Robustness to Adversarial Inputs and Tail Risk via Boosting - Machine Learning Model Defense Strategies

Explore a 47-minute conference talk by Dr. Pradeep Ravikumar on enhancing machine learning model robustness against adversarial inputs and tail risk. Delve into innovative strategies for deploying ML models in high-stakes environments, focusing on an ensemble approach using neural networks to defend against threats while maintaining performance on challenging samples. Gain insights into critical areas such as algorithmic fairness, class imbalance, and risk-sensitive decision-making. Cover topics including distribution shift, unknown sub-populations, CVAR&DRO, randomized approaches, solving zero-sum games, and Distributional Outlier Robust Optimization (DORO). Ideal for professionals and enthusiasts in machine learning, AI, data science, and related fields seeking to advance their understanding of robust AI techniques and adversarial defense strategies.

- Intro
- Distribution Shift
- Unknown Sub-populations, Tail Risk
- CVAR&DRO
- Deterministic to Randomized
- Solving Zero Sum Games
- DORO: DIstributional Outlier Robust Optimization
- Summary