June 10, 2025 – September 29, 2026

Microsoft Research New England Generative Modeling & Sampling Seminar

10:30 AM – 11:30 PM every Tuesday

위치: One Memorial Drive, Cambridge, MA, Room Clara Barton on the M floor

In recent years, generative models have made remarkable strides—not only in generating high-fidelity images, videos, and text, but also in powering chatbots and autonomous agents. Beyond these advances, the growing connections between generative modeling and broader problems in measure transport, sampling, and stochastic control have led to exciting developments in fine-tuning, inference-time alignment, and solutions to scientific simulation and inference tasks.

This seminar aims to bring together researchers from machine learning, statistics, applied mathematics, and the physical sciences to explore this rapidly evolving and interdisciplinary space—emphasizing both theoretical foundations and practical applications.

Nominate of self-nominate if you would like to give a talk: https://forms.gle/xkpKqwkPerTPvW7f6 (opens in new tab) 

Based in Microsoft Research New England (MSR NE), we welcome a diverse set of speakers and participants and aim to foster cross-disciplinary conversations and collaborations. The seminar takes place in the New England / Boston area with in-person participation from the local research community and virtual attendance from a global audience. We host speakers in person whenever possible. 

This seminar is currently organized by Carles Domingo-Enrich (opens in new tab) and Yuanqi Du (opens in new tab) at Microsoft Research New England, and first established by Carles Domingo-Enrich (opens in new tab), Dinghuai Zhang (opens in new tab), and Yuanqi Du (opens in new tab). 

Online attendance

Fill out this 1-minute form (opens in new tab) (if it doesn’t work try it on a different browser/device). The email you input will be added to the Google Group for the seminar, which means you will get Teams invites and emails about upcoming talks. It may take a few hours (up to a day) for you to get the welcome email.

In-person attendance

If you are a Microsoft New England employee, just show up. If you are an external guest, fill out this 1-minute form (opens in new tab) (if it doesn’t work try it on a different browser/device). Provide your name and ID to the building reception at the entrance, take the elevators to the M floor, and ask the receptionist at the M floor for a visitor tag and for Room Clara Barton. Bring a personal ID!

If you have any questions, send an email to carlesd@microsoft.com and yuanqidu@microsoft.com.

