performance guarantees
Formal assurances about the expected performance of an AI algorithm under certain conditions or assumptions. They are important for establishing trust and reliability in deploying AI systems in critical applications.
- $O(\sqrt{T})$ Static Regret and Instance Dependent Constraint Violation for Constrained Online Convex Optimization
- A General-Purpose Theorem for High-Probability Bounds of Stochastic Approximation with Polyak Averaging
- Active Seriation: Efficient Ordering Recovery with Statistical Guarantees
- Asymptotic theory of SGD with a general learning-rate
- Data-Dependent Regret Bounds for Constrained MABs
- Eluder dimension: localise it!
- Greedy Sampling Is Provably Efficient For RLHF
- Improved Best-of-Both-Worlds Regret for Bandits with Delayed Feedback
- Kernel-based Equalized Odds: A Quantification of Accuracy-Fairness Trade-off in Fair Representation Learning
- Learning-Augmented Facility Location Mechanisms for the Envy Ratio Objective
- Non-Clairvoyant Scheduling with Progress Bars
- Preference-based Reinforcement Learning beyond Pairwise Comparisons: Benefits of Multiple Options
- Provable Meta-Learning with Low-Rank Adaptations
- Revisiting Agnostic Boosting
- Risk-Averse Total-Reward Reinforcement Learning
- Robust Satisficing Gaussian Process Bandits Under Adversarial Attacks
- Split conformal classification with unsupervised calibration
- Statistical Parity with Exponential Weights
- Tight analyses of first-order methods with error feedback