MSA Bootcamp 14 April Slides


2020/04/15 发布于 技术 分类

1. Welcome to MSA AI Bootcamp 1. We will be commencing shortly 3. If you have any questions, please select “Ask a question” or upvote by liking existing ones. 2. 4. 5. 6. 7. 8. This is a Microsoft Teams Non-tech Live event, so you are placed on mute for the entire duration. Captions can be turned on anytime by clicking the “CC” icon. At any point where you get lost, you can rewind the live stream to any point in time. This Bootcamp is being recorded and all recordings can be accessed via the same link you used to access this live event. Having this Bootcamp being recorded, should you not consent, please feel free to leave if you wish. For updates on the program, please join our Facebook group at We hope you enjoy the session!
2. presents Artificial Intelligence and Analytics
3. Agenda 01 Keynote 05 Project Walk-through 02 Introduction to MSA 06 Instructions 03 AI & ML Concepts 07 Questions and Answers 04 Break Project Submission & Assignment
4. Speaker Keynote
5. Joseph Gurney Graduated from the University of Tasmania in 2018, Joseph is now working at Microsoft as a Premier Field Engineer in Data & AI. Role Premier Field Engineer - Data & AI Company Microsoft – Sydney, Australia
6. Disruption is the 4th industrial revolution 1780s STEAM 1870s ELECTRICITY 1970s ELECTRONICS & IT 2015+ DIGITAL
9. Introduction to MSA
10. Overview The Microsoft Student Accelerator (MSA) program works with students to provide industry relevant training and put those skills in practice to solve a real-world problem. Training Start with 6 months of bootcamps, workshops and projects Imagine Cup Mentorship Top performing students will then be mentored for the Imagine Cup preparation
11. Structure  Consists of a series of bootcamps and complimenting workshops.  Participants are expected to complete at least 2 bootcamps to be certified as having completed the MSA Program.  Each bootcamp will require students to finish a project and corresponding MS Learn modules. Bootcamp Project Workshop
12. Bootcamps vs Workshops Bootcamp Topics AI & Advanced Analytics Cloud Computing with Azure Web App and APIs Supporting Workshops Git & GitHub DevOps and Project Workflow Presentation and Soft Skills
13. Timeline Advance Analytics & AI Bootcamp (13th & 14th April) April Azure/ Cloud Fundamentals Bootcamp (20th May) Web Apps + Cognitive Services Bootcamp (End of July) May July Industry Night Imagine Cup Training Aug Oct
14. Points  Part of the MSA program and is an indicator of your progress. You can accumulate points through completing projects and Microsoft Learn modules.  The top students with the highest number of points will be mentored to participate in the Microsoft Imagine Cup 2020. Project Microsoft Learn Modules
15. Prizes  Based on the points accumulated, while top students can earn exclusive Microsoft swags and goodies, there are special prizes for the top 3 participants. 01 02 Surface Book 2 Xbox One X 03 Surface Headphone
16. Support and Resources  Access to exclusive tools and resources such as GitHub, Microsoft Learn and helpdesks.  GitHub will be your primary source of information  Helpdesks enable participants to ask questions and get help from our volunteers. GitHub MS Learn Helpdesks
17. AI & ML Concepts
18. Story Time!
20. Artificial Intelligence  Theoretical concept of “smart” or “sentient” machines  Machine which simulate human-like behaviour  Earliest examples date as early as Greek Mythologies (Pandora’s box)
21. AI vs Machine Learning vs Deep Learning Artificial Intelligence  Terms often used interchangeably Machine Learning  Are subsets of each other Deep Learning
22. Machine Learning  Practical application of AI using programming, mathematics and statistics  Building mathematical models and algorithms in order to identify patterns or generalize information  Can make decisions or predictions based on those knowledge
23. Deep Learning  Subfield of ML inspired by the structure and function of the brain  “Deep” refers to the number of layers through which the data is transformed  Commonly applied for computer vision, speech recognition, natural language processing etc.
