Beyond the Algorithm: What Makes Machine Learning Work?
summary
Anecdotes and case studies show the difference between a lab model and production value: a Raspberry Pi running a TensorFlow+Inception v3 classifier on thousands of labeled images detects chickens but takes 30 seconds latency — too slow to stop a jay — highlighting SLA-driven design. Preserve raw data, run decoy and canary models, use streaming/Kafka and a rendezvous architecture for model selection, and automate auditing of mislabeled entries to recover tens of millions saved; heavy emphasis on data engineering and deployment logistics, not just algorithms.
AI-written summary · about
chapters (19)
- 0:00Intro
- 1:14Contact Information
- 1:49Data Engineer lan Downard
- 8:13Value from ML: what about action? When image classification indicates a jay in the henhouse, what would you do to chase away the predator?
- 9:36What makes a difference for impact?
- 16:36Spot the Important Difference?
- 17:03Another Example
- 18:56Big Industry, Big Data, Big Value
- 24:56Simple But Valuable: Accounting Audit Targeting Big industrial company with a lot of machinery • Tracking actions & parts - label as repairs, contracted services, delivery of supplies, etc. • Which are to be taxed, expensed or counted as revenues - Mislabelling can cost millions of dollars • Use machine learning to target potential mislabelling for audit review
- 29:32Is it the algorithm? the model? the ML tool?
- 31:11What Does Streaming Do for You?
- 34:08Enter Rendezvous...
- 34:40Rendezvous server making continuous decisions
- 34:51Machine Learning Logistics: Model Management in the Real World
- 35:34Canary Model Helps Gauge Performance of New Model
- 36:10Rendezvous: Mainly for Decisioning Type Systems • Decisioning style machine learning - Looking for "right answer" - Simpler than interactive machine leaming (think self-driving car)
- 36:32Update on Machine Learning Logistics
- 37:54DataOps: Brings Flexibility & Focus
- 40:14Streaming Architecture: New Designs Using Apache Kafka & MapR Streams
related talks
36mApplied Machine Learning Playbook: From Idea to Production
08mConsiderations for producing the best AI and Machine Learning models
33mRunning Machine Learning in Production - a Journey to Success
38mBuilding robust machine learning systems
26mWhy Machine Learning Algorithms Fall Short...
06mMachine learning is about your data and deployment
04mThe age of machine learning
33mAI with a DevOps mindset - experimentation, sharing ..