![Owain Brennan Profile](https://pbs.twimg.com/profile_images/1862439642319945728/6YbfyOZu_x96.jpg)
Owain Brennan
@BrennanOwain
Followers
347
Following
4K
Statuses
3K
🇬🇧🇺🇸25 | Solving Hard Trade Problems | Exited Founder (Flytta) - 99 Strength, 99 Fishing, 99 Fletching
Middlesbrough
Joined October 2021
For my Data Science MSc dissertation a few years ago I focused on the effect of synthetic data in model training, in my case computer vision as LLMs weren’t cool at that time. I looked at ship crash detection which there was a limited training set for. I did an experiment training a model with: 100% real 25% synthetic, 75% real 50% - 50% 25% real 75% synthetic, 100% synthetic I used a super early image generator model to generate the synthetic set. I never ended up publishing the paper as I was launching my data science lab Seer at the time and had fallen out of love with academia. My research and experiment found that the 75% real 25% synthetic data split led to the best classification accuracy. I really should have published that thing in hindsight. Should I dig through a drive and just publish the whole thing models and data? It will be stored somewhere.
0
0
0
RT @growing_daniel: If we’re friends and you have multiple kids I have a favorite. You can’t but I can. Also a least favorite
0
13
0