This class will be offered in Spring 2021, in the Department of Computer Science, Tezpur University. We are designing the course upto the latest industry standards and hope to prepare our undergrad students to tackle the most challenging problems in the industry as ML Engineers.
You are eligible to take this class for credit, if you have taken these classes before:
- Programming-1, Programming-2
- Data Structures
- Algorithms
- Introduction to Machine Learning
- Engineering Mathematics-1, Engineering Mathematics-2
- Software Engineering and Networking.
Keeping in mind the complexity of the class, only B.Tech students in their 3rd year (6th semester +) and M.Tech students in their 2nd Year will be allowed to take the class for credit. Video lectures of the class will be made available in YouTube. Notes and assignments will be handled over Github for the active class. Please note that, candidates not taking the class for credit can still attempt the assignments with automated evaluation, but we will not be going over the other assignments which aren't evaluated automatically, for non-credit students.
- Learning theory, frequentist versus bayesian approach of learning
- Types of learning, Inductive, deductive, reinforcement learning
- Supervised and unsupervised learning and the common learning algorithm overviews and most common modern day usages
- Learning from data and decision making frameworks, how to extract signals from data, and use it in a decision system. Simple example and walkthrough.
- Fundamentals of what a lerning system does
- Example of an learning system
- A search engine : algorithm to implementation
- Netflix recommendation engine : competition, algorithm, implementation
- Assignment-1: (Term paper, no kidding!) General survey of learning systems, pick up 3 examples of learning systems of an app/product you use, and come up with how they built a learning system to extract information from a user pipeline and use it to offer best services to the user. What would be an ideal report? An article covering the business aspects as well as complex technical details explained in a way which should be understandable to a first year undergrad. Your report, has to be good as to publish it in WSJ, as well as something you can present to your CEO.
- Classification, Regression and Clustering : Theory, known and unknown applications
- How were traditional ML systems designed? What were the drawbacks faced? The age before deep-learning.
- ML tasks in the modern world, problems and common solutions.
- Classification methods : theory and implementation
- Regression methods : theory and implementation
- classification and regression model
- Training, Testing, Inference and examples
- Distributed file systems and distributed processing.
- Map-reduce: theory, and applications, implementation (assignment-2 prep)
- Classifier and Regressor, distributed implementation.
- (Assignment-2) Introduction to Hadoop. Setting up a HDFS server, basic access ops, PySpark and basics of Spark jobs in Scala.
- Common large scale product/business problems mapped to simple classification/regression objectives.
- What kind of problems do ML solve usually? Scenarios and use-cases.
- Common learning systems and their problem definitions
- Object detection and masking : self driving
- Question answering and dialogue generation : voice assistants
- Demand prediction and forecasting : rideshare
- How are the systems designed to tackle modern problems?
- The model
- Training, Testing, Inference
- Designing the ML inference system using task based preTrained models
- Example-2
- We'll design a tool for task-specific question answering using the open source tools we can find (transformers, server and client framework for chatbot, evaluation)
(this will be a class of 6 week indstructions. More content coming up)