Machine Learning

Efficient Adaptive Data Acquisition via Pretrained Belief Representations

POLAR uses pretrained belief representations to learn efficient adaptive data acquisition policies.

Jun 23, 2026

Loss-Driven Bayesian Active Learning

A loss-driven framework for Bayesian active learning that targets downstream decision losses directly.

May 8, 2026

SigDiffusions: Score-Based Diffusion Models for Time Series via Log-Signature Embeddings

A novel approach combining score-based diffusion models with log-signature embeddings for advanced time series modeling and generation.

Jan 22, 2025