SlideShare a Scribd company logo

Meetup 18/10/2018 - Artificiële intelligentie en mobiliteit

•
•
Digipolis Antwerpen
Digipolis Antwerpen

This document discusses a self-learning computer vision AI that aims to make deep learning solutions more accessible. It outlines how current expert consulting is limited and expensive. The AI would use techniques like transfer learning, hyperparameter optimization, and Bayesian optimization to "learn how to learn" models without human experts. This could expand access to computer vision applications while also researching technologies that may make expert knowledge obsolete over time.

Related slideshows

Taking the plunge: Why you should use new technology on client projects
Taking the plunge: Why you should use new technology on client projectsTaking the plunge: Why you should use new technology on client projects
Taking the plunge: Why you should use new technology on client projects
 
The obstacles of developer productivity.pptx
The obstacles of developer productivity.pptxThe obstacles of developer productivity.pptx
The obstacles of developer productivity.pptx
 
Customer segmentation scbcn17
Customer segmentation scbcn17Customer segmentation scbcn17
Customer segmentation scbcn17
 
1 of 17
Download to read offline
Self-Learning Computer Vision AI By Jonatan
WebMonks Social Business in Computer Vision ● GOAL - jobs for Venezuelans ● Image Annotation ● Computer Vision Consulting
The Problem Access to Deep Learning Solutions ● DL very powerful ● Limited availability of experts ● Implications? ○ Subject matter expertise ○ Screen applications ○ Waste of experts’ time ● Lots of untapped potential!
We’re DL experts That offer consulting While researching tech That might make it obsolete
What does it look like today? 1. No more feature engineering 2. Repurposable through Transfer Learning !
What is the problem? 1. Seemingly easy a. Play with hyperparameters to make a model ‘learn’ b. Make the blue curve go up! 2. Hyperparameter space is huge 3. Building intuition requires insight & experience 4. Paying us to make a computer learn
Can’t a computer learn how to learn? ● What Google suggests ○ Neural Architecture Search ○ 100-1,000x computational power ● Stick with Transfer Learning!
Can’t a computer learn how to learn? ● Important aspects in CV ○ Collecting & cleaning data ○ Selecting a model ○ Configure your model ○ Hyperparameter Optimization (HPO) ● Computer Vision is an ideal candidate!
How does it work? ● Black box optimization ○ We can’t evaluate the derivative of the objective function → Like searching in the dark
How does it work? ● Many options! ○ 1 evaluation = how well does the model learn with this set of hyperparameters? ■ Excludes many optimization strategies!
How does it work? ● Bayesian Optimization ○ Build a probability model of the objective function ○ Treat OF as random ○ Prior → Posterior → Acquisition function ○ Spearmint - SMAC - Hyperopt - RoBO
How does it work? ● Problem with Bayesian ○ Sequential ○ Random takes many evaluations ○ Solution: behave like a gambler on slot machines → Hyperband
How does it work? ● Hyperband ○ Run random candidate settings ○ Limited iterations per candidate ○ Revisit good candidates → 70 times faster than random
How does it work? ● Hyperband on steroids = BOHB ○ Bayesian Optimization (BO)+ Hyperband (HB) ○ Speed of HB + optimization performance of BO
Our Experience With existing tools ● Client work ● (+) Within 5% accuracy range ● (-) Costly ● (-) Closed-source ● (-) Can’t export models
Our Experience With own research ● Traffic Analysis ○ Future project ○ Implemented simple POC ○ Combining with Synthetic Data ● Feel we can do better ○ Interested parties to collaborate?
Interested in any shape or form? jonatan@webmonks.io

More Related Content

Meetup 18/10/2018 - Artificiële intelligentie en mobiliteit

  • 2. WebMonks Social Business in Computer Vision ● GOAL - jobs for Venezuelans ● Image Annotation ● Computer Vision Consulting
  • 3. The Problem Access to Deep Learning Solutions ● DL very powerful ● Limited availability of experts ● Implications? ○ Subject matter expertise ○ Screen applications ○ Waste of experts’ time ● Lots of untapped potential!
  • 4. We’re DL experts That offer consulting While researching tech That might make it obsolete
  • 5. What does it look like today? 1. No more feature engineering 2. Repurposable through Transfer Learning !
  • 6. What is the problem? 1. Seemingly easy a. Play with hyperparameters to make a model ‘learn’ b. Make the blue curve go up! 2. Hyperparameter space is huge 3. Building intuition requires insight & experience 4. Paying us to make a computer learn
  • 7. Can’t a computer learn how to learn? ● What Google suggests ○ Neural Architecture Search ○ 100-1,000x computational power ● Stick with Transfer Learning!
  • 8. Can’t a computer learn how to learn? ● Important aspects in CV ○ Collecting & cleaning data ○ Selecting a model ○ Configure your model ○ Hyperparameter Optimization (HPO) ● Computer Vision is an ideal candidate!
  • 9. How does it work? ● Black box optimization ○ We can’t evaluate the derivative of the objective function → Like searching in the dark
  • 10. How does it work? ● Many options! ○ 1 evaluation = how well does the model learn with this set of hyperparameters? ■ Excludes many optimization strategies!
  • 11. How does it work? ● Bayesian Optimization ○ Build a probability model of the objective function ○ Treat OF as random ○ Prior → Posterior → Acquisition function ○ Spearmint - SMAC - Hyperopt - RoBO
  • 12. How does it work? ● Problem with Bayesian ○ Sequential ○ Random takes many evaluations ○ Solution: behave like a gambler on slot machines → Hyperband
  • 13. How does it work? ● Hyperband ○ Run random candidate settings ○ Limited iterations per candidate ○ Revisit good candidates → 70 times faster than random
  • 14. How does it work? ● Hyperband on steroids = BOHB ○ Bayesian Optimization (BO)+ Hyperband (HB) ○ Speed of HB + optimization performance of BO
  • 15. Our Experience With existing tools ● Client work ● (+) Within 5% accuracy range ● (-) Costly ● (-) Closed-source ● (-) Can’t export models
  • 16. Our Experience With own research ● Traffic Analysis ○ Future project ○ Implemented simple POC ○ Combining with Synthetic Data ● Feel we can do better ○ Interested parties to collaborate?
  • 17. Interested in any shape or form? jonatan@webmonks.io