Overview

Today artificial intelligence is a term used widely across the world. However it also means different things for different people. The only thing that does not change is, how to transform a business requirement to a machine learning problem statement. In this context it is important to understand the facts and myths behind the AI, ML, DL and RL methodologies to understand scope and limitation. The workshop is structured to cater to the clarification learning requirement of the machine learning enthusiasts.

Objectives

  • To obtain the skill of translating a business requirement to a machine learning problem statement
  • To understand the scope and limitation of machine learning methodology
  • To learn mathematical essentials of deep neural networks
  • To get initialized into reinforcement learning and unsupervised learning approaches
  • Summer School Topics to be covered

    Supervised machine learning as pattern mapping problem via examples - vector representation, loss function minimization via gradient descent methodology; multiclass classification via logistic regression and multivariate linear regression; data issues - feature characteristics - homogenous, non-homogenous, missing values, class imbalance, noise, label corruption; bias and variance, cross validation, metrics & plotting, concept of automatic differentiation for gradient computation.
    Introduction to reinforcement learning methodology - concepts of agent, environment, state space, reward; Markov and Q-learning approaches; Examples of RL formulation.
    Introduction to neural networks - backpropagation and weight update, CNNs for object detection; Encoder-decoder networks for image segmentation and GAN methodology.
    Machine learning facts and myths, sharing of ML experiences by Industry personnel, sharing of ML problems by Institute faculty and interactive sessions.

    Prerequisites for Summer School

    The audience are expected to self-help to learn the basic calculus, coordinate geometry and python to be able to follow the lectures. Those who have had prior background on these topics but lost touch recently, need to refresh their memory and aptitude levels. There will be people interested in machine learning but have not got exposure to the above topics and they need to definitely practice well from the scratch before the summer school. We have also seen people using ML jargon a lot but not at all have even basic mathematical concepts clear, they need to calm down and go through prerequisite weblinks listed below carefully before the summer school.
    Mathematics: College level mathematics, Partial derivatives
    Python: Learn Python, Numpy tutorials, Pandas Library
    Machine Learning: Pytorch Library, Hands On, Scikit Learn, Coordinate Geomerty

    Expected outcomes

    After successfully completing the summer school, the participants would be able to

    Identify vectorization requirements in business scenarios
    Define a mapping function, loss function and formulate a machine learning solution
    Understand classification and regression settings, standard metrics and visualization
    Understand and build simple deep neural networks and convolutional neural networks
    Get an idea of reinforcement learning methodology and unsupervised learning
    Have a glimpse of problem statements in Industry and faculty laboratories
    Obtain practice codes for hands-on exercises

    Schedule

    Click here to Download the Schedule


    Day 10 - 11:30 11:30 - 11:40 11:40 - 13:10 13:10 - 14:00 14:00 - 15:30 15:30 - 15:40 15:40 - 17:10
    July 25, 2022 Business problem to ML problem formulation - Vector representation; The Five Steps of Supervised Learning; Types of Data; Types of Mapping; Loss functions; Use case scenarios Tea Break ML Methodology - Gradient descent; Multi variate linear regression; Multi-class classification (Logistic regression); Standard ML Metrics & Plots Lunch Break Code and data demonstrations in SKLEARN - Data sets for conceptual and real for text, numerical, image, audio and video types; Classification models; Model comparison Tea Break Data complexities - Noise in data, missing values, class imbalance, wrong labels; Preprocessing; Production data deviations and setting alarms
    July 26, 2022 Concept of automatic differentiation; Parametric and non-parametric methods; Decision Trees - Classification and Regression Tea Break Ensemble methods - random forests, gradient boosting; Bias and Variance trade off; Cross validation; Hyper parameter search; Feature reduction - PCA Lunch Break Code and demonstration - Random forests, AdaBoost and GradientBoost, Bias and Variance, PCA, Feature selection and reduction Tea Break Unsupervised learning - clustering association and recommendation systems
    July 27, 2022 Introduction to Neural Networks, Back propagation, Weight initialization and update Tea Break Introduction to convolutional neural network, mathematical formulation, filter update process Lunch Break Convolutional Neural Networks for Computer Vision: Image Classification, Encoder-Decoder Networks for Image Segmentation, Object Detection using ConvNets, GANs Tea Break Code demonstration using Pytorch - Neural Networks
    July 28, 2022 Code demonstration using Pytorch - Convolutional Neural Networks Tea Break Introduction to reinforcement learning, Markov decision process, Policy evaluation, Optimal control, Bellman Equation, Value Iteration, Policy Iteration Lunch Break Reinforcement learning setting, Learning from sample trajectories, Q-learning, SARSA, TD-learning, Eligibility traces, Function approximation techniques, Deep learning, Solving tic-tac-toe game using RL Tea Break Facts and Myths about ML/DL capability; Deductive reasoning; The idea of vector transformations as fundamental to deep neural networks
    July 29, 2022 Industry Sessions on ML/DL Tea Break IIT Tirupati faculty projects in machine learning and demonstrations Lunch Break Hands-on production maintenance issue of an ML model & debugging Tea Break Quiz and discussion