Trustworthy Deep Learning

Explore the critical aspects of robust and trustworthy deep learning in this comprehensive lecture from MIT's Introduction to Deep Learning course. Delve into the challenges of algorithmic bias, including class imbalance and latent feature imbalance, and learn about innovative solutions like the debiasing variational autoencoder (DB-VAE). Examine the importance of uncertainty in deep learning, distinguishing between aleatoric and epistemic uncertainty, and discover methods for estimating both types. Gain insights into evidential deep learning and understand how companies like Themis AI are revolutionizing risk-awareness in AI systems. Conclude with an introduction to Capsa, an open-source risk-aware AI wrapper, and explore the future of trustworthy AI development.

- Introduction and Themis AI
- Background
- Challenges for Robust Deep Learning
- What is Algorithmic Bias?
- Class imbalance
- Latent feature imbalance
- Debiasing variational autoencoder DB-VAE
- DB-VAE mathematics
- Uncertainty in deep learning
- Types of uncertainty in AI
- Aleatoric vs epistemic uncertainty
- Estimating aleatoric uncertainty
- Estimating epistemic uncertainty
- Evidential deep learning
- Recap of challenges
- How Themis AI is transforming risk-awareness of AI
- Capsa: Open-source risk-aware AI wrapper
- Unlocking the future of trustworthy AI