Download as PDF, PPTX
Uploaded byTamir Taha
Intro to machine learning
This document provides an introduction to machine learning, including definitions, types, and case studies. It begins with an agenda and overview of artificial intelligence applications. It then defines machine learning as a field that allows computers to learn without being explicitly programmed. The main types of machine learning are described as supervised, unsupervised, semi-supervised, and reinforcement learning. Example case studies on Netflix recommendations, cancer diagnosis, and Amazon inventory are outlined. The document concludes with tips on prerequisites and resources for studying machine learning, including mathematics, programming tools, and course recommendations.
In this document
Powered by AI
Overview of the presentation by Hesham Gowaily; agenda includes AI introduction, ML types, and case studies.
Introduction to AI, discussing applications such as face recognition, chatbots, medical diagnosis, and autonomous cars.
History of AI development; highlights significant advancements like ImageNet and GPT-3, showcasing their scale.
Explains AI jargon and related terms including AI, ML, DL, DS, and Big Data, defining their roles and interconnections.
Definition of ML as computers learning from data; contrasts ML with traditional programming.
Illustrates how ML determines housing prices using a data-driven approach with parameters.
Describes various ML types: supervised, unsupervised, semi-supervised, and reinforcement learning.
Examples include Netflix's recommendation system, cancer diagnosis, and Amazon's inventory optimization.
Provides guidance for studying ML, including math prerequisites, key resources, and courses.
Q&A session, followed by contact details for further engagement and gratitude for participation.
More Related Content
Similar to Intro to machine learning
![1_Introduction to Machine Learning [Autosaved].pptx](https://cdn.slidesharecdn.com/ss_thumbnails/1introductiontomachinelearningautosaved-250910004933-3913b711-thumbnail.jpg?width=640&height=640&fit=bounds)
Intro to machine learning
- 1. Intro to MachineLearning By: Hesham Gowaily
- 2. Agenda - Introduction toAI. - What is Machine Learning. - Types of Machine Learning. - ML Case Studies. - How to study ML.
- 3.
- 4. AI Applications • FaceRecognition • handwriting analysis • Automatic Captioning • Object Tracking • Image Styling • Machine Translation • Sentiment Analysis • Chatbots • Automatic Text Generation • Virtual Assistants • Voice Search • Speech generation • Medical Diagnosis • Personalized Medicine • Early detection • Trading bots • Stock predictions • Fraud Detection • Personalized Ads • Product Recommendations • Autonomous cars • Robotic Surgery
- 5.
- 6. The Big DataEra Data • Large volumes of data is produced everyday. • Everyone has a phone packed with several sensors. Infrastructure • The computing power of GPUs has increased dramatically. • Cloud providers offer online computing (IaaS). Services • User applications: YouTube, Gmail, Facebook, Twitter. • Online storage available for free or low cost.
- 7. Notable AI Achievements ImageNetis a database of 14 million images with over 20,000 categories. GPT-3 is a language model with 175 billion learning parameters.
- 8.
- 9. Overlapping AI RelatedTerminology • Artificial Intelligence (AI) Trying to simulate human intelligence. • Machine Learning (ML) Learn by example from experience and historic data. • Deep Learning (DL) Learn patterns using multi-layered data processors. • Data Science (DS) Uses a variety of scientific methods, processes and systems to solve problems involving data. • Big Data Analyze data sets that are too large or complex. Artificial Intelligence Machine Learning Data Science Deep Learning Big Data
- 10. What is MachineLearning
- 11. What is MachineLearning The subfield of computer science that “gives computers the ability to learn without being explicitly programmed”.(Arthur Samuel, 1959) Using previous data for answering future questions Historic may contain answers or may not contain answers Training Prediction Labeled Unlabeled
- 12. Machine Learning vs.Traditional Programming Traditional Programming Machine Learning Data Rules Answers Data Answers Rules Traditional Programming: • Business requirements and data are analyzed. • A set of hard-coded rules are programmed and tested. • Program process new data based on the coded rules. Machine Learning: • Data and their labels (answers) are fed into a model. • Model “learns” useful features and frequent patterns to predict answers. • Trained model is used to “predict” answers for new data.
- 13. • Traditional Approach: Price= 1.2 x Area + 0.7 x # Bedrooms + 0.3 x # Bathrooms Pricing formula is known beforehand and is explicitly hard- coded. The formula can be deducted by manual analysis or SME domain experience. • Machine Learning: Price = A x Area + B x # Bedrooms + C x # Bathrooms Pricing Formula is unknown at the beginning and would need the model to be trained to “Learn” the formula attributes A,B and C. HOUSING PRICES Estimate housing prices based on 3 features (properties): Area of the House, Number of Bedrooms, Number of Bathrooms
- 14. Housing Prices Area #Bedrooms#Bathrooms Price 130 3 1 1,200 160 3 2 1,500 90 2 1 900 … Model Hyperp aramet ers Optimiz ation Price = A x Area + B x # Bedrooms + C x # Bathrooms A = 1.247 B = 0.682 C = 0.319 Input Dataset: Contains prepared historic data of actual house sales. Model: Model Learns appropriate “Parameters” to “Fit” the input data. Output: Parameters that completes the formula and can be generalized to predict unsold houses.
