Oswin So (@oswinso) profile picture

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

Cambridge, Massachusetts
Joined April 2013
Photo by Oswin So (@oswinso)Photo by Oswin So (@oswinso)Photo by Oswin So (@oswinso)Photo by Oswin So (@oswinso)Photo by Oswin So (@oswinso)Photo by Oswin So (@oswinso)Photo by Oswin So (@oswinso)Photo by Oswin So (@oswinso)Photo by Oswin So (@oswinso)
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
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
2,759
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
8,306
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
71
20,362
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
162
25,194
🏆 Huge congratulations to the #RSS2025 Award Winners! roboticsconference.org/progr…
1
10
1
78
6,901
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 🌉🌝
😀 #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
5,402
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…
16
97
36
469
186,243
An @MITAeroAstro team developed a machine-learning technique that can autonomously fly a plane through a difficult “stabilize-avoid” scenario while avoiding obstacles. Read more at MIT News news.mit.edu/2023/safe-and-r…
2
11
41
5,401
📣 Our #ICML2023 workshop on Learning, Control and Dynamical Systems are #calling for #papers! 🗓️ Submission Deadline: May 29, 2023 📝Author Guideline: frontiers4lcd.github.io/subm… Drop me messages or email [email protected] if you have any questions! #Frontiers4LCD
✨Call for Papers✨ We are now accepting submissions for our #Frontiers4LCD workshop at #ICML2023! Guidelines for the authors can be found here frontiers4lcd.github.io/subm… Submission deadline: May 29, 2023
1
6
31
9,690
Tomorrow I'll be in Poster Session 1 #NeurIPS2022 to present "Deep Generalized Schrödinger Bridge" with @iamct_r @oswinso. The work got selected as #Oral🙂. Drop by or DM if interested and let's have a chat 🙂! 📍Hall J, #528 🕚11/29 Tue, 11am - 1pm
"Deep Generalized Schrödinger Bridge" got #Oral in #NeurIPS2022, and we just release our codes in github.com/ghliu/DeepGSB! I'll give talks on SB in the IBM research seminar (mid Nov) and NeurIPS workshop on score-based methods (Dec 2nd). Many thanks for the kind invitations 🙂
2
7
I'll be at #NeurIPS2022 from Mon and give a talk in Score-based Method Workshop on Fri! Happy to discuss topics on Neural ODE/SDE/PDE, Schrodinger bridge, diffusion model, stochastic optimal control for ML applications! I'm also exploring Neural Operator🙂! DM if interested 📨
2
4
38
"Deep Generalized Schrödinger Bridge" got #Oral in #NeurIPS2022, and we just release our codes in github.com/ghliu/DeepGSB! I'll give talks on SB in the IBM research seminar (mid Nov) and NeurIPS workshop on score-based methods (Dec 2nd). Many thanks for the kind invitations 🙂
📢Presenting "Deep Generalized Schrödinger Bridge" (#NeurIPS2022) -- a new Diffusion Model for social science problem like population modeling & #opinion_depolarization. It solves Mean-Field Games up to 1000 dim (highest dim to my best knowledge). arxiv.org/abs/2209.09893 (1/n)
1
8
1
36
Deep Generalized Schr\"odinger Bridge. (arXiv:2209.09893v1 [stat.ML]) ift.tt/OL4dypB
8
17
Sorry everyone! There was a time zone configuration issue in the settings for the paper submission site causing it to close early. The submission deadline has now been pushed back to noon AoE on January 29 (a 12 hour extension)
1
4
1
6