View Proposal

Proposer
Jamie Gabbay
Title
AI as approximation
Goal
Learn stuff about AI
Description
Simplifying greatly, modern AI is based on multiplying a long vector by a large dense matrix. This is fine, but the problem is that multiple behaviours are encoded in that large dense matrix, and matrix multiplication can be expensive. I propose to approximate the matrix by a more sparse version which approximates the desired output for a particular class of vectors. Essentially, we decompose the denser matrix as a sum of sparser "expert" matrices, where each expert is specialised in handling vectors of a particular form. I am open to suggestions on how to improve this; part of the project would be to figure out how much of this is on-point and useful, and adapt the project accordingly.
Resources
Background
Url
Difficulty Level
High
Ethical Approval
None
Number Of Students
0
Supervisor
Jamie Gabbay
Keywords
Degrees
Back to List