- 15. Types of MachineLearning
- 16. Types of Learning •SupervisedLearning •Unsupervised Learning •Semi-supervised Learning •Reinforcement Learning
- 17. Supervised Learning • Learnthrough examples collected from historic data. • Examples contain the desired output (labels) that will be predicted for future data. • Is this a cat or a dog? • Is this email a spam or not? • What is the market value of a house given its area and number of bedrooms? Supervised Unsupervised Semi- Supervised Reinforcement
- 18. Supervised Learning Output iscontinuous. Predicts numerical values such as prices or temperature. Supervised Unsupervised Semi- Supervised Reinforcement Regression Classification Output is discrete. Predicts categorical labels such as: Cat or Dog.
- 19. Unsupervised Learning • Usinghistoric data that has no labels. • Discovers the intrinsic links of data. • Group photos into 20 groups based on their metadata. • Segment customer profiles based on their demographics and purchase behavior. • Find an anomaly in credit card usage patterns. Supervised Unsupervised Semi- Supervised Reinforcement
- 20. Unsupervised Learning • Usefulfor learning structure in the data (clustering) or detecting outliers (anomaly). Supervised Unsupervised Semi- Supervised Reinforcement Anomaly
- 21. Semi-Supervised Learning • Historicdata has a small amount of labeled data, and a large amount of unlabeled data. • The cost of manually labeling all data is prohibitive. • The problem is initially treated as Unsupervised to group data to different structure. • After that, available labels are used to label entire clusters. Supervised Unsupervised Semi- Supervised Reinforcement
- 22. Reinforcement Learning • Anagent interacts with an environment and watches the result of the interaction. • Environment gives feedback via a positive or negative reward signal. • The agent learns to optimize its interactions to maximize the reward. • An autonomous vehicle learns to put safety first, minimize ride time, and obey the rules of law. • An stock trading agent can decide to buy, sell or hold based on market status and transactions profit/loss. Supervised Unsupervised Semi- Supervised Reinforcement
- 23.
- 24. ML Productizing AI-First AI-Inside Actionable Insights AItech is at the center and is essential to the product function. Examples: Virtual assistants, Chatbots, self- driving cars. AI adds a useful function that enhances user experience. Example: Recommendation engines, process automation, Fraud detection AI leveraging data that you collect to make informed decisions. Examples: Sales forecast, Churn analysis
- 25. Case Study: NetflixRecommendation Personalized recommendation using Collaborative filtering Scale: • Volume: 13,612 titles (2019) • Subs: 159 million (2020) Results: • High engagement rate, Low churn • Personalization and recommendations save Netflix more than $1Billion per year.
- 26. Case Study: InfervisionCancer Diagnosis Predominantly used in early-stage lung cancer screening. Employs more than 50 deep learning algorithms to determine each diagnosis Scale: • Trained using over 200,000 scans in trials at 20 hospitals. Results: • Helped reduce the rate of missed cancer diagnoses by 50 percent.
- 27. Case Study: AmazonInventory Optimization ML-powered inventory optimization ensures that inventory preemptively caters for forecast demands. Scale: • ship an average of 10 million packages per day. Results: • Store 40% more inventory. • Fulfill 1 and 2 days shipping on time.
- 28.
- 29. Study Pre-requisites Math ProgrammingTools/Concepts Probability and Statistics Python Modern IDE’s Linear Algebra R Data tools Calculus
- 30. Mathematics Study tips •Probability Book A First Course In Probability 9th ed. • Mathematics for Machine Learning 3Blue1Brown • Coursera: Mathematics for Machine Learning, by Imperial College of London https://www.coursera.org/specializations/mathematics-machine-learning • Book: Mathematics for Machine Learning https://mml-book.github.io/
- 31. • Coursera: MachineLearning, offered by Stanford https://www.coursera.org/learn/machine-learning • YouTube: Stanford CS 229 – Machine Learning (Math focused) https://www.youtube.com/playlist?list=PLoROMvodv4rMiGQp3WXShtMGgzqpfVfbU • Book: Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow, 2nd Edition https://www.oreilly.com/library/view/hands-on-machine-learning/9781492032632/ Machine Learning Study tips
- 32.
- 33.