ONLINE GIAN COURSE | FAIRNESS IN MACHINE LEARNING
FAIRNESS IN MACHINE LEARNING
Organized by Department of Data Science, Indian Institute of Technology Palakkad
If you are interested, please fill this simple form to enroll yourself.
OVERVIEW
Machine Learning (ML) has become pervasive in today’s digital era. With massive amounts of data being accumulated electronically, machine learning systems that can identify and exploit patterns in digital databases have expanded their ambit from niche applications to almost every sphere of human activity. For example, data-driven ML algorithms, through personalized and pro-active recommendations, influence the news we consume, the route that we choose while driving, the movies we watch, and the consumables we buy. These algorithms work by identifying patterns and regularities on the historical data they learn over and use them in order to effectively address the task at hand. While supervised learning involves correlating data regularities against manually entered labels in the data that encode the gold-standard results, unsupervised learning focuses on identifying and leveraging patterns in the absence of labels. Digital devices such as mobile phones, traffic sensing devices, medical wearables, and computers are allowing the collection of digital data at an unforeseen rate, making it impossible for manual labeling efforts to keep pace with them. This data avalanche which will only intensify as time progresses increases the importance of unsupervised learning in the days to come. The underlying data and the labels – the raw material for ML algorithms – are derived from human activities and judgments. Thus, such data would inevitably embody biases, stereotypes and conventions that are rooted within the society they are sourced from. Allowing ML algorithms to exploit data in an `unregulated` manner would lead them to internalize such biases, stereotypes and conventions, in their decision-making, potentially making them behave in an unacceptable manner when viewed from the perspective of ethics. The increasing relevance of such questions within an increasingly data-driven era has led to much recent interest in the ethics of machine learning algorithms. Particular attention has been accorded to operational notions within the broader ethics umbrella, such as fairness and transparency in ML-based decision-making.
In this course, our focus will be on fairness in ML-based decision-making. This is a highly interdisciplinary field of inquiry, given that notions of fairness have their foundations in political philosophy, and fairness issues manifest in profiundly different ways across various application domains such as healthcare, education, smart cities, policing and immigration. This course will cover ML fairness issues broadly including theoretical foundations and case studes, and will particularly focus on fairness issues in unsupervised machine learning tasks – e.g., clustering, representation learning, retrieval and outlier detection - towards the latter half. Overall, the course will provide a well-rounded overview of the field of fairness in machine learning.
OBJECTIVE
The primary objectives of the course are as follows:
- Exposing participants to the exciting new field of fairness in machine learning (ML),
- Providing exposure to participants on a variety of fairness notions and their inter-relationships,
- Enable participants to identify and critique ML algorithms from the perspective of fairness,
- Enable participants to understand state-of-the-art computational methodologies to embed fairness-oriented constructs into ML algorithms, in particular, for unsupervised ML,
- Expose participants to methods of evaluating the fairness of ML algorithms based on their empirical performance, with a focus on unsupervised ML,
- Get participants to develop a well-rounded understanding of the discipline of fairness in ML, with knowledge of the state-of-the-art in the field.
COURSE TOPICS
- Machine Learning and Fairness
- Sources of Unfairness in Machine Learning Techniques
- Philosophical and Legal Underpinnings of Fairness
- Streams of Fairness: Individual and Group FairnessFairness Testing
- Costs of Fairness: The Accuracy vs. Fairness Trade-off
- Fairness in Classification and Fair Algorithms for Classification
- Fairness in Clustering and Representation Learning
- Fairness in NLP; Text Embeddings and Fairness
- Fairness in Similarity Search and Information Retrieval
- Fairness in Outlier and Anomaly Detection
- Research Gaps in ML Fairness
INSTRUCTED BY
Prof. Deepak Padmanabhan, Queen’s University Belfast, UK
Prof. Deepak P is an assistant professor of computer science at Queen’s University Belfast, UK. He has 15+ years of research experience in Machine Learning and Data Analytics, which includes 10 years as a Research Scientist at IBM Research. He has published 90+ research publications in various areas of data analytics, authored 3 books, and has 12 patents to his credit. He has won several awards including best paper awards and the INAE Young Engineer Award and is a senior member of the ACM and IEEE, as well as a fellow of the HEA (UK). His research interests are in ML applications and AI ethics.
Dr. Sahely Bhadra, IIT Palakkad, India
Dr. Sahely Bhadra works in the broad area of Machine Learning and Optimization. She is interested in multi-view kernel learning from noisy incomplete data with structural properties. Dr. Sahely has published in top venues in ML and bioinformatics such as ICML, JMLR, MLJ, ICDM, Bioinformatics.
| YOU SHOULD ATTEND IF YOU ARE |
|
|
DATE-TIME (Tentative) |
JANUARY 24 – FEBRUARY 4, 2022 3pm - 6pm (IST) |
| MODE |
The course will be completely online. Students need good internet and a computer with python at their end. The LInk will be sent to registered candidates. If you are interested, please fill this simple form to enroll yourself. |
| COURSE FEES AND REGISTRATION |
The participation fees for taking the course is as follows:
|
If you are interested, please fill this simple form to enroll yourself.
CONTACT
Dr. Sahely Bhadra (email can be found on her webpage)
