Assistant Professor, Machine Learning Department
Carnegie Mellon University
4:00 PM Panel discussion: Understanding, managing and effectively mitigating the hidden risks associated with AI
- The risks of the person behind the AI/model: how to remove bias?
- Coping with tail risks
- Can you ever model all circumstances in an undefinable and constantly moving market?
- How can AI measure a risk it hasn’t seen yet?
- Balancing the compliance and risk demands of creating human understanding of complex models
- Overfitting: Have you applied diversification in your quant strategies? A variety of asset classes? Appropriate algorithm choices?
- Would a failure of model or liquidity mean you have inappropriate risk management in place?
- Underspecified and unclear: have you appropriately defined your model and the properties it satisfies?
- Understanding causal structure: validating the decisions your machine made
- Resilience and robustness: The importance of factoring in domain adaptation and distribution shift
- Anomaly detection: quantifying and actively compensating by modifying your beliefs and constraints