Building
Mouse Trajectory Synthesis
ML research into generating synthetic human mouse movements indistinguishable from real behavior.
In building the ecommerce bot, I realized that mouse movements were one variable that bot detection systems examine.
Karpathy had just released his Autoresearch framework and I was wondering about potential applications for my own work. I thought it'd be interesting to see if I, as a non-engineer with zero ML training whatsoever, could leverage Autoresearch to create a generative model that outperforms existing published research.
Needless to say, it was incredibly confusing and difficult. It's hard to optimize if I have no idea what I'm doing. And my GTX 4070 was insufficient as my laptop crashed and overheated numerous times.
For my personal use case, I deployed corpus replay into my own tools to achieve AUC 0.50. The generative model is something I've put on the backburner for now.
Karpathy had just released his Autoresearch framework and I was wondering about potential applications for my own work. I thought it'd be interesting to see if I, as a non-engineer with zero ML training whatsoever, could leverage Autoresearch to create a generative model that outperforms existing published research.
Needless to say, it was incredibly confusing and difficult. It's hard to optimize if I have no idea what I'm doing. And my GTX 4070 was insufficient as my laptop crashed and overheated numerous times.
For my personal use case, I deployed corpus replay into my own tools to achieve AUC 0.50. The generative model is something I've put on the backburner for now.