Deep Learning to Solve Challenging Problems
summary
Overview of concrete claims, tools, and numbers: Google Brain shows vision models drop ImageNet error from ~26% (2011) to ~3% (2016); robot grasping rises from ~65% (2015) to 96% grasp success using pooled robot data; retinal models match retinal specialists and predict age, sex, blood pressure; quantum chemistry surrogate nets run ~300,000× speedup vs simulator; AutoML finds architectures outperforming hand designs; training hardware scale: TPU v3 pod: train ImageNet in <2 minutes.
AI-written summary · about
chapters (12)
- 0:00Deep learning overview
- 4:42Grand engineering challenges
- 5:57Autonomous vehicles
- 7:48Robotics and grasping
- 10:54Advanced healthcare
- 16:57Scientific discovery
- 18:38TensorFlow ecosystem
- 21:25Sparse activation models
- 24:37Automated machine learning
- 30:15Specialized hardware
- 36:19Future AI directions
- 38:26Responsible AI principles
related talks
40mMachines are Learning
25mDeep Learning in Biomedicine: Challenges and Opportunities
56mTutorial on Deep Learning II
42mState of the Art of Innovation in Computer Vision
01h 02mTutorial on Deep Learning I
01h 00mDeep Learning Revolution: Rethinking Machine Learning Pipelines
42mDeep Learning for Computer Vision
58mDeep Learning in Real Life