- Level: Intermediate - Choice - Specialisation
- Prerequisites:Introduction to Interaction AI or any introduction course.
- Understanding the nature, diversity, and process of working of algorithms that are used to train various intelligences from an Interaction AI Lens.
- Assessments: Micro Assessment weekly, Full Assessment last week. Midterm, and Completion.
Week 1- The current State of AI
Learning OutcomeLearning Outcome Students should be able to understand the various families of algorithms, their working, shortcomings, and new approaches in AI.
1.1 Algorithms leading to AI Brain Technology.
1.2 The edge of AI Brain Technology
1.3 Explainability of AI Brain Models, process, and working.
Practical Component
Practical demonstrations on different projects made with different algorithms and their edge for various tasks and use cases.
Week 2 -The process of training
Learning OutcomeLearning Outcome What does it mean to train an algorithm, and what are the key parameters and processes in training effectively.
2.1 What Happens During Training. Achieving the Weights.
2.2 Training Speed, Size, Algorithm Performance difference for various tasks.
2.3 Choosing the optimal structure for a given task.
Practical Component
Understand the training process and how various algorithms perform after training to be able to make keener selections of first-time working solutions for problems
Week 3 - Evaluating Training
Learning OutcomeLearning Outcome Has the training happened as expected, and achieved the outcome intelligence required. Here we delve deeper into understanding the outcomes.
3.1 Evaluating a trained model.
3.2 Could we have done better through another model, more training, or data.
3.3 Planning Test Case Selection, for various models, limitations, and methods
Practical Component
Practically view outcomes of well-assessed models, and the difference in predictions. Identify and understand potential flaws before starting.
Week 4- Fixing Problems that arise in process
Learning OutcomeLearning Outcome Depict understanding of common problems that arise through training, and how to detect and fix them.
4.1 Overfitting, low test accuracy.
4.2 Lower predictability, high variance in output.
4.3 Concept understanding failure
Practical Component
Showcase algorithms that have these flaws, and realize constructs to detect and evaluate them.
Week 5 - State of the Art Algorithms
Learning OutcomeLearning Outcome Which are the algorithms powering current high-end products, and what are their advantages, drawbacks, and opportunities.
5.1 Conversational AI Exemplary Models.
5.2 High-End and Tiny Structures for task-specific outcomes.
5.3 Non-Conventional Approaches for outcomes.
Practical Component
Here we practically assess, and examine various algorithms, understanding their working, drawbacks, and advantages to start critically thinking about best training practices.
Week 6 - Project and Assessment
Learning OutcomeLearning Outcome Here students work on a custom project of choice selected by the group to finish within a week from end to end with peer review, and complete presentation of outcomes.
6.1 Select a task, and appropriate algorithm with data to train.
6.2 Initial Self Assessment, and Peer Assessment and review of outcomes.
6.3 Final Presentations and Scoring of outcome achieved
Practical Component
Practical demonstrations of student-made models at various stages, with fixes, and live demonstrations to a committee of peers or prospective recruiters.