Vladimir Boza

Contact: <surname>1 (at) uniba.sk

Thesis topics

If you are interested in any of these or have your own topic in ML or data structure area, send me an email.

Matrix decomposition in neural networks

• Lot of operations in neural networks are just dot products of vector x and matrix W.
• Here we are looking for decomposition of matrix W as a product or sum of multiple matrices, such that computing the final matrix vector product is fast and accuracy of network is preserved.
• Examples include: Butterfly matrices or Monarch matrices or just a common low-rank decomposition
• We want to improve upon these ideas
• Expect a lot of programming here
• Prereqs: Python, ML/neural networks course