You are cordially invited to attend the biweekly Brookhaven AI Mixer (BAM). BAM includes three short talks on AI research happening at BNL, followed by an open mixer over coffee and snacks for everyone to network and discuss all things AI. The first half hour will consist of presentations that will be available via ZOOM, and the second half hour will be for in person only networking.

Join us every other Tuesday at noon in CDSD's Training Room (building 725, 2nd floor) to learn about interesting AI methods and applications, engage with potential collaborators, prepare for pending FASST funding calls, and build a community of AI for Science at BNL.

Speakers

Kriti Chopra, Computing & Data Sciences (CDS)
Thomas Flynn, Computing & Data Sciences (CDS)
Wenjie Liao, Chemistry Division

Tuesday, January 7, 2025, 12:00 pm -- CDS, Bldg. 725, Training Room

Join ZoomGov Meeting: https://bnl.zoomgov.com/j/1615289117?pwd=Hqkbj9itxWrFnkhZ8rQXHPInO2gxdF.1

Meeting ID: 161 528 9117
Passcode: 991382

Abstract:
Coarse grained (CG) models alleviate the drawbacks of all-atom simulations. The latter still pose challenges because they are computationally expensive and give access to limited spatiotemporal scales, despite the use of modern high-performance computing clusters. CG models ignore some of the atomistic degrees of freedom, leading to fewer interatomic interactions, hence less computing time. Introducing such models emphasizes the need to properly manage these multiple scales, by carefully deriving potentials and reconstructing conformations from their CG representations, usually with the help of Machine Learning. Following a bottom-up and force matching approach, we train a Physics-Informed Neural Network to extract the CG force field parameters from all-atom simulation data. We verify our approach by applying it to fibrin monomers to study multiple-fibrin polymerization in solution at the microsecond scale, after modifying the force field to incorporate further non-bonded interactions, not present in the training data. Access to these scales will allow us to study the effects of some of the molecules' components. Furthermore, we modify recent solutions in data-driven protein backmapping. Taking advantage of the developments in graph neural networks and variational inference, we introduce an intermediate step in the all-atom reconstruction of a molecule given its CG configuration, in an attempt to more accurately de-coarsen structures whose atom-to-CG-beads ratio is very high. The combined effect of our new forward and inverse coarse graining methodology will enable the in silico study of many phenomena that are highly dynamic and intrinsically multiscale.

Bio:
Georgios Kementzidis is a third year PhD student in the Department of Applied Mathematics and Statistics at Stony Brook University. His advisor is Dr. Yuefan Deng. His research interests lie at the intersection of Computational Science, molecular dynamics (MD) simulations, and Machine Learning (ML) applications to Computational Biophysics. He is particularly interested in coarse-graining and multi-scale simulations.

*Note: this seminar will be held in-person (food provided on a first-come, first serve basis) and online*

Join Zoom Meeting https://stonybrook.zoom.us/j/99510099036?pwd=EyowuLBGvUVLZDBlG6F6chkMICFOZ7.1
Meeting ID: 995 1009 9036
Passcode: 132419
Recently, large-scale language data combined with modern machine learning techniques have shown strong value as means for studying human psychology and behavior. For example, language alone has been shown predictive in mental health, personality, and health behaviors. However, many applications for such language-based assessments have readily available and important data beyond language (i.e. extra-linguistics), such as predicting the subjective well-being of a community using tweets, where one can take into account their age, education, and demographic attributes. Language may capture some characteristics while extra-linguistic variables captures others. We believe that effectively integrating linguistic and extra-linguistic data can yield benefits beyond either independently. In this thesis, we develop methods which effectively integrate extra-linguistic data with language data focused primarily on social scientific applications. The central challenge is dealing with the size and heterogeneity of, often sparse and noisy, language data versus the, often low-dimensional and non-sparse, extra-linguistic variables. First, we consider structured extra-linguistics, like socioeconomic (income and education rates) and demographics (age, gender, etc.), and propose two integration methods, named residualized controls (RC) and residualized factor adaptation (RFA), to be used in county-wise prediction tasks. Demonstrating techniques that integrate information at both the model-level and data-level, we found consistently strong improvement over naively combining features, for example, increasing county level well-being predictions by over 12%. Next, we consider unstructured extra-linguistic data. In the first part, we incorporate social network connections and language over time to propose a novel metric for quantifying the stickiness of words - their ability to spread across friendship connections in a social network over time (or in other words, stick in ones vocabulary after seeing friends use it). We obtain which language features are more probable to disseminate through friendship and show such a metric is useful for predicting who will be friends and what content will spread. In addition, we analyze language content over time by proposing a novel dynamic content-specific topic modeling technique that can help to identify different sub-domains of a thematic scope and can be used to track societal shifts in concerns or views over time.

