061: Interpolation, Extrapolation and Linearisation (Prof. Yann LeCun, Dr. Randall Balestriero)
Machine Learning Street Talk (MLST)
Hybrid learning systems with continuous and discrete elements
The apparent dycodomy between continuous and discreet systems is like the human brain. It has evolved both analogue and discreet computations. It's difficult to train fixed depth neural networks for discreet computations. The hybrid system is efficient since it clusters different settings and applies discreet and continuous reasoning accordingly. Both systems need to interact for optimal results. Humans lack a discreet system for playing chess, but symbolic and statically coded discreet systems have limitations. Deep learning models can guide discreet programm search, but it's not feasible within the continuous domain.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
