Machine Learning

Adaptive labeling minimizes uncertainty in out-of-distribution model evaluation by strategically selecting which data points to label, leading to more efficient and reliable assessments.

GVFExplorer: An adaptive behavior policy efficiently learns multiple GVFs by minimizing return variance, optimizing data usage and reducing prediction errors.

Adaptive Depth Networks with Skippable Sub-Paths: Train once, deploy efficiently! This paper proposes a novel training method to create adaptive-depth networks, enabling on-demand model depth selectio…

Q-Aided Conditional Supervised Learning (QCS) effectively combines the stability of return-conditioned supervised learning with the stitching ability of Q-functions, achieving superior offline reinfor…

Score-based diffusion models are improved by a novel coefficient design, enabling efficient adaptation to unknown low-dimensional data structures and achieving a convergence rate of O(k²/√T).

Adam-Rel: A novel optimizer for RL, dramatically improves performance by resetting Adam’s timestep to 0 after target network updates, preventing large, suboptimal changes.

Ada-MSHyper: A novel adaptive multi-scale hypergraph transformer significantly boosts time series forecasting accuracy by modeling group-wise interactions and handling complex temporal variations.

ActSort: Active learning dramatically accelerates cell sorting in massive calcium imaging datasets, minimizing human effort and improving accuracy.

This paper introduces anytime-valid risk-controlling prediction sets for active learning, guaranteeing low risk even with adaptive data collection and limited label budgets.

Active Set Ordering: Efficiently discover input subsets (maximizers, top-k) of expensive black-box functions via pairwise comparisons, using a novel Mean Prediction algorithm with theoretical guarante…

Active learning efficiently orders items using contextual attributes, minimizing comparison needs and improving generalization.

Active learning for general halfspaces is surprisingly hard; membership queries are key to efficiency.

Boost global sensitivity analysis efficiency by 10x with novel active learning methods targeting derivative-based measures for expensive black-box functions!

Researchers developed an active learning method using two-photon holographic optogenetics to efficiently identify neural population dynamics, achieving up to a two-fold reduction in data needed for ac…

Slash deep learning’s memory footprint! This paper introduces a novel activation map compression technique via tensor decomposition, significantly boosting on-device training efficiency for edge AI.

Distributional RL’s sensitivity to high-frequency decisions is unveiled, with new algorithms solving existing performance issues in continuous-time RL.

First tractable algorithm achieves minimax optimal regret in average-reward MDPs, solving a major computational challenge in reinforcement learning.

D-AdaST: A novel distributed adaptive minimax optimization method achieves near-optimal convergence by tracking stepsizes, solving the inconsistency problem hindering existing adaptive methods.

A novel parameter-free decentralized optimization algorithm achieves linear convergence for strongly convex, smooth objectives, eliminating the need for hyperparameter tuning and improving scalability…

Cert-LSVI-UCB achieves constant regret in RL with linear function approximation, even under model misspecification, using a novel certified estimator.