Confidence Sequence

Confidence sequences (CSs) provide anytime-valid confidence intervals, crucial for sequential decision-making and online learning where data arrives incrementally. Current research focuses on extending CSs to more complex models, including generalized linear models and kernelized settings, and improving their efficiency and robustness, particularly in the presence of heavy-tailed data or outliers. These advancements are impacting diverse fields, from bandit algorithms and financial auditing to natural language processing, by enabling reliable uncertainty quantification in adaptive and online systems. The development of tighter, more efficient CSs remains a key area of ongoing investigation.