
Episode 33: Tri Dao, Stanford: On FlashAttention and sparsity, quantization, and efficient inference
Generally Intelligent
The Kernel Theory of Data Augmentation
Data augmentation is one way you can encode knowledge into the model. In computer vision, it might be that if you blur an image or rotate them a little bit they shouldn't change their identity. The simpler way to encode this kind of knowledge into themodel is just directly modifying the data. And people have found that to be a little more intuitive and so far, that approach has been a little bit more popular.
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.