Formal Guarantee
Formal guarantees in machine learning and related fields focus on developing methods and algorithms that provide mathematically provable assurances about the performance, safety, or reliability of systems. Current research emphasizes developing algorithms with such guarantees for various applications, including reinforcement learning (e.g., using Lyapunov functions or compositional methods), causal inference (leveraging interventional data and faithfulness assumptions), and robust optimization (addressing distributional uncertainty and worst-case scenarios). This work is significant because it moves beyond empirical evaluations, providing stronger confidence in the behavior of complex systems and enabling their deployment in high-stakes applications like robotics, autonomous systems, and medical diagnosis.
Papers
Fine-tuning Language Models over Slow Networks using Activation Compression with Guarantees
Jue Wang, Binhang Yuan, Luka Rimanic, Yongjun He, Tri Dao, Beidi Chen, Christopher Re, Ce Zhang
Sparse Mixed Linear Regression with Guarantees: Taming an Intractable Problem with Invex Relaxation
Adarsh Barik, Jean Honorio