Brief

Researchers have developed a new framework for Online Convex Optimization (OCO) that can improve worst-case regret even with a limited and noisy budget of pairwise probes. The proposed method unifies sublinear best-expert queries and pairwise feedback, showing that a sublinear, noisy probe budget can provably enhance regret in the full feedback OCO regime. The analysis quantifies the benefit of probing through variance reduction and a second-order analysis of Continuous Exponential Weights, yielding tight regret guarantees. AI

Researchers have developed a novel principle for online inventory optimization (OIO) that achieves optimal performance on general convex sets. This method, which involves maintaining a hidden target and projecting it onto the feasible order-up-to set, improves regret guarantees for OIO and offers the first polylogarithmic regret for strongly convex losses. The analysis introduces a 'norm alignment' principle, reducing the problem to one-dimensional queue control, which has been validated through experiments on both synthetic and real-world inventory data. AI

Two new research papers explore advanced methods for improving AI agent decision-making and learning. The first paper, "Trivium," introduces temporal regret as a key objective for causal-memory controllers, aiming to log and correct errors more effectively than outcome-based methods. The second paper, "Parameter-free Dynamic Regret," presents a novel algorithm for online convex optimization that handles time-varying movement costs, delayed feedback, and memory, achieving improved dynamic regret bounds. AI