Eyke Hüllermeier

Unknown GP hyperparameters can drastically distort tail-risk estimates in CVaR optimization, but a new algorithm robustly navigates this uncertainty to achieve near-oracle performance.

Uncertainty in loss distributions cripples standard risk-averse ML; this work shows credal sets can reliably prevent catastrophic decisions without sacrificing average performance.

Transferring latent group priors across cohorts eliminates the standard bandit cold-start penalty while actively preventing the premature catalog exhaustion that plagues real-world deployments.

Human participation isn't just a stopgap for AI's shortcomings; it's a vital component in shaping outcomes that emerge through collaboration.