Humanizing Machine Learning - Real-World Applications
3 likes920 views
The document discusses the application of machine learning in various industries, highlighting examples like real estate dynamic pricing, Con Edison’s manhole prediction model, and Netflix's customer engagement strategies. It emphasizes the importance of data discovery and supervised learning techniques, as well as applications like anomaly detection and natural language understanding. The key takeaways include the necessity of a robust discovery phase for model accuracy and the potential of machine learning to enhance decision-making processes.
Humanizing Machine Learning - Real-World Applications
- 2. MACHINE LEARNINGARTIFICIAL INTELLIGENCE != Myths of Machine Learning
- 3. • Programming with Data not Heuristic • Features and Observations • Black Box Algorithm (of sorts) • Training & Validation What is it? f(x) Features Outputs
- 5. Example 1: Real Estate • Dynamic Pricing Suggestions • Image Analysis on Listing Photos • Two-Sided Marketplace Personalization • Hot Homes • Home Recommendations • Zestimate Calculated Using ML Model Crowdsourced Competition
- 6. Example 2: Con Edison • Randomly, Manholes Explode in NYC • Developed Machine Learning Model to Predict Which Manhole Covers are Likely to Blow • The Top 2 Percent of Manholes Ranked as Vulnerable by the Algorithm Included 11 Percent of the Manholes That Had Recently Had a Fire or Explosion. PREDICTING SERIOUS MANHOLE EVENTS
- 7. Example 3: Netflix • Every List is Influenced by Machine Learning • Using Machine Learning, They Were Able to Reduce Customer Churn and Spread Viewership Across More Videos • Gain User Trust by Providing Reasons • A/B Testing in Production
- 8. How We’re Leveraging Machine Learning
- 9. Big Data Architecture Historical DataFeed Blend Streaming Platform API Cached Data 3 Party APIrd ApplicationsMachine Learning Inference Machine Learning Training / Analytics
- 10. Fee Generator IDEA • Conducted a data discovery and built a predictive model to generate pricing points. MODEL AND TECHNIQUES • Supervised learning using random forest model. KEY TAKEAWAY • Discovery phase is crucial to producing relevant and accurate model. OTHER APPLICATIONS • Anomaly detection, predictive analytics, complex regression. * *
- 11. Chat Analyzer IDEA • Perform analysis on internal messaging platform to produce a model that categorizes each message by content and sentiment. MODEL AND TECHNIQUES • Natural language understanding (NLU) and sentiment analysis model built using a deep neural network with embedding, recurrent, and dense layers. KEY TAKEAWAY • Reduce the scope of the problem domain for faster training and better results. OTHER APPLICATIONS • Translation (Google, Facebook), Spam Detection, Information Extraction.
- 12. Epoch 1 Output <end> a <end> , . <end> i the <end> i <end> <end> and <end> , and <end> ’ , the <end> ? <end> <end> <end> <end> <end> <end> ? a ? a a to <end> <end> i <end> the , <end> ’ ’ <end> <end> <end> a . <end> <end> , ? . ’ <end> <end> . <end> <end> ? <end> <end> and the <end> <end> <end> the <end> and ’ <end> <end> <end> <end> <end> ’ <end> i a <end> the to <end> <end> , ’ ’ <end> <end> i and <end> . <end> <end> <end> a <end> <end> ,
- 13. Epoch 3 Output <end> i think <end> . <end> is a channel <end> < <link> > <end> commented <link> | i the go to get a the be . , . <end> < <link> > <end> < <link> | <end> , <end> the have the the to be to a have to . to ’ <end> > has < <link> > <end> < <link> > t file m joined . <end> the can to go <end> to ’ m file m be <end> , the have to <end> <end> <end> <end> i ’ ve be to the ’ m i ’ s channel <end>
- 14. Epoch 12 Output <end> i ’ m not to have the , that was <end> /gif <end> i think it ’ s in it <end> <end> the one <end> the of is a bit . <end> i was going to the <end> that can have that to the <end> < <link> > <end> i ’ m going to be a <end> i would do you have a good , but <end> it would go . <end> i ’ m not sure i think it ’ s the best is going to go with the , but i need it to be in this one
- 15. Epoch 16 Output that is a cool for my phone . <end> i ’ ve been a new fan of the office . i would be in the <end> i think the . <end> / gif that i am <end> i ’ m just in it to be <end> < <link> > <end> i ’ m in <end> /gif i have to get a good . <end> / gif i ’ m down for the , but it ’ s not <end> i ’ m just in one of the <end> i was <end> < <link> > <end> i ’ m going to get the <end>
- 16. Epoch 17 Output i am . <end> that has not <end> /gif <end> it ’ s <end> i ’ ll be there about the next of the same time <end> i just have my phone . <end> so <end> /gif that was a good , and i ’ d love to the <end> < <link> > <end> < <link> > <end> < <link> > <end> <end> /gif <end> <end> < <link> > <end> < <link> > <end> i am going to get the , <end> this is awesome ! <end> yes <end> < <link> > <end> that is amazing <end> i ’ m not
- 17. Epoch 25 Output < <link> > <end> < <link> > <end> < <link> > <end> < <link> > <end> this is a great <end> i ’ m not going to go . <end> i ’ m in the same <end> /gif <end> i am <end> <end> /gif that is that <end> is a <end> /gif <end> i have a bit to the office <end> <end> it ’ s a little idea that i ’ ve got that . <end> /gif <end> i ’ d like a new shirt and it ’ s not a little <end> <end> i know <end> /gif <end> /gif <end>
- 18. Epoch 27 Output yeah . i think this was a good thing , <end> that would be good , you should go to the game ! < <link> > <end> /gif <end> <end> this song was pretty neat <end> < <link> > <end> this is a lot of the one of the <end> and they can ’ t have the same <end> < <link> > <end> < <link> > <end> < <link> > <end> < <link> > <end> < <link> > <end> @martin pinned a message to this channel - > <end> this is <end> /gif <end> /gif <end> the one is @here <end>
- 19. Epoch 28 Output i think they could have a lot of of the way . <end> the of the of the <end> the first guy is to me with it and then i ’ m not in that . <end> yeah i have a good idea and i think the . the of my life is a <end> but if anyone needs to take your own to go to the ? <end> that would be good , but i ’ m gonna be at least . the of the <end> /gif <end> i ’ m <end> i was a good idea <end> i don
- 20. PROPERTY OF DIALEXA LLC ©2017 - CONFIDENTIAL & PROPRIETARY HONESTY If it’s not true, don’t say it TRANSPARENCY Open dialogue is the oxygen that great companies and great ideas breathe RESILIENCE Greatness requires toughness INTEGRITY Doing the right thing even if no one is watching EXCELLENCE Success is a product of execution, not luck TRUST Believe in each other, and strive to be worthy of that belief CREATIVE FREEDOM The status quo is never sacred HUNGER Big results require big ambition Any Questions? Greg Corley VP OF TECHNICAL OPERATIONS https://www.dialexa.com Rowdy Howell SOLUTIONS ENGINEER https://www.dialexa.com