Recent Solutions Using Deep Learning Models/ The How and Why
summary
Shows concrete wins and limits: convolutional nets speed virtual drug screens, raising initial hit rates from 0.02% → 30% and helping find inhibitors for resistant bacteria; drug development still costs about $1.8B and 13 years. Kepler light-curve CNNs score ~98.8% on human-curated data but fall to 81% on simulated exoplanet transits and 33% on binary-star cases, revealing data-bias failure modes. A DeepBeat MLP trained on 600,000 lines from 104 rappers hits 17% next-line prediction versus 0.03% random.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:35HISTORY OF MACHINE LEARNING
- 3:02NEURAL NETWORKS
- 3:59DEEP LEARNING
- 17:40TRANSIT PHOTOMETRY
- 18:15KEPLER 90
- 30:20DEEP BEAT
related talks
54mIntroduction to Understanding Deep Learning
41mDeep Learning to Solve Challenging Problems
40mMachines are Learning
43mWhy you can't afford to ignore deep learning
25mBridging Gaps: Classic Approaches Beyond Deep Learning Constraints
59mDeep Neural Networks for Video Applications
58mDeep Learning in Real Life
01h 01mA practical guide to deep learning