10 Things Every PHP Developer Should Know About Machine Learning
2 likes117 views
This document provides an overview of machine learning and discusses 10 things every PHP developer should know about machine learning. It begins with introducing the presenter and providing background on machine learning. It then covers topics such as what machine learning is, why developers should care about it, how it works, different machine learning approaches, where it is used, and how to get started. For each of the "10 things", it provides more detailed explanations and examples related to recommendations, content analysis, computer speech, and computer vision. It encourages attendees to get started with machine learning and provides various open-source and cloud-based options to do so.
10 Things Every PHP Developer Should Know About Machine Learning
- 1. 10 THINGS EVERY PHP DEVELOPER SHOULD KNOW ABOUT MACHINE LEARNING
- 2. Fill in the gaps and squash hype around M.L. Build the case for using it now. And provide easy ways to get started. TODAY’S GOALS
- 3. Background #1: What it is #2: It’s taking over #3: How it works #4: Different approaches #5: Where it’s used #6: How to get started OUR JOURNEY Code #7: Recommendations #8: Content analysis #9: Computer speech #10: Computer vision
- 5. ABOUT ME chris.mohritz@10xeffect.com ● Lifelong entrepreneur ● Deep technology background (strategy, not developer) ● Using A.I. (machine learning) in business since 2009 ● Opening a startup accelerator in Vegas
- 6. HOW I GOT STARTED Apache Mahout (Decision Forest) Behavior prediction Suite of mobile apps Determine most relevant (highest-converting) sales offer to present to each individual user — and the best (highest-converting) time to present it. circa 2009
- 7. Will the current user buy “Madden NFL” right now? WHAT IS A DECISION FOREST? is male? is age > 16? is Y app installed? is X app installed? end has used > 30 days? was X function used? was Y function used? no yes no yes no yes no yes end (better ways to do this now) no yes end do it
- 8. #1: WHAT IS IT?
- 9. “Field of study that gives computers the ability to learn without being explicitly programmed.” ~ Arthur Samuel, 1959 MACHINE LEARNING IS... analyticsvidhya.com/blog/2015/07/difference-machine-learning-statistical-modeling
- 10. Training your computer to do stuff, just like you would train a pet. IN OTHER WORDS...
- 11. SIMILAR TO HOW WE LEARN Data System Output Model Question Answer Emotions Mindset Algorithm The reference data pattern (decision-making stuff) Process the computer uses to ‘learn’ the model The model is built from historical data Training data Life experience Perspective Algoritm
- 12. AT THE END OF THE DAY... It’s all pattern recognition.
- 13. #2: WHY SHOULD I CARE?
- 14. SOFTWARE IS EATING THE WORLD Everything is becoming code. Software automates, simplifies, and accelerates business.
- 15. BUT... Someone needs to write the code/logic.
- 16. THE TRADITIONAL WAY Handwritten logic. If / Then / Else
- 17. THE PROBLEM Complex situations require complex2 software logic.
- 18. THE SOLUTION Trained logic using historical data. Model tripID hasEggs eggsBough t milkBought 1 1 6 1 2 0 0 2 3 1 6 1 4 1 8 1 5 0 0 1 6 1 6 1 7 0 0 3 0011001101011101 0101100111010 11001001101 Stored as a mathematical model. Finds patterns in the data.
- 19. WHAT IT LOOKS LIKE console.aws.amazon.com/machinelearning/home?region=us-east-1#/datasources
- 20. M.L. IS EATING THE SOFTWARE All applications are becoming “smart” — with unprecedented complexity in logic. Machine learning automates, simplifies, and accelerates software.
- 21. ENDLESS OPPORTUNITIES Everything mankind has ever invented — including all software apps — will be reinvented using A.I.
- 22. A QUICK NOTE... In the near future, A.I. be writing it’s own code.
- 23. #3: HOW DOES IT WORK?
- 24. STILL A MURKY LANDSCAPE Artificial Intelligence Machine Understanding (?) Pattern recognition Classification Prediction Can only do one thing Brute-force approach Autonomous decisions Universally applicable Intuition approach Google DeepMind Amazon Machine Learning Natural language processing Computer vision Optimization IBM Watson Classic learning Multi-tiered deep learning neural networks Deep learning neural network Explicit ProgrammingHandwritten Machine Learning logiccomplexity
- 25. IT’S ALL CLASSIFICATION via: wjscheirer.com
- 26. “Features” Points of differentiation within the data. How would you teach a child to recognize the differences? ● Distance between eyes ● Width of nose ● Shape of cheekbones ● etc. HOW DOES IT CLASSIFY?
- 27. “Probability” Each potential answer gets a numeric probability calculated for it. Higher probability means greater confidence. HOW DOES IT MAKE DECISIONS?
- 28. ● Supervised learning — Labeled training data ● Unsupervised learning — Unlabeled training data ● Transfer learning — Applying aspects across models ● Reinforcement learning — Reward-based training TRAINING gym.openai.com
- 29. #4: WHAT ARE THE DIFFERENT APPROACHES?