Teams link

Schedule

DateTimeSpeaker in-person?SpeakerTitle
June 23, 202610:30 AMYesYunyi Shen, MIT
June 16, 202610:30 AMYesYuanqi Du, Microsoft Research
Carles Domingo-Enrich, Microsoft Research		Rare Event Analysis via Stochastic Optimal Control
June 9, 202610:30 AMYesMichelle M. Li, Harvard University
June 2, 202610:30 AMYesMouyang Cheng, MIT
May 26, 202610:30 AMYesRunzhong Wang, MITDe novo Generation for Molecular Structure Elucidation from Mass Spectrometry
May 19, 202610:30 AMYesAimee Maurais, MITDesigning Dynamic Measure Transport for Sampling | Watch
May 12, 202610:30 AMYesLingkai Kong, Harvard UniversityGenerative AI for High-Stakes Decision-Making with Applications in One Health | Watch
May 5, 202610:30 AMNoLuca Ambrogioni​, Donders Institute for Brain, Cognition and BehaviourHow Out-of-Equilibrium Phase Transitions can Seed Pattern Formation in Trained Diffusion Models | Watch
April 28, 202610:30 AMYesEmma Finn, Harvard UniversityWhere the Score Lives: What Wavelets Reveal About Diffusion Models | Watch
April 21, 2026No talk (ICLR week)
April 14, 202610:30 AMYesMujin Kwun, Harvard University
Carles Domingo-Enrich, Microsoft Research		Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Models | Watch
April 9, 20264:00 PMYesYixuan Wang, CaltechData-Driven Discovery and Verification of Singularities in Nonlinear Partial Differential Equations | Watch
April 7, 202611:00 AMNoTristan Bereau, Heidelberg UniversityTractable Mapping Entropy and Generative Backmapping via Split-Flows | Watch
March 31, 202610:30 AMNoMichael Plainer, ELIZA Zuse School
Winfried Ripken, TU Berlin
Gregor Lied, TU Berlin		Generative Models for Molecular Dynamics Across Timescales | Watch
March 24, 202611:00 AMNoYinuo Ren, Stanford UniversityA Unified Approach to Analysis and Design of Denoising Markov Models | Watch
March 17, 202611:00 AMNoQiyang (Colin) Li, UC BerkeleyQ-learning with Flow-Matching Policies | Watch
March 10, 202610:30 AMNoJiequn Han, Flatiron InstituteGenerative Modeling without Clean Data: Self-Consistent Transport under Black-Box Corruptions | Watch
March 3, 202610:30 AMYesDavid Layden, IBM ResearchWavefunction Flows: Efficient Quantum Simulation of Continuous Flow Models | Watch
February 24, 202610:30 AMYesLorenz Richter, dida, Zuse Institute BerlinA non-Markovian approach to diffusion-based sampling | Watch
February 17, 202610:30 AMYesAram-Alexandre Pooladian, Yale UniversityBlind denoising diffusion models and the blessings of dimensionality | Watch
February 10, 2026No talk (break)
February 3, 202610:30 AMYesPeter Potaptchik, University of Oxford, Harvard UniversityMeta Flow Maps | Watch
January 27, 2026No talk (ICML deadline)
January 20, 202610:30 AMNoLuca Eyring, Technical University of MunichReward fine-tuning of Few-step Diffusion Models with Noise Hypernetworks
January 13, 202610:30 AMYesAayush Karan, Harvard UniversityReasoning with Sampling: Your Base Model is Smarter Than You Think
January 6, 202611:00 AMNoFan Chen, MITThe Coverage Principle: How Pre-Training Enables Post-Training
December 30, 2025No talk (winter break)
December 23, 2025No talk (winter break)
December 16, 2025No talk (winter break)
December 9, 202512:00 PMYesRunqian Wang, MITEquilibrium Matching: Generative Modeling with Implicit Energy-Based Models
December 2, 2025No talk (NeurIPS)
November 25, 202510:00 AMYesPeter Holderrieth, MITGLASS Flows: Transition Sampling for Alignment of Flow and Diffusion Models
November 18, 202511:00 AMNoJorge L. Rosa-Raices, UC BerkeleyNonadiabatic flow matching for free-energy estimation in and out of equilibrium
November 11, 202511:00 AMYesPeter Potaptchik, University of Oxford, Harvard UniversityTilt Matching for Scalable Sampling and Fine-Tuning
November 4, 202511:00 AMYesAlex Berlaga, University of ChicagoTargeting Low-Energy Protein Ensembles with Adjoint Matching
October 28, 202511:00 AMNoDenis Blessing, KarlsruheSampling with trust region constraints
October 21, 202511:00 AMYesChin-Wei Huang, Microsoft ResearchAccurate and scalable density functional with deep learning
October 14, 202511:00 AMYesJaeyeon Kim, Harvard University
Brian Lee Cheuk-Kit, Harvard University		Any-Order, Any-Length, Any-Time: Extending Masked Diffusion Models with Flexibility and Self-Correction
October 7, 202511:00 AMNoAdam Block, Columbia UniversityEMA Without the Lag: Stabilizing Optimization for Behavior Cloning and Language Models
September 30, 202511:00 AMYesAva Amini, Kevin K. Yang, Microsoft ResearchThe Dayhoff Atlas: scaling sequence diversity for improved protein generation
September 23, 2025No talk (ICLR deadline)
September 16, 202511:00 AMYesSonglin Yang, MITToward More Expressive yet Scalable RNNs: DeltaNet and Its Variants
September 9, 202511:00 AMYesZongyi Li, MITNeural Operator for Scientific Computing
September 2, 202511:00 AMYesKwangjun Ahn, Microsoft ResearchDion: The distributed orthonormal update revolution is here
August 26, 202511:00 AMYesZhengyang Geng, Carnegie Mellon UniversityMean Flows for One-Step Generative Modeling
August 19, 202511:00 AMNoJiajun He, University of Cambridge
Yuanqi Du, Cornell, Microsoft Research
Francisco Vargas, University of Cambridge, Xaira		Fantastic Path RND and Where to Find Them
August 12, 202511:00 AMYesLuhuan Wu, FlatironA training-free diffusion-based SMC sampler for unnormalized distributions
August 5, 202511:00 AMYesRaghav Singhal, NYU, Microsoft ResearchInference-Time Steering of diffusion models
July 29, 202511:00 AMYesSitan Chen, Harvard UniversityWhat does guidance do?
July 22, 202511:00 AMNoGuan-Horng Liu, Meta FAIRSampling with Schrödinger Bridge — An Adjoint-Matching Perspective
July 15, 202511:00 AMYesTianhong Li, MITBroadening the Scope of Autoregressive Models in Computer Vision and Beyond
July 8, 202511:00 AMNoJiaxin Shi, Google DeepmindDiscrete Generative Modeling with Masked Diffusions
July 1, 202511:00 AMYesYilun Du, Harvard UniversityInference Time Reasoning with Generative Models
June 24, 202511:00 AMYesMichael Albergo, Harvard UniversityGenerative Flow Maps via Self-distillation
June 17, 202511:00 AMNoKirill Neklyudov, MILASolving Many-Body Schrödinger Equation from the Probabilistic Perspective
June 10, 202511:00 AMYesMarta Skreta, University of Toronto, Microsoft ResearchControlling Diffusion Models at Inference Time