DPLS Fixes It | AI Trend Blend Adversarial Robustness · Journal of Machine Learning Research 26 (2025) 1–48 · 16 min read Why Hard Training Examples Are Secretly Sabotaging Your Neural Network’s Robustness A team from Seoul National University and Ewha Womans University pinpointed a
Why Hard Training Examples Hurt Neural Networks — And How DPLS Fixes It Read More »
Teaching Machines That the World Keeps Changing: Supervised Learning with Evolving Tasks and Performance Guarantees | AI Trend Blend AITrendBlend Machine Learning Math About Continual Learning · Journal of Machine Learning Research 26 (2025) 1–59 · BCAM · University of the Basque Country · 22 min read Teaching Machines That the World Keeps Changing: One
An Axiomatic Definition of Hierarchical Clustering: Why Hartigan Was Right All Along | AI Trend Blend AITrendBlend Machine Learning Math Computer Vision About Statistical Learning Theory · Journal of Machine Learning Research 26 (2025) 1–26 · 18 min read Three Rules That Define What a Cluster Actually Is: The Axiomatic Case for Hartigan’s Cluster Tree
An Axiomatic Definition of Hierarchical Clustering: Why Hartigan Was Right All Along Read More »
Bayes Meets Bernstein: Why Meta-Learning Finally Gets Fast — A PAC-Bayes Breakthrough | AI Trend Blend AITrendBlend Machine Learning Math About Machine Learning Theory · Journal of Machine Learning Research 26 (2025) · University of Tokyo, ESSEC Business School & CNRS / Sorbonne Université · 16 min read Why Meta-Learning Suddenly Gets Smarter the More
Bayes Meets Bernstein: Why Meta-Learning Finally Gets Fast — A PAC-Bayes Breakthrough Read More »