The University's Main Commencement Ceremony will take place on Friday, May 23, 2025 at 11 am at Kenneth P. LaValle Stadium. Gates open at 10 am.

All guests need a valid ticket to enter LaValle Stadium - no exceptions. Children age 1 and older require a ticket. Seating is first-come, first-served.

Register here.

Abstract: Recent progress in large language and vision models demonstrates how far we can go by scaling with vast internet-scale data. In contrast, physical AI, agents that perceive and act in the real world, still lags far behind. Today, both academia and industry primarily pursue generalizable physical AI by scaling up: collecting large-scale action-video datasets or training world models that enable interaction through learned environments. However, this paradigm is inherently inefficient and will soon reach a data ceiling. In this talk, I argue for a shift from scaling up to scaling out. I introduce reality world simulators, a new paradigm that converts real-world videos into diverse, interactive simulation environments. Instead of relying on more data collection, this approach expands data through structured reconstruction and recomposition, enabling both higher data efficiency and physically grounded interaction. I will present a three-pronged approach: 1) Scaling out via Digital Twins: reconstructing controllable, interactive environments from monocular videos to support diverse agent exploration. 2) Scaling out via Digital Cousins: disentangling scene structure into compositional elements to generate large-scale variations of real-world environments. 3) Scaling out via Embodied Humans: incorporating realistic human dynamics to improve safety and social compliance in robot learning. Finally, I will outline a roadmap toward building generalizable and safe physical AI systems for open-world deployment.

Bio: Dr. Wayne Wu is a postdoctoral researcher at UCLA Computer Science, working closely with Bolei Zhou, and collaborating with Trevor Darrell (UC Berkeley EECS) and Jiaqi Ma (UCLA CEE). He received his Ph.D. in Computer Science and Technology from Tsinghua University in June 2022 and was previously a visiting Ph.D. student at Nanyang Technological University. He also spent seven years in industry, where he led the research and development of products that reached more than 10 million end users worldwide. His research lies at the intersection of computer vision, robotics, and computer graphics. He focuses on developing infrastructure and methods to scale physical AI, enabling robots to work reliably and safely in the open world. He has published over 50 papers at top-tier venues including CVPR, ICCV, ICLR, NeurIPS, and ICRA, with over 9,500 citations and 10,000 GitHub stars. His work has received a CVPR Best Paper Candidate and multiple Oral, Spotlight, and Highlight presentations. He was also honored with the 2025 UCLA Chancellor's Award for Postdoctoral Research, recognizing the best postdocs at UCLA, and he was the only awardee from the School of Engineering. He serves as an Area Chair at CVPR 2026.

