Is It Impossible for Learning Complex Representations?

Neural networks often work like locality sensitive hashing in reinforcement learning because it's the easy thing to learn. However, the dominance of gradient descent as the current paradigm may not be ideal for complex representations. Making larger stochastic changes and evaluating them over time can allow for learning beyond what gradient descent can achieve.

Transcript

Play full episode

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.