Statistical Learning
Preventing Dimensional Collapse in Self-Supervised Learning via Orthogonality Regularization
SSL endeavors to produce meaningful representations without the guidance of labels.
Diffusion-Reward Adversarial Imitation Learning
Despite its encouraging results, GAIL training is often brittle and unstable.
Effective Exploration Based on the Structural Information Principles
The SI2E analyzes value differences in the agent's policy between state-action
Online Composite Optimization Between Stochastic and Adversarial Environments
We study online composite optimization under the Stochastically Extended Adversarial (SEA) model.
ac4106bcfff33140de7799d03daeb8a4-Paper-Conference.pdf
Cross-V alidation (CV) is the default choice for estimate the out-of-sample performance of machine learning models.
Accelerating Augmentation Invariance Pretraining Jinhong Lin Cheng-En Wu
In ImageNet, our method achieves speedups of 4 in MoCo, 3.3 in SimCLR, and 2.5 in DINO, demonstrating substantial efficiency gains.
Test-time Adaptation in Non-stationary Environments via Adaptive Representation Alignment
Continual test-time adaptation methods have shown promising results by using reliable pseudo-labels, but they still fall short in exploring representation alignment with the source domain in non-stationary environments.