Location: NCS 120

Abstract: Pre-trained diffusion and flow matching models have made visual generation remarkably powerful, enabling high-fidelity synthesis of images and videos from natural language prompts. However, their behavior is still largely dictated by the pre-training data distribution and likelihood objective, which do not directly encode downstream desiderata such as fine-grained semantic alignment, controllability, or realism. This gap motivates post-training: starting from a base generator and further optimizing it with additional supervision signals derived from human or reward model preferences.This work presents post-training for visual generative models through two complementary case studies. First, Hummingbird addresses the problem of fine-grained contextual alignment in image-text-to-image generation. We introduce a multimodal context evaluator that scores the consistency between rich contextual descriptions and generated images, capturing fine-grained alignment beyond global CLIP similarity. By directly backpropagating these differentiable rewards through the diffusion sampler, Hummingbird substantially improves semantic faithfulness while preserving high visual quality.
Second, PISCES tackles post-training for text-to-video generation, where alignment is inherently semantic-spatio-temporal. We show that naive VLM-based rewards suffer from distributional mismatch and token-level misalignment, leading to reward hacking and suboptimal optimization. PISCES introduces a bi-objective, Optimal Transport (OT)-aligned reward module: distributional OT using Neural Optimal Transport to align text and video embedding distributions, and discrete, partial OT over a spatio-temporal cost matrix to capture semantic alignment at the token level. These rewards are integrated into both direct backpropagation and GRPO-style optimization to post-train state-of-the-art text-to-video generators. Together, Hummingbird and PISCES provide a unified view of how carefully designed visual reward models, coupled with OT-based representation alignment, can reliably improve the downstream behavior of pre-trained image and video generators.

Speaker: Minh Quan Le

Location: NCS 220

Zoom: https://stonybrook.zoom.us/j/94798224254?pwd=CFraer25qnpORbJ14aAVHRwaSJOjJM.1

Professor Petar M. Djuric, SUNY Distinguished Professor and Savitri Devi Bangaru Professor in Artificial Intelligence at Stony Brook University, has been selected as a plenary speaker at the upcoming 23rd IEEE Statistical Signal Processing Workshop (SSP 2025). The event will be held from June 8-11, 2025, in Edinburgh, Scotland, and is one of the premier international forums for the latest advances in statistical signal processing.

Professor Djuric's plenary talk, titled Quantifying causal relationships: Dynamic strengths, attributions, and confounders, will take place on June 10 from 9:00 AM to 10:00 AM EST. His presentation addresses foundational challenges in data-driven causality, proposing novel methodologies for quantifying causal strength in both static and dynamic systems, with special attention to latent confounders and attribution analysis.

This work has broad implications across disciplines including healthcare, economics, and climate science--areas where causal understanding drives critical decisions and innovations.

Professor Djuric has been a long-standing leader in the fields of machine learning and signal and information processing. After receiving his Ph.D. from the University of Rhode Island, he joined the faculty at Stony Brook University, where he served as Chair of the Department of Electrical and Computer Engineering from 2016 to 2023. He is also the founding Editor-in-Chief of the IEEE Transactions on Signal and Information Processing Over Networks and a Fellow of IEEE, EURASIP, AAIA, and AIIA.

Early bird registration for the workshop is open until April 30, 2025. For more information, visit the official SSP 2025 website.

Abstract: Many unresolved legal questions over LLMs and copyright center on memorization: whether specific training data have been encoded in the model's weights during training, and whether those memorized data can be extracted in the model's outputs. While many believe that LLMs do not memorize much of their training data, recent work shows that substantial amounts of copyrighted text can be extracted from open-weight models. However, it remains an open question if similar extraction is feasible for production LLMs, given the safety measures these systems implement. We investigate this question using a two-phase procedure: (1) an initial probe to test for extraction feasibility, which sometimes uses a Best-of-N (BoN) jailbreak, followed by (2) iterative continuation prompts to attempt to extract the book. We evaluate our procedure on four production LLMs -- Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, and Grok 3 -- and we measure extraction success with a score computed from a block-based approximation of longest common substring (nv-recall). With different per-LLM experimental configurations, we were able to extract varying amounts of text. For the Phase 1 probe, it was unnecessary to jailbreak Gemini 2.5 Pro and Grok 3 to extract text (e.g, nv-recall of 76.8% and 70.3%, respectively, for Harry Potter and the Sorcerer's Stone), while it was necessary for Claude 3.7 Sonnet and GPT-4.1. In some cases, jailbroken Claude 3.7 Sonnet outputs entire books near-verbatim (e.g., nv-recall=95.8%). GPT-4.1 requires significantly more BoN attempts (e.g., 20X), and eventually refuses to continue (e.g., nv-recall=4.0%). Taken together, our work highlights that, even with model- and system-level safeguards, extraction of (in-copyright) training data remains a risk for production LLMs.

Speaker: Xinyue

Location: CS2311