Overcoming the Barriers to Deep Learning in Production
summary
Overview of concrete tactics and failures: commercial engagements build vision-guided smart agricultural machinery that increases yield without new land; early wins use modest projects like speech recognition and photo OCR (Google Speech, Maps) to build momentum; tooling stack centers on Python, GPUs, TensorFlow/PyTorch and the need for better data versioning; workflows adopt a one-day sprint cadence for overnight experiments; risk is managed with FMEA scoring — probability, severity, detectability — and attention to small-data and distribution-shift problems in production.
AI-written summary · about
chapters (9)
- 0:00Introduction and welcome
- 1:04Andrew Ng’s background and projects
- 3:31Landing AI’s mission and work
- 8:04Choosing the right AI projects
- 13:18Challenges in enterprise AI adoption
- 16:29Scaling AI in production
- 21:18Iterating quickly in ML workflows
- 27:18Managing AI deployment risks
- 32:12Future of AI beyond software
related talks
16mDeep Learning in Production Seen by a Data Engineer
41mBeyond the Algorithm: What Makes Machine Learning Work?
33mRunning Machine Learning in Production - a Journey to Success
05mDeep Learning in the Wild
39mKeynote: Solving the problems of AI in production
37mPanel Discussion: Developing with AI
28mLeading machine learning research journeys with happy (into-production) endings
42mDeep learning beyond the learning