Neural Network Training
Ishan emphasizes the importance of data and data augmentation techniques over specific neural network architectures, noting that many architectures yield similar results depending on the task. He discusses the challenges of training large models, highlighting the need for efficient distributed computing strategies to minimize communication costs during training. The conversation reveals that scaling up models effectively involves careful consideration of both hardware and system design.In this clip
From this podcast
Lex Fridman Podcast
Ishan Misra: Self-Supervised Deep Learning in Computer Vision | Lex Fridman Podcast #206
Related Questions