Kunal M.

AI researcher| Eng @usemesh_ai | Prev:

[Kunal M. @MundadaM]

@paulg The more we make abstract, the harder it becomes to be in control of the machine we wish to instruct.

[Kunal M. @MundadaM]

@biraj21_ @rachiitfr I would guess that this would be easy to implement.

[Kunal M. @MundadaM]

@skalskip92 This crash out is 100% valid!!

[Kunal M. @MundadaM]

@augmentaltech Please someone play OSU on it!!

[Kunal M. @MundadaM]

@AkshayIyer22 is this like Paxel productized?

[Kunal M. @MundadaM]

@AlessandroDuico Oh makes sense👍

[Kunal M. @MundadaM]

@AlessandroDuico I thought you can't get access to Claude on airplane wifi.

[Kunal M. @MundadaM]

@AlessandroDuico Software.

[Kunal M. @MundadaM]

@aman_shakesbeer what course is this?

[Kunal M. @MundadaM]

Come find us at the poster session!
📅 Wed, Jul 8, 2026 • 10:30 PM to 12:15 AM PDT
📍 Hall A

🌐 Project page: https://t.co/xZw4qk9pN7

Huge thanks to my coauthors Kai Fukazawa and @ImanSoltaniPhD for letting me be part of this project.

[Kunal M. @MundadaM]

Excited to share that our paper "RAMAC: Multimodal Risk-Aware Offline Reinforcement Learning and the Role of Behavior Regularization" has been accepted to ICML 2026!
What if a policy performs well on average but still produces rare catastrophic outcomes? 🧵

[Kunal M. @MundadaM]

Our work takes a closer look at a standard but often implicit ingredient in offline RL: behavior regularization as a mechanism for OOD control. Beyond behavior matching, our analysis shows it helps bound OOD action probability and stabilize lower tail optimization under shift.

[Kunal M. @MundadaM]

We introduce RAMAC, which uses lower tail return information to suppress risky behavior modes, while behavior regularization keeps the policy close to data supported behavior.

[Kunal M. @MundadaM]

At the same time, how can we control out of distribution (OOD) actions without sacrificing the multiple valid behaviors captured by expressive diffusion or flow policies in offline RL?

[Kunal M. @MundadaM]

@sammo3x @benjamin_bolte Coursera

[Kunal M. @MundadaM]

@sammo3x @benjamin_bolte Databricks

[Kunal M. @MundadaM]

@DataChaz this isn't recent, right?

[Kunal M. @MundadaM]

@iamsahaj_xyz Ohh 😮.. I have gotten red-pilled by Shift+Delete in other places too. So I just switch to vim mode

[Kunal M. @MundadaM]

@iamsahaj_xyz I am also annoyed by this.