AI, Robot
Google DeepMind: The Podcast
00:00
Training Neural Networks to Provide Feedback and Assess Performance
The speaker describes their approach to training neural networks to provide feedback to an agent attempting a task. They acknowledge the challenge of having a human provide feedback as it is slow and requires significant resources. Instead, they propose training a second neural network to provide feedback, allowing the primary network to learn from it. This approach leverages the ability of neural networks to spot patterns and assess performance, potentially eliminating the need for constant human intervention in providing feedback.
Transcript
Play full episode
