Fuzzy Systems: Crossing the barrier of 100000 Variables - Professor Nikhil Pal

Fuzzy Systems: Crossing the barrier of 100000 variables!

2024 UTS FEIT Research Excellence Lecture Series - Guest Lecture

Abstract

In this presentation, we discuss the limitations of large language models (LLMs), especially in areas requiring interpretability, explainability, and trustworthiness, such as healthcare and nuclear diagnostics. While LLMs have proven effective, fuzzy rule-based systems offer better interpretability and trust. However, fuzzy systems struggle with high-dimensional data due to numerical underflow. We propose a method to design fuzzy systems for high-dimensional datasets without dimensionality reduction, ensuring underflow is avoided. This approach, tested on datasets with over 100,000 dimensions for classification and regression tasks, aims to improve trustworthiness and explainability, though its full effectiveness remains under evaluation.

Speaker Biography

Nikhil R Pal, is a professor at Techno India University, Calcutta, and an Honorary Visiting Professor at South Asian University, New Delhi. He was previously a professor and the founding head of the Center for Artificial Intelligence and Machine Learning at the Indian Statistical Institute. His research focuses on brain science, computational intelligence, machine learning, and data mining. Professor Pal was Editor-in-Chief of IEEE Transactions on Fuzzy Systems (2005–2010) and has served on numerous journal editorial boards. He has received prestigious awards, including the IEEE CIS Fuzzy Systems Pioneer Award (2015) and Meritorious Service Award (2021). He has held leadership roles in IEEE CIS, including President (2018-2019), and is a Fellow of multiple esteemed academies, including IEEE, USA, and IFSA.