Chapter
The Struggles of Working with Neural Nets
25:28 - 31:10 (05:41)
The speaker discusses the difficulties and mistakes made when experimenting with neural nets, such as improperly initializing weights and making the network too small.
Clips
The speaker emphasizes on the investment required to create software platforms to experiment with neural nets and create network architectures.
25:28 - 29:57 (04:28)
Neural Networks
Summary
The speaker emphasizes on the investment required to create software platforms to experiment with neural nets and create network architectures. Making basic mistakes like bad weight initialization and small network size are part of the learning process.
ChapterThe Struggles of Working with Neural Nets
EpisodeYann LeCun: Deep Learning, Convolutional Neural Networks, and Self-Supervised Learning
PodcastLex Fridman Podcast
The US patent office used to allow the patent of software as long as it was embodied, however, both Facebook and Google no longer believe in this kind of patent and fire patents for defensive purposes only.
29:57 - 31:10 (01:13)
Patenting, Software Ideas, Defensive Purposes
Summary
The US patent office used to allow the patent of software as long as it was embodied, however, both Facebook and Google no longer believe in this kind of patent and fire patents for defensive purposes only.