- 30. REMEMBER... Data System Output Model Question Answer Emotions Mindset Algorithm The reference data pattern (decision-making stuff) Process the computer uses to ‘learn’ the model The model is built from historical data Training data Life experience Perspective Algoritm
- 31. Who wants to be a data scientist? ENDLESS ALGORITHMS docs.microsoft.com/en-us/azure/machine-learning/machine-learning-algorithm-choice machinelearningmastery.com/a-tour-of-machine-learning-algorithms
- 35. (SIMPLE) NEURAL NETWORK Each layer performs a discrete function ≥ 1 input neurons ≥ 1 output neurons ≥ 1 hidden layers Output “fires” if all weighted inputs sum to a set “threshold” Each connection applies a “weighted” influence on the receiving neuron Layers build on each other (iterative) Each input can be a separate “feature” Each neuron takes in multiple inputs Hidden layers can’t directly “see” or act on outside world cs231n.github.io/neural-networks-1
- 36. HOW MUCH IS A HOUSE WORTH? Decisions based on combinations. 3 bedrooms 37 years old 1450 ft2 $191,172 Is it “old” or “historic?” Is it “small” or “open floor plan?” $32,108 per bedroom $64,251 per acre Need a lower weight for “old” Apply initial abstractions Set values cs231n.github.io/neural-networks-1
- 37. #5: WHERE IS IT USED?
- 38. ENDLESS USES ● Classifying DNA sequences ● Economics ● Fraud detection ● Medical diagnosis ● Search engines ● Speech recognition ● Job search ● Spam filtering ● Risk prediction ● Visual product search ● Create art / music ● Industrial design ● Image caption generation ● Facial recognition ● Colorization of b&w images ● Adding sound to silent movies ● Language translation ● Image editing ● Vehicle navigation ● Error detection
- 39. IN THE WILD Recommender (pick from list) Classifier (binary) Visual recognition (deep learning)
- 40. #6: WHERE SHOULD I START?
- 41. ● You don’t need a supercomputer ● You don’t need to write a ton of code ● You don’t need to invest massive amounts of time ● You don’t need a data science degree ● You don’t need to be a math whiz ● You don’t need mountains of data MYTH BUSTING
- 42. IT’S EASIER THAN YOU THINK Forget theory, just do it.
- 43. ● Amazon Artificial Intelligence ● Google Cloud Machine Learning ● Microsoft Cognitive Services ● IBM Watson * ● DiffBot * - PHP library is 3rd-party SaaS OPTIONS
- 44. ● TensorFlow * ● Amazon DSSTNE * ● H2O * ● PredictionIO ● Apache Mahout ● Scikit Learn ● Caffe * OPEN SOURCE OPTIONS ● Microsoft CNTK * ● Torch * ● Theano * ● MXnet * ● Chainer * ● Keras * ● Neon * * ANN / Deep learning
- 45. ● archive.ics.uci.edu/ml ● deeplearning.net/datasets ● mldata.org ● grouplens.org/datasets ● cs.toronto.edu/~kriz/cifar.html ● cs.cornell.edu/people/pabo/movie-review-data ● yann.lecun.com/exdb/mnist (handwriting) ● kdnuggets.com/datasets/index.html (long list) ● image-net.org (competition) OPEN SOURCE DATASETS
- 47. A CUSTOMER-DRIVEN WORLD Today, consumers control the brand-customer relationship. They choose when and how they interact. Brands need to create attractive experiences that draw consumers in — through highly relevant communications and products.
- 48. PRODUCTS
- 49. CONTENT
- 50. ENDLESS APPLICATIONS ● Visitors who viewed this product also viewed ● Visitors who viewed this product ultimately bought ● You might also like ● Recently viewed ● Trending in category ● Site-wide top sellers ● Customer also bought ● Other customers who bought this product also bought ● Items viewed with items in your cart ● Top sellers from your recent categories on homepage
- 52. Recommendations Build FBT Build Model Application Training Catalog Training Usage HOW IT WORKS Related item recommendations Recommendations API Frequently bought together recommendations
- 53. PLAN PLAN LIMITS PRICE Free 10,000 calls / mo Free S1 Standard 100,000 calls / mo $75 / mo (overage at $0.75 / 1000 calls) S2 Standard 1,000,000 calls / mo $500 / mo (overage at $0.75 / 1000 calls) S3 Standard 10,000,000 calls / mo (overage at $0.75 per 1K calls) $2,500 / mo (overage at $0.75 / 1000 calls) S4 Standard 50,000,000 calls/mo $5,000 / mo (overage at $0.75 / 1000 calls) PRICING
- 57. IBM WATSON NATURAL LANGUAGE UNDERSTANDING ibm.com/watson/developercloud/natural-language-understanding.html ● Code: github.com/10xNation/ibm-watson-natural-language-understanding-php ● GUI: natural-language-understanding-demo.mybluemix.net
- 61. GOOGLE CLOUD VISION cloud.google.com/vision Guide: gigaom.com/2017/02/06/harnessing-visual-data-using-google-cloud Code: github.com/GoogleCloudPlatform/php-docs-samples/tree/master/vision/api
- 62. CLOSING
- 63. Intro blog posts: ● Artificial Intelligence 101 (the big picture) ● Machine Learning 101 (what you’ll actually use) New ‘How to Apply A.I. in Your Business’ blog series: ● Voice-Powered Products w/ Amazon Alexa ● Predictive Social Media w/ IBM Watson(live) ● Image Recognition w/ Google Cloud ● Recommendation Engine w/ Microsoft Azure GO DEEPER