Machine Learning

New method quantifies & reduces bias in constant-step stochastic approximation algorithms with Markovian noise, improving accuracy and efficiency.

Computation-Aware Gaussian Processes (CaGP) achieve linear-time inference and model selection, enabling efficient training of GPs on large datasets without compromising uncertainty quantification, sur…

Goal-conditioned RL gets a temporal upgrade with compositional DFAs (cDFAs), enabling zero-shot generalization and faster policy specialization via novel graph neural network embeddings and reach-avoi…

Communication-efficient algorithms for federated group distributionally robust optimization (FGDRO) are introduced, achieving lower communication complexity and superior performance on real-world task…

Distributed Lion: Training large AI models efficiently by communicating only binary or low-precision vectors between workers and a server, significantly reducing communication costs and maintaining co…

Boosting neural network prediction reliability, this research ingeniously combines statistical depth and Fermat distance for superior uncertainty quantification, eliminating the need for distributiona…

Collaborative Refining for Learning from Inaccurate Labels (CRL) refines data using annotator agreement, improving model accuracy with noisy labels.

CODA: A novel modeling scheme tackles subpopulation shifts in machine learning by disentangling spurious correlations, augmenting data strategically, and using reweighted consistency loss for improved…

CoBo: A novel bilevel optimization algorithm for collaborative learning surpasses existing methods by efficiently selecting helpful clients, resulting in superior performance and scalability.

This paper introduces novel non-adaptive algorithms for clustering using subset queries, achieving near-linear query complexity and improving upon existing limitations of pairwise query methods.

RecPiece: Relational Clustering for Better Knowledge Graph Embedding Anchors

Boosting in-distribution generalization is achieved by strategically altering the training data distribution to reduce simplicity bias and promote uniform feature learning.

CE-NAS: A novel framework minimizes the carbon footprint of Neural Architecture Search by dynamically allocating GPU resources based on predicted carbon intensity, achieving state-of-the-art results w…

This paper presents novel causal imitation learning algorithms using partial identification to achieve expert performance even when unobserved confounders affect Markov Decision Processes.

Causal CPC: a novel method for accurate and efficient counterfactual regression over time using RNNs, CPC, and InfoMax, achieving state-of-the-art performance.

Statistical Flow Matching (SFM) uses information geometry to create a new flow-matching framework for generating discrete data, achieving superior sampling quality and likelihood compared to existing …

RLHF’s KL regularization fails to prevent ‘catastrophic Goodhart’—policies achieving high proxy reward but low actual utility—when reward errors have heavy tails.

Energy-based models’ training reveals a cascade of phase transitions, progressively learning data features, offering new insights into deep learning dynamics.

Truthful comparison data is hard to obtain without ground truth. This paper presents novel peer prediction mechanisms using bonus-penalty payments that incentivize truthful comparisons, even in networ…

This paper proposes cardinality-aware top-k classification, improving accuracy and efficiency by dynamically adjusting prediction set sizes.