Stein Discrepancy ML Optimization for High Latency Systems

Inspiration

Modern computing systems today use massive compute power for stock trading and performance training(ai), etc. These lead to needs in power optimization for these distributed systems. I would like to use ML agents to run a probability distribution (using stein discrepancy and noise contrastive estimation) in order to maximize an equilibrium for performance and energy usage. Also aim to optimize latency in these configurations by running routing algorithms. In doing these tasks, I hope to allow the internal hardware like the SRAM memory cells to have significantly more efficient cores by testing, and wireshark to test network latency over higher uses.

What it does

This project tests ml agents to find optimal energy levels during a certain workload. This project also tests a workload to reduce its latency.

How we built it

At the core, this project uses ml agents to find and pinpoint areas in a system. I wrote the majority of this code in python. I wrote test cases for uses in network latency.

Challenges we ran into

Understanding different types of workloads and what happens to cpu cores under high energy. It got me thinking about certain questions for later. How are certain cores in a computer constrainted by latency? Which agents can find these cores to track these constraints?

Accomplishments that we're proud of

What we learned

What's next for Stein Discrepancy ML Optimization for High Latency Systems

Built With

• ai
• numpy
• python
• wireshark