24. Why Do We Need Machine Learning?  The amount of data is growing exponentially  Finding patterns and other useful information in data is becoming considerably difficult for humans  Machine Learning gives us the promise to derive meaning from the data  Frees humans to engage in more creative or decision making tasks
25. Image recognition & analysis Common Uses of Machine Learning Fraud detection Recommendation systems Text & speech systems Bioinformatics
26. How Does Machine Learning Work? A typical model can be broken down to: 1. 2. 3. 4. 5. Identify the problem then gather and prepare data for the problem we’re trying to solve Select an appropriate ML model based on the problem Train our model on the training data Test our model and optimise Launch it to the real world!
27. Data Collection and Processing  Very important to collect a wide variety of data  If not provided in a correct format, the algorithm would perform incorrect analysis  Missing data can significantly decrease the accuracy of the model  Focus on the features or values we want and drop the insignificant ones
28. Common Categories of Machine Learning Supervised Learning Unsupervised Learning Reinforcement Learning
29. Supervised Learning  Is like learning with a teacher  Use past data to predict future outcomes  Uses a trial and error bases approach  Often used for prediction and classification
30. When Supervised Learning Is Used? Classification: Machine is trained to classify something into some class.  classifying whether a patient has disease or not  classifying whether an email is spam or not Regression: Machine is trained to predict some value like price, weight or height.  predicting house/property price  predicting stock market price
31. Unsupervised Learning  Is like learning without a teacher  The machine learning through observation &  Tries to learn some type of structure from the data  No specific way to compare model performance in most unsupervised learning methods.
32. When Do We Use Unsupervised Learning? Clustering: A clustering problem is where you want to discover the inherent groupings in the data  Such as grouping customers by purchasing behavior Association: An association rule learning problem is where you want to discover rules that describe large portions of your data  Such as people that buy X also tend to buy Y
33. Reinforcement Learning (RL)  The models consists a decision process(s) and reward system  An agent (model) interacts with the environment to maximise the total rewards.  RL is usually modelled as a Markov Decision Process (image below):
34. When Not To Use Reinforcement Learning? You can’t apply reinforcement learning model in all situation. Here are some condition when you should not use it: • When you have enough data to solve the problem with supervised learning • You need to remember that reinforcement learning is a computingheavy and time-consuming Common real-world applications of RL: • A machine learning model to play chess • Intelligent traffic control system Input Generated
35. Importance of Proper Testing  How the model generalizes unseen data is very important – Overfitting & Underfitting (Demystified in later slides)  Fine tune model’s parameters and optimise  Choose the best performing model Common Model Evaluation Techniques: Holdouts, Cross-Validation. Read more about them here.
36. Demystifying Common Jargons
37. Weights, Biases & Hyperparameters Weights: Initialised values that decide much influence the input will have on the output. They are updated as our model trains itself. Biases: Constant values which act like pre-determined notions. And additional input Hyperparameters: Parameters whose value is set before the learning process begins. Example: learning rate, batch-size etc.
38. Train, Test & Validation Train: Go through the training datasets to generate output Validation: Validate the output generated during training and update weights and biases Test: Previously unseen data. Used to determine model’s accuracy. Tune hypermeters to optimise
39. Overfitting & Underfitting Overfitting: When the model memorizes the features and information of the training data rather than generalizing it. Underfitting: When the model can neither generalize the training data nor unseen data
40. Thank You! Please ask any questions in the Q&A section of the Livestream
42. Break (10 mins)
43. Project walkthrough
44. Project Walkthrough  A sample to teach you how to use AzureML and Azure Notebooks in your project  Uses MNIST Dataset – Predicting Handwritten digits  *If you get lost, you can always re-watch this recording later at your own pace
45. Project Submission & Assignment Instructions
46. Assignment Instructions  In order to complete this Bootcamp, you will need to complete its corresponding Microsoft Learn modules or tutorial and a project.  All of these will be detailed on GitHub which you can access via a link we sent to you via email. 100PT Project 3PT Microsoft Learn Module
47. Project Submission  Submit GitHub links and screenshots of completed tasks here:  The assessment of this bootcamp is due by 13th June 2020
48. Questions and Answers
49. Thank you for attending! 1. The recording for this Bootcamp can be accessed via the same link used to access this live event. 2. All resources can be found on our GitHub repo which can be accessed via the link we sent to your email.