@oswinso

Graduate Researcher with Chuchu Fan at MIT @mit_REALM. Bringing Guarantees to Safe Reinforcement Learning 🇭🇰

 Cambridge, Massachusetts
oswinso.xyz
Joined April 2013
  • Tweets51
  • Following112
  • Followers164
9Media
Neither @oswinso nor I could make it to ICLR this year, but our poster did make the trip 🤣✈️🇧🇷 Come by our poster today about #DAM (#Discrete #Adjoint #Matching)—a unifying AM for discrete generative models (e.g., dLLM) !! 🕒 3:15 PM – 5:45 PM 📍 Pavilion 3 P3 #1711
Guan-Horng Liu@guanhorng_liuFeb 17
Adjoint Matching works great for fine-tuning diffusion models with reward gradients. How about #AM for #diffusionLLMs with #nondifferentiable #rewards? Does "discrete adjoint" even exist ... and how? 🤔 📢 Introduce #DiscreteAdjointMatching (#DAM)—a unifying AM for discrete generative models, accepted to #ICLR2026 🇧🇷 Work done with my amazing intern @oswinso and @RickyTQChen, Brian, Chuchu 🙌 📰 arxiv.org/abs/2602.07132
4
18
2,749
Adjoint Matching works great for fine-tuning diffusion models with reward gradients. How about #AM for #diffusionLLMs with #nondifferentiable #rewards? Does "discrete adjoint" even exist ... and how? 🤔 📢 Introduce #DiscreteAdjointMatching (#DAM)—a unifying AM for discrete generative models, accepted to #ICLR2026 🇧🇷 Work done with my amazing intern @oswinso and @RickyTQChen, Brian, Chuchu 🙌 📰 arxiv.org/abs/2602.07132
1
11
1
94
8,292
At #NeurIPS from Dec 2 to Dec 7 in San Diego! Looking forward to catching up and meeting new friends. Excited to chat about safety for robotics, constraint satisfaction in RL, and (stochastic) optimal control. Feel free to DM me to grab coffee or have a chat!
5
360
Robots can plan, but rarely improvise. How do we move beyond pick-and-place to multi-object, improvisational manipulation without giving up completeness guarantees? We introduce Shortcut Learning for Abstract Planning (SLAP), a new method that uses reinforcement learning (RL) to discover shortcuts in the planning graphs induced by task and motion planning (TAMP) skill libraries. It is a plug-and-play module that can be trained on top of existing planners to speed up execution through learned shortcuts. (1/5)
1
22
6
70
20,282
Meet Casper👻, a friendly robot sidekick who shadows your day, decodes your intents on the fly, and lends a hand while you stay in control! Instead of passively receiving commands, what if a robot actively sense what you need in the background, and step in when confident? (1/n)
9
35
6
163
25,112
🏆 Huge congratulations to the #RSS2025 Award Winners! roboticsconference.org/progr…
1
10
1
78
6,898
#almostsure blog post: On the integral ∫I(W ≥ 0) dW This looks at the mentioned integral, which displays properties particular to stochastic integration and which may seem counter-intuitive. almostsuremath.com/2023/11/0…

On The Integral ∫I(W ≥ 0)dW

In this post I look at the integral Xt = ∫0t 1{W≥0} dW for standard Brownian motion W. This is a particularly interesting example of stochastic integration with connections to local times, option p…

