Videet Mehta

cs @ mit | prev @liquidai | research @MIT_CSAIL | @paretoholdings

[Videet Mehta @VideetM]

@cory Interested in joining

[Videet Mehta @VideetM]

@mikefutia CLAY

[Videet Mehta @VideetM]

@AlessandroLavis relief

[Videet Mehta @VideetM]

@spect3ral clay

[arXiv Sound @ArxivSound]

Videet Mehta, Liming Wang, Hilde Kuehne, Rogerio Feris, James R. Glass, M. Jehanzeb Mirza, "CALM: Class-Conditional Sparse Attention Vectors for Large Audio-Language Models," https://t.co/aHej47hWn1

[Videet Mehta @VideetM]

we achieved a 4x speedup from traditional diffusion policies and ~200% improvement in survival steps with extremely low number of function evaluations (~5) . also better guidance control.

[Videet Mehta @VideetM]

cool quick work we did in robotic manipulation class on using flow matching models for loco-manipulation. have a read: https://t.co/cKrEYDRyMU

[Videet Mehta @VideetM]

we teach our own fm policy to walk via prev work in co-diffusion where we diffuse across both states and actions together

[Videet Mehta @VideetM]

Wrote an in-depth blog of the GRPO vs GSPO objective functions including the gradient derivation for both

https://t.co/prwuWLnXF4

[Videet Mehta @VideetM]

Compressed Latent Reasoning: https://t.co/3CGANrNOM1

[Videet Mehta @VideetM]

been writing couple of blogs about various ai papers/concepts that I've encountered this summer

Flow Matching Models: https://t.co/DZFE0KGfBW

[Videet Mehta @VideetM]

Gaussian Mixture Models https://t.co/M480Awfgia