#206
Ishan Misra - Self-Supervised Deep Learning in Computer Vision
Sat, Jul 31, 2021 ·
Loading views
Ishan Misra is a research scientist at FAIR working on self-supervised visual learning.
Timestamps
- 2:27 - Self-supervised learning
- 11:02 - Self-supervised learning is the dark matter of intelligence
- 14:54 - Categorization
- 23:28 - Is computer vision still really hard?
- 27:12 - Understanding Language
- 36:51 - Harder to solve: vision or language
- 43:36 - Contrastive learning & energy-based models
- 47:37 - Data augmentation
- 51:57 - Fixed audio spike by lowering sound with pen tool
- 1:00:10 - Real data vs. augmented data
- 1:03:54 - Non-contrastive learning energy based self supervised learning methods
- 1:07:32 - Unsupervised learning (SwAV)
- 1:10:14 - Self-supervised Pretraining (SEER)
- 1:15:21 - Self-supervised learning (SSL) architectures
- 1:21:21 - VISSL pytorch-based SSL library
- 1:24:15 - Multi-modal
- 1:31:43 - Active learning
- 1:37:22 - Autonomous driving
- 1:48:49 - Limits of deep learning
- 1:52:57 - Difference between learning and reasoning
- 1:58:03 - Building super-human AI
- 2:05:51 - Most beautiful idea in self-supervised learning
- 2:09:40 - Simulation for training AI
- 2:13:04 - Video games replacing reality
- 2:14:18 - How to write a good research paper
- 2:18:45 - Best programming language for beginners
- 2:19:39 - PyTorch vs TensorFlow
- 2:23:03 - Advice for getting into machine learning
- 2:25:09 - Advice for young people
- 2:27:35 - Meaning of life
Listen on Spotify
Hide video