almostsuremath.com
2
7
1
52
14,029
Suppose now that Xₜ is started from ε: Xₜ ≔ ε + ∫₀ᵗ 1{Wₛ >= 0} dWₛ Since Xₜ = ε + max(W_t,0) - ½ L(t) and L(t) strictly increases only when Wₜ=0, does there exist a time u such that Xᵤ ≥ Wᵤ AND Xᵤ < 0? xitter.cf/oswinso/status/1721265…
Oswin So@oswinso5 Nov 2023
More observations and questions on the following stochastic integral: Xₜ ≔ ∫₀ᵗ 1{Wₛ >= 0} dWₛ Numerically simulating this does confirm that E[Xₜ]=0 and Xₜ does go negative. What I did not expect, however, is the distribution of Xₜ to look the way it does.
2
1
15
4,250
Using intuition from the discrete case, "Xᵤ downcrosses 0 when Wᵤ also downcrosses 0", and so u exists. However, I have no idea whether this holds in the continuous limit... Numerical simulations show that u exists, but I feel like this is due to numerical error?
237
More observations and questions on the following stochastic integral: Xₜ ≔ ∫₀ᵗ 1{Wₛ >= 0} dWₛ Numerically simulating this does confirm that E[Xₜ]=0 and Xₜ does go negative. What I did not expect, however, is the distribution of Xₜ to look the way it does.
Oswin So@oswinso1 Nov 2023
Small question about Ito integrals: Consider Xₜ ≔ ∫₀ᵗ 1{Wₛ >= 0} dWₛ where Wₜ is a Brownian Motion and 1 is the indicator. Xₜ is a martingale, so E[Xₜ] = 0. I would think that Xₜ is non-negative, but that doesn't seem to be true?
5
6
2
89
37,495
Small question about Ito integrals: Consider Xₜ ≔ ∫₀ᵗ 1{Wₛ >= 0} dWₛ where Wₜ is a Brownian Motion and 1 is the indicator. Xₜ is a martingale, so E[Xₜ] = 0. I would think that Xₜ is non-negative, but that doesn't seem to be true?
3
3
1
32
43,272
I realized theres a typo: I mean to put 1{X_s>=0} instead of 1{W_s>=0}. That changes the question significantly though.
1
1
1
6
3,029
Momentum Schrödinger Bridge is a nice framework for multi-marginal distribution matching (e.g. population modeling) that overcomes the stiff trajectories induced by most pair-wise distribution matching methods. Fun project with @iamct_r @MoleiTaoMath & Evangelos 🌉🌝
Tianrong Chen 陈天荣@iamct_r12 Oct 2023
😀 #NeurIPS2023 Introducing our work #DMSB (arxiv.org/abs/2303.01751)! #DMSB is an extension of Schrödinger Bridge algorithm (arxiv.org/abs/2110.11291 and arxiv.org/abs/2106.01357) in phase space to tackle trajectory inference task!
7
1
35
5,396
New blog post: Yet Another ICML Award Fiasco The story of the @icmlconf 2023 Outstanding Paper Award to the D-Adaptation paper with worse results that the ones from 9 years ago Please share it to start a needed conversation on mistakenly granted awards parameterfree.com/2023/08/30…

Yet Another ICML Award Fiasco

EDIT 4/2/26I thought I would write here what happened after 3 years.Absolutely Nothing.The ICML board discussed my formal appeal, but never replied to it.Actually, it got worse. The authors of the …

parameterfree.com
16
97
36
469
186,216
What is variational optimization? Why can continuous dynamics help? Optimization is already a profound field, what can it bring in? Check out blog itsdynamical.github.io/artic… Comment/Retweet/Like will be deeply appreciated! 1/6

Variational Optimization, and How It Simplifies Manifold Optimization (Part I: the continuous side...

TL; DR Gradient Descent (GD) is one of the most popular optimization algorithms for machine learning, and momentum is often used to accelerate its convergence. In this blog, we will start with a...

itsdynamical.github.io
1
40
157
20,757
Epigraph form provides a stable way of solving constrained optimal control problems to find safe, stabilizing controllers. We investigate this in our work accepted to #RSS2023. Project: mit-realm.github.io/efppo Paper: arxiv.org/pdf/2305.14154.pdf MIT News: news.mit.edu/2023/safe-and-r…
2
2
11
569
more replies
We test EFPPO in simulation on challenging underactuated systems such as the "Top Gun: Maverick" inspired F16 fighter jet, and find up to ten-fold improvements in stability performance compared to baseline methods. (7/8)
1
190
I'm excited to be presenting this work in #RSS2023 Daegu at the Controls & Dynamics section that starts at 1:30 pm tomorrow. Hope to see you there! (8/8)
154
Load more