Capsa: A Unified Framework for Quantifying Risk in Deep Neural Networks

Dec 2, 2022



The deployment of large-scale deep neural networks (NNs) in safety-critical scenarios requires quantifiably calibrated and reliable measures of trust. Unfortunately, existing algorithms to achieve risk-awareness of NNs are complex and ad-hoc. We present Capsa, an open-source and flexible framework for unifying these methods and instilling models with risk-aware capabilities. We unify state-of-the-art risk algorithms under the Capsa framework, propose a composability method for combining different risk estimators together in a single function set, and benchmark on high-dimensional perception tasks.


Store presentation

Should this presentation be stored for 1000 years?

How do we store presentations

Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%


Recommended Videos

Presentations on similar topic, category or speaker

Interested in talks like this? Follow NeurIPS 2022