Friday, March 17, 2017

The business of deep learning

The Business of Deep Learning
Understanding Deep Learning and Discovering Real World Applications
By Matt Coatney
Publisher: O’Reilly Media
Final Release Date: February 2017
Run time: 3 hours 30 minutes
Video: $64.99

course note

Overview: The first 2 lessons (~30 min)are free and give you an introduction and the 5 areas that will be covered:
  • characterize the world (36 min)
  • identify and group things together (27 min)
  • predict outcome and behavior (18 min)
  • human-machine interaction (30 min)
  • robotics and real-world interaction (22 min)
The details are charged. Then ends with implications (30 min) and conclusion(14 min).

Why now?
  1. Business need: hyper competition, knowledge economy
  2. data: natively digital, sensors, internet of things
  3. methods: large neural nets, ensemble/panel of experts
  4. compute power: exponential growth, GPU
  • more than just deep learning
  • ensemble of techniques approaching human intelligence
  • no clear definition or finish
4 possibilities for business models: (risk from high to low)
  1. platform( expensive, IBM Watson, Google tensor flow, Microsoft cntk, caffe, torch)
  2. consulting service (customized solutions, people intensive, not scalable)
  3. technique core (backbone for other specific technology)
  4. enablement (integration in application: Netflix,google, facebook, amazon)
  • which business models are most applicable to your industry?
  • what initial ideas do you have for monetizing deep learning?

characterize the world

extract features: image, audio, video, text

identifying and grouping things:

  • segmenting(clustering)
  • categorizing(classification)
dimensionality reduction: principle component analysis,
predicting outcomes/behavior
  • making recommendations
  • making prediction

building buy-in for AI projects

  • fear of change
  • identify a pressing,burning need
  • avoid pitching/building a platform
  • incrementally target new high-value needs
  • use past success as internal social proof
  • look for adjacent opportunity
  • follow the money

Driving User Acceptance

  • Teach: time, empathy, awareness, communication, help
  • If you communicate about 10 times as much as you should, you are starting to get it right.
  • create the climate for change: create urgency, form a powerful coalition, create a vision for change.