Taming Dataset Bias via Domain Adaptation

Explore dataset bias and domain adaptation techniques in this 43-minute lecture from MIT's Introduction to Deep Learning course. Delve into the occurrence and real-world implications of dataset bias, and learn strategies to mitigate its effects. Discover adversarial domain alignment, pixel space alignment, and few-shot pixel alignment methods. Examine approaches that move beyond alignment and enforce consistency in machine learning models. Gain valuable insights from Prof. Kate Saenko of the MIT-IBM Watson AI Lab on taming dataset bias to improve the robustness and fairness of deep learning systems.

​​ - Introduction
- When does dataset bias occur?
- Implications in the real-world
- Dealing with data bias
- Adversarial domain alignment
- Pixel space alignment
- Few-shot pixel alignment
- Moving beyond alignment
- Enforcing consistency
- Summary and conclusion