Francesco Sacco (@francescosacco1) 's Twitter Profile
Francesco Sacco

@francescosacco1

Physicist working as ML reasercher

ID: 355585692

calendar_today15-08-2011 15:51:13

60 Tweet

222 Followers

166 Following

Francesco Sacco (@francescosacco1) 's Twitter Profile Photo

Now it's the right time to go back to youtube for content Hundreds of high quality math videos are being published for the Summer of Math Eposition

Now it's the right time to go back to youtube for content

Hundreds of high quality math videos are being published for the Summer of Math Eposition
Francesco Sacco (@francescosacco1) 's Twitter Profile Photo

This few days I've been going back to just trying to do video diffusion in pixel space. No VAE. Just Diffusion. Either it works or not. Right? So, on simple environment like lunar lander, it does work, but on harder environments like counterstrike I can't make it to work

This few days I've been going back to just trying to do video diffusion in pixel space. No VAE. Just Diffusion.

Either it works or not. Right?

So, on simple environment like lunar lander, it does work, but on harder environments like counterstrike I can't make it to work
Francesco Sacco (@francescosacco1) 's Twitter Profile Photo

Lately I've been reading some papers about off-policy RL methods. They don't really work well. Today I sketched some math equations and I think that people have been optimizing the Bellman equation wrong this whole time. It's time for a new weekend research project.

Lately I've been reading some papers about off-policy RL methods. They don't really work well.

Today I sketched some math equations and I think that people have been optimizing the Bellman equation wrong this whole time.

It's time for a new weekend research project.
Francesco Sacco (@francescosacco1) 's Twitter Profile Photo

In Q-learning people usually predict just the expected Q-value. So I though: Is it possible to predict the entire probability distribution of Q? The answer is yes. Stay tuned-- tomorrow I'll publish the repository with everything x.com/FrancescoSacco…

In Q-learning people usually predict just the expected Q-value.

So I though: Is it possible to predict the entire probability distribution of Q? The answer is yes. 

Stay tuned-- tomorrow I'll publish the repository with everything

x.com/FrancescoSacco…
Francesco Sacco (@francescosacco1) 's Twitter Profile Photo

One of the side projects I'm working on is a B2B SaaS for health professionals. It's an AI medical scribe and it's way cheaper and simpler than the competition: noteician.com How do you guys get leads in this field?

Francesco Sacco (@francescosacco1) 's Twitter Profile Photo

Since my project that took me a weekend to do is popping off here on twitter, you might want to take a look at the project that took me 6 months to do x.com/FrancescoSacco…