09I03-03-V04-00562

Statistical Approach to Monitor Quantisation in Neural Network Training


Project title:
Statistical Approach to Monitor Quantisation in Neural Network Training

Call code:
09I03-03-V04

Component:
9: More effective managing and strengthening research, development and innovation funding

Investment:
3. Excellent science

The code of the request for funds from the mechanism:
09I03-03-V04-00562

Requester identification (name, address, reggon):
Žilinská univerzita v Žiline, Univerzitná 8215/1 010 26 Žilina, IČO: 00397563

Principal investigator:
Ing. Roman Budjač, PhD.

Total amount of funds requested from the mechanism:
148 306,80 €

Maximal approved funds:
148 306,80 €

Start of project implementation:
1. 6. 2024

End of project implementation:
31. 5. 2026

Project goal:
The aim of the project is to bring new insights into neural network compression using statistical methods before and after the neural network compression process, and to implement these techniques into the neural network training process.

Project deliverables:
D1.1 Seminar (report)
D1.2 Results report
D1.3 Source code
D1.4 Publication at a scientific conference
D2.1 Seminar (report)
D2.2 Report of results
D3.1 Source code
D2.4 Publication at a scientific conference
D2.5 Interim report of project results
D3.1 Source code
D3.2 Seminar (report)
D3.3 Report of results
D4.1 Report of results
D4.2 Seminar (report)
D4.3 Results report
D4.4 Publication in a scientific journal
D4.5 Final report


Project milestones:
M1 Analysis and test of suitable statistical for weight distribution comparison
M2 Analysis of implementation weight distribution comparison into network training process.
M3 Implementing a new training strategy of neural network based on statistical method
M4 Proving the concept by Implementing statistical methods into the training process

Project activities:
Will be continuously updated

Project data (open science):
Will be continuously updated