Model Self-Criticism
Moustapha emphasizes the importance of regularization in models to prevent them from learning harmful biases from poor-quality data. He draws an analogy between how children learn to discern good from bad food and how models should be trained to select valuable data points. The conversation highlights the need for models to develop self-criticism and introspection to improve their learning processes.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Security and Safety in AI: Adversarial Examples, Bias and Trust with Moustapha Cissé - #108
Related Questions