Know What You Don't Know: Tools to Understand Uncertainty in DL
summary
Bridges practical deep learning with Bayesian inference by introducing Bayesian Neural Networks and the fundamentals of Bayesian inference. Explains how dropout and other stochastic regularization methods can be interpreted as inexpensive ways to estimate model uncertainty, shows real-world applications (healthcare, trading, autonomous vehicles), and includes code snippets demonstrating uncertainty-aware techniques for exploration-exploitation in reinforcement learning and active sampling for annotation efficiency.
AI-written summary · about
chapters (12)
- 0:00Uncertainty in deep learning
- 6:44Models can't say I don't know
- 12:22Aleatory vs epistemic uncertainty
- 19:58Bayesian models and neural networks
- 24:15Dropout as Bayesian approximation
- 29:56Measuring model uncertainty
- 31:40Uncertainty in reinforcement learning
- 33:48Active data selection
- 36:43Summary and tools
- 38:07Q&A on rejection classification
- 41:59Aleatory vs epistemic uncertainty
- 44:16Financial data modeling challenges
related talks
05mThe Frontiers of Machine Learning and AI
35mUncertainty in deep learning
25mDiscovering Your Model's Known Unknowns and Unknown Unknowns
39mBayesian Deep Learning with Edward (and a trick using Dropout)
44mAre You Sure about That?! Uncertainty Quantification in AI
07mProbabilistic Reasoning and Learning for Trustworthy AI
32mStructured Deep Learning with Probabilistic Neural Prog
34mBayesian Optimization and it's application to Neural Networks