Abstract: The remarkable success of large foundational models, such as LLMs and diffusion models, is built on their learning over vast amounts of static data from the Internet. However, human learning and problem-solving are fundamentally interactive processes--humans learn by engaging with their environment, tools, search engine, and feedback loops, iteratively refining their understanding and decisions. This gap between the interactivity of human learning and the static nature of model training raises a critical question: how can we imbue foundational models with the capacity for meaningful interaction?

In this talk, I will explore methods to enhance foundational models by incorporating interaction with the external environment. I will discuss strategies such as leveraging external tools, compilers, function calls to provide dynamic feedback to enhance foundation models. By drawing inspiration from human's interactive learning processes, I demonstrate how interaction-driven learning can lead to models that are not only more accurate but also more adaptable to real-world applications.

This work bridges the gap between static training paradigms and the dynamic, iterative nature of human intelligence, paving the way for a new generation of interactive AI systems.

Bio: Wenhu Chen has been an assistant professor at the Computer Science Department in University of Waterloo and Vector Institute since 2022. He obtained the Canada CIFAR AI Chair Award in 2022 and CIFAR Catalyst Award in 2024. He has worked for Google Deepmind as a part-time research scientist since 2021. Before that, he obtained his PhD from the University of California, Santa Barbara under the supervision of William Wang and Xifeng Yan. His research interest lies in natural language processing, deep learning and multimodal learning. He aims to design models to handle complex reasoning scenarios like math problem-solving, structure knowledge grounding, etc. He is also interested in building more powerful multimodal models to bridge different modalities. He received the Area Chair Award in AACL 2023, the Best Paper Honorable Mention in WACV 2021, the Best Paper Finalist in CVPR 2024, and the UCSB CS Outstanding Dissertation Award in 2021.
AI is everywhere and so are the privacy concerns that come with it. At its core, the most common forms of AI we use today are online digital services and thus inherit the usual privacy risks. We'll take a look at indirect prompt injection- a technique that can trick AI tools into revealing or extracting private information as well as techniques being used in academic contexts to manipulate systems and even mislead researchers.

Register here for the online session.

Abstract:
Deep learning models have achieved remarkable success across a wide range of computer vision tasks, including image classification, semantic segmentation, etc. However, such success highly relies on a large amount of annotated data, which are expensive to obtain. Moreover, their performance often degrades when there exist distribution shifts between training and test data. Domain Adaptation overcomes these issues by transferring knowledge from a label-rich source domain to a related but different target domain. Despite its popularity, domain adaptation is still a challenging task, especially when the data distribution shifts are severe, while the target domain has no or few labeled data.

In this thesis, I develop four efficient domain adaptation approaches to improve model performance on the target domain. Firstly, inspired by the large-scale pretraining of Vision Transformers, I explore Transformer-based domain adaptation for stronger feature representation and design a safe training mechanism to avoid model collapse in the situation of a large domain gap. Secondly, I observe that source models have low confidences on the target data. To address this, I focus on the penultimate activations of target data and propose an adversarial training strategy to enhance model prediction confidences. Thirdly, I study using weak supervision from prior knowledge about target domain label distribution. A novel Knowledge-guided Unsupervised Domain Adaptation paradigm is devised, and a plug-in module is designed to rectify pseudo labels. Lastly, I step into the task of Active Domain Adaptation, where the labels of a small portion of target data can be inquired. I propose a novel active selection criterion based on the local context and devise a progressive augmentation module to better utilize queried target data. The robustness of domain adaptation approaches, in addition to accuracy, is critical yet under-explored. To conclude the thesis, I empirically study set prediction in domain adaptation using the tool of conformal prediction and conformal training.


Location: New Computer Science Bldg., Room 120
Zoom Link: https://stonybrook.zoom.us/j/92736258273?pwd=ipDdh1CTG6dRYmqa3ltUvooei8OfaT.1Meeting ID: 927 3625 8273
Passcode: 466399
The Future of Learning: Rethinking Practice in a Changing World

Thursday, March 26, 2026 (Workshops)
Friday, March 27, 2026 (Symposium)

Open to Stony Brook University Faculty, Staff, and Graduate Students. Hosted by the Center for Excellence in Learning and Teaching, Office of the Provost.

Thursday, March 26, 2026
Workshop: AI Tools and Techniques
  • Open to all faculty & staff
  • Hands-on, exploratory
  • Registration only limited to the size of the room
  • Location: In-person, TBD
  • Time: 10 AM - 12 PM
  • Registration required

Friday, March 27, 2026
Keynote: Teaching and Thinking with AI
  • Faculty, TAs, postdocs, and academic staff
  • In-person on-campus conference venue
  • Location: SAC Balroom
  • Time: 9 AM - 3 PM
  • Registration required

Keynote Speaker: José Antonio Bowen

José Antonio Bowen has been leading innovation and change for over 40 years at Stanford, Georgetown and the University of Southampton (UK), as a dean at Miami University and SMU and as President of Goucher College. Bowen has worked as a musician with Stan Getz, Dave Brubeck, and many others and his symphony was nominated for the Pulitzer Prize in Music (1985).
Bowen holds four degrees from Stanford and has written over 100 scholarly articles and books, including the Cambridge Companion to Conducting (2003), Teaching Naked (2012 and the winner of the Ness Award for Best Book on Higher Education), Teaching Naked Techniques with C. Edward Watson (2017) and Teaching Change: How to Develop Independent Thinkers using Relationships, Resilience and Reflection (Johns Hopkins University Press, 2021).
Bowen has appeared in The New York Times, Forbes, The Wall Street Journal, and has three TED talks. Stanford honored him as a Distinguished Alumni Scholar (2010) and he has presented keynotes and workshops at more than 300 campuses and conferences 46 states and 17 countries around the world. In 2018, he was awarded the Ernest L. Boyer Award (for significant contributions to American higher education). He is a senior fellow for the American Association of Colleges and Universities.

Register here.
Title: Sustainable NLP

Time: Friday 4/29, 2:40 PM

Location: NCS 120

Abstract:


Natural language processing (NLP) technology has supercharged many real-world applications ranging from intelligent personal assistants (like Alexa, Siri, and Google Assistant) to commercial search engines such as Google and Bing. But current NLP applications use extremely large neural models, making them (i) expensive to deploy on servers, requiring large amounts of compute resources and power, and (ii) impossible to run on mobile devices, making on-device, privacy-preserving applications impractical.

In the first part of the talk, I will describe systems optimizations we have developed that significantly reduce the compute and memory requirement of NLP models. The optimizations we developed can be applied broadly and results in over 10x reduction in latency when deployed on mobile devices. In the second part of the talk, I will describe our recent work on predicting energy consumption of NLP models. Existing energy prediction approaches are not accurate, making it difficult for developers and practitioners to reason about their models in terms of power. We use a multi-level regression approach that produces highly accurate and interpretable energy predictions.



Bio:
Aruna Balasubramanian is an Associate Professor at Stony Brook University. She received her Ph.D from the University of Massachusetts Amherst, where her dissertation won the UMass outstanding dissertation award and was the SIGCOMM dissertation award