Luca Scimeca

AI Research Scientist @ Sapient
Gen AI, Vision, Learning Representations, Bias, Scientific Discovery, Robotics

[Siddarth @siddarthv66]

Come check out our poster this Wednesday at 4:30pm @icmlconf !!
Happy to chat about diffusion, GFlowNets and RL!

[Siddarth @siddarthv66]

Is there a universal strategy to turn any generative model—GANs, VAEs, diffusion models, or flows—into a conditional sampler, or finetuned to optimize a reward function?
Yes! Outsourced Diffusion Sampling (ODS) accepted to @icmlconf , does exactly that!

[Luca Scimeca @ScimecaLuca]

Come check out our neurips poster today! We will be at West Ballroom #7101 from 4:30pm - 7:30pm.

Website: https://t.co/4GPZySbQiD

[Siddarth @siddarthv66]

Resharing this thread now that our paper has been accepted to #NeurIPS! We propose a new (off-policy) objective "Relative Trajectory Balance" to train a diffusion prior with a reward function to obtain the Bayesian posterior. We can do KL regularized RLHF similar to LLMs now!

[Siddarth @siddarthv66]

Can we perform unbiased bayesian posterior inference with a diffusion model prior? We propose Relative Trajectory Balance (RTB) which allows us to directly optimize for this posterior model. We apply this to several tasks in image, language and control!🧵https://t.co/h7elRgivcC

[Sky News Breaking @SkyNewsBreak]

Labour has called on Liz Truss to step down as Prime Minister

Follows three Conservative MPs demanding her resignation yesterday

For more on this and other news visit https://t.co/NEDMP2vmWu

[Seong Joon Oh @coallaoh]

[ICLR 2022] Which Shortcut Cues Will DNNs Choose? A Study from the Parameter-Space Perspective
https://t.co/pkZeWvs3sm
@ScimecaLuca @SanghyukChun @MichaelPoli6 @oodgnas

TL;DR Why would DNNs choose colour over shape as the cue for recognition? Answer below.

[Dmytro Mishkin 🇺🇦 @ducha_aiki]

Which Shortcut Cues Will DNNs Choose? A Study from the Parameter-Space Perspective

Luca Scimeca, @coallaoh, @SanghyukChun, @MichaelPoli6, Sangdoo Yun
tl;dr: provide CNN entangles features, check which it uses for prediction. Color & ethnicity rules.
https://t.co/9UiRVKT2NR

[Seong Joon Oh @coallaoh]

Neural estimation of dynamical systems =
data-driven parameter estimation + strong maths modelling
This is a fascinating topic, isn't it?

I'm glad to contribute to this topic with great talents @MichaelPoli6 @ScimecaLuca @SanghyukChun

https://t.co/aAChhEf4hF (NeurIPS 2021)
1/