HOW DO YOU HAVE AN AI FORGET - RESPONSIBLY?

Typically, AIs are trained using data from corpuses on the Internet and/or data which has been labelled by humans. This leads to problems if the information is originally labelled incorrectly or the source on the Internet is incorrect. This is particularly pertinent if the original source of information turns out to be misinformation. Alternatively, a previous fact changes over time or just becomes out of date. So how do you get an AI to forget, responsibly?

This project will look at developing a framework to enable an AI to recognise when a previous piece of information is out of date or has been superseded by other information. This requires understanding how information relates to another piece of information in each context. The problem can alternatively be considered from the point of view of a machine learning model. That is the process needed is a process which moves the embeddings or in the case of a symbolic model the nodes of the decision path to a different embedding or path. Gaining a greater understanding of the nature of this problem would provide insights into how machine learning models arrive at the decisions that they do.

Skills required

A good understanding of existing computational and algorithmic techniques in Deep Learning
Strong coding skills - preferably including Java and Python
A sound foundation on using statistics in code.
A broad understanding of evolutionary computation techniques would be advantageous.

Background reading

Kevin P Murphy, “Probabilistic Machine Learning: An Introduction”, MIT Press, 2022 - Amazon, downloadable PDF draft

Ian Goodfellow et al., ”Deep Learning (Adaptive Computation and Machine Learning Series)”, MIT Press,2017 - Amazon, also available for download from https://www.deeplearningbook.org/

Saurabh Shintre, Kevin A. Roundy, Jasjeet Dhaliwal,”Making Machine Learning Forget”,Privacy Technologies and Policy, 2019, Volume 11498 Read Online

Zihao Cao, Jianzong Wang, Shijing Si, Zhangcheng Huang and Jing Xiao,” Machine Unlearning Method Based On Projection Residual”,2022 IEEE 9th International Conference on Data Science and Advanced Analytics (DSAA), 2022, p.1 Read Online

Heng Xu, Tianqing Zhu, Lefeng Zhang, Wanlei Zhou and Philip S. Yu, “Machine Unlearning: A Survey”, ACM Computing Surveys, 2023 Read Online