Evidential Deep Learning and Uncertainty

Explore evidential deep learning and uncertainty estimation in this comprehensive lecture from MIT's Introduction to Deep Learning course. Delve into probabilistic learning, distinguishing between discrete and continuous target learning, and understanding the difference between likelihood and confidence. Examine various types of uncertainty, including aleatoric and epistemic, and learn about Bayesian neural networks. Discover advanced techniques beyond sampling for uncertainty estimation, with a focus on evidential deep learning for both regression and classification tasks. Gain insights into evidential model training and its practical applications. Compare different approaches to uncertainty estimation and grasp the importance of these concepts in the field of deep learning.

Syllabus

- Introduction and motivation
- Outline for lecture
- Probabilistic learning
- Discrete vs continuous target learning
- Likelihood vs confidence
- Types of uncertainty
- Aleatoric vs epistemic uncertainty
- Bayesian neural networks
- Beyond sampling for uncertainty
- Evidential deep learning
- Evidential learning for regression and classification
- Evidential model and training
- Applications of evidential learning
- Comparison of uncertainty estimation approaches
- Conclusion