Noisy distributed learning on noisy data: A unified mathematical framework for dealing with arbitrary noise

Project Description

The goal of this project is to develop a unified mathematical framework for dealing with arbitrary noise in a distributed network of autonomous AI systems. We shall consider arbitrary noise not just in datasets, but also in the model parameters of AI models. The scope of our project is inspired by fundamental technical challenges that must be resolved before personalized AI systems are able to interact and learn from one another (e.g. exchange AI model parameters), without any unintended failures due to noise. Specifically, we shall build a comprehensive theory for the trustworthiness of data and AI models, develop general methods for noise detection tiered by user-selected thresholds, and design robust systems for noisy distributed learning on noisy data that are computationally efficient. We shall deal with all types of noise, whether known or unknown, via a “universal” approach.

Research Technical Area

  • Heuristic search and optimization
  • Knowledge representation and reasoning
  • Machine learning

Benefits to the society

Imagine that in the near future, every individual has a personalized AI system that can learn directly from user-generated data, as well as interact and learn from other AI systems in an autonomous manner. Ideally, such a personalized AI system should be robust to inherently noisy real-world data. It should be able to distinguish information from misinformation. It should also be resilient to adversarial noise, especially when downloading external AI model parameters from other potentially compromised AI systems. Our project tackles all these technical challenges, so as to pave the way for such personalized AI systems, which would ultimately unlock an entire exciting new ecosystem in personalized AI.

Project’s Publications

Team’s Principal Investigator

Assistant Professor Ernest Chong Kai Fong

Singapore University of Technology and Design

Ernest Chong is an assistant professor in the Information Systems Technology and Design pillar at the Singapore University of Technology and Design (SUTD). He obtained his bachelor’s degree in Mathematics (summa cum laude) from Cornell University in 2009, and his Ph.D. in Mathematics from Cornell University in 2015. His current research deals with both Mathematics and Artificial Intelligence (AI). Within AI, his primary interests are theoretical and computational aspects of deep learning; and unsupervised and semi-supervised learning. He is also particularly interested in developing an algebraic theory for deep learning that incorporates methods from computational commutative algebra.

 

Recent Notable Awards

  • Eleanor Norton York Award, 2014
  • NRF-ISF Singapore-Israel joint research grant award, 2017

The Team

Co-Principal Investigator

Associate Professor Tony Quek, Singapore University of Technology and Design
Research Focus: Heuristic search and optimisation, Machine Learning, Edge computing, Network Intelligence

Collaborator

Assistant Professor Shaowei Lin, Singapore University of Technology and Design
Research Focus: Cognitive modeling and systems, knowledge representation and reasoning, machine learning