ML Starter Pack

Presented at GDG Waterloo.

Basics of ML

1. ML Starter Pack A comprehensive guide to Machine Learning April 22, 2023 GDG Waterloo, Canada
2. I'm Charmi! • ML Grad at Mila and UdeM • GDE for ML • International Speaker • Worked @ AWS, Shipmnts, ISRO • Tech, Art, Travel
3. Plan of the Day! • What is Machine Learning? • Types of ML • Supervised and Unsupervised techniques • Classification and Regression • Basics of Deep Learning • Basics of NLP, CV, ASR • The 3D ML Pipeline • Fun Applications
4. ? 90% have failed to solve this... = = = 33 30 36
5. 99% have failed to solve this...
6. Retrospection time!! What we just did? We solved logical math puzzle in a fraction of time
7. Think about, How we solved them? We had inputs We also had outputs
8. Think about, How we solved them? We created Rules and Patterns in our brain!!
9. Can machines create rules on their own?
10. Can machines create rules on their own? No, if it’s Classical Programming...
11. Can machines create rules on their own? YES, if it’s Machine Learning!!!
12. Classical Programming vs ML Classical Programming Machine Learning Rules Rules Data Data Answers Answers
13. Artificial Intelligence Deep Learning Machine Learning Any Technique that enables computers to mimic human intelligence & behaviour A subset of ML exposing multilayered neural networks to vast amount of data A subset of AI that includes statistical techniques to solve the tasks using experience
15. Basic Terminologies • Features • Labels • Examples ◦ Labelled example ◦ Unlabelled example • Data Split (Train, Valid, Test) • Models (Train and Test) ◦ Classification model ◦ Regression model
16. Supervised Learning • Supervised Learning deals with prediction of values based on given combinations of data and labels given beforehand • ML systems learn how to combine input to produce useful predictions on never-before-seen data • It is like learning with a teacher
17. Regression and Classification • A regression model predicts continuous values by fitting a line ◦ What is the value of a house in California? ◦ What is the probability that a user will click on this ad? • A classification model predicts discrete values by creating boundaries ◦ Is a given email message spam or not spam? ◦ Is this an image of a dog, a cat, or a hamster?
18. Iterative Learning
19. Overfitting vs Underfitting • An overfit model gets a low loss during training but does a poor job predicting new data • Overfitting is caused by making a model more complex than necessary • The fundamental tension of machine learning is between fitting our data well, but also fitting the data as simply as possible
20. Unsupervised Learning • It deals with clustering values or forming groups of values • One aims to infer patterns from the data rather than predicting values • It is like learning on your own
21. Evaluation Metrics https://en.wikipedia.org/wiki/Precision_and_recall, https://www.kdnuggets.com/2018/06/right-metric-evaluating-machine-learning-models-2.html
22. Evaluation Metrics https://en.wikipedia.org/wiki/Precision_and_recall, https://www.kdnuggets.com/2018/06/right-metric-evaluating-machine-learning-models-2.html
24. Deep Learning
25. Neural Network https://gfycat.com/gifs/search/neural+network
26. When to use or not use DL? • Deep Learning outperforms other techniques if the data size is large. But with small data size, traditional Machine Learning algorithms are preferable • Finding large amount of “Good” data is always a painful task • Deep Learning techniques need to have high end infrastructure to train in reasonable time • When there is lack of domain understanding for feature introspection, Deep Learning techniques outshines others as you have to worry less about feature engineering
27. When to use or not use DL? • Model Training time: a Deep Learning algorithm may take weeks or months whereas, traditional Machine Learning algorithms take few seconds or hours • DL never reveals the “how and why” behind the output- it’s a Black Box • Deep Learning really shines when it comes to complex problems such as image classification, natural language processing, and speech recognition • DL excels in tasks where the basic unit (pixel, word) has very little meaning in itself, but the combination of such units has a useful meaning
28. Natural Language Processing • Sentiment analysis • Chatbots • Machine translation • Speech recognition • Text summarization
29. Computer Vision • Facial recognition • Object detection • Medical imaging • Autonomous vehicles • Segmentation
30. Speech Understanding • Voice-controlled personal assistants • Virtual agents • Dictation software • Closed captioning • Subtitling for movies • Transcribing
31. The 3D ML Pipeline Design - Develop - Deploy
32. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
33. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
34. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
35. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
36. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
37. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
38. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
39. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
40. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
41. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
42. Data acquisition Model Deployment Data Cleaning Feature Engineering Model Validation Model Monitoring Model Selection Model Testing Model Training Hyper parameter tuning
43. It’s time for exploring some FUN applications
44. Classification on Custom Data https://teachablemachine.withgoogle.com/train
45. Sketch-RNN https://magenta.tensorflow.org/assets/sketch_rnn_demo/index.html
46. From where the data came Quick Draw: https://quickdraw.withgoogle.com/
47. RUNN = 🏃Run + 🤖RNN http://vibertthio.com/runn/
48. Experiments with Google https://experiments.withgoogle.com/collection/ai
49. The project Magenta https://magenta.tensorflow.org/
50. https://developers.google.com/machine-learning/
52. Let's connect! @CharmiChokshi Charmi Chokshi @charmichokshi coda.io/@charmi-chokshi/hi
53. Thanks! Questions? Download the deck