Why Deep Learning is driving demand for Human in the Loop
summary
Explains why Imagenet 2012 and Hinton’s jump from a few hidden layers to eight catalyze deep learning’s hunger for labeled data, driving demand for human in the loop annotation teams; shows models with billions to trillions of parameters that require continual retraining because supervised models degrade unless relabeled data are fed back; describes quality controls like double annotation and exception processing (OCR handles ~95% with humans for the rest) and cites annotator throughput examples of 45 per hour, plus ag‑tech and accessibility apps that already work.
AI-written summary · about
chapters (10)
- 0:00Deep learning and human in loop
- 3:25Supervised learning myths
- 6:102012 changed deep learning
- 10:05Massive labeled datasets needed
- 13:13Data annotation beyond training
- 16:48Human in loop missing in education
- 19:16Expert annotators' value
- 23:12Human in loop use cases
- 25:40Deep learning in everyday life
- 27:45Autonomous vehicles future
related talks
51mEfficient Deep Learning with Humans in the Loop
30mActive Learning and Human-in-the-Loop Machine Learning
45mIntroduction To Deep Learning
28mChallenges & future trends of Deep Learning
37mBest Practices for Building AI Datasets
08mAgents and Robots: The Human Effort Putting AI Into Production
54mIntroduction to Understanding Deep Learning
32mAugmented Intelligence - A Marriage between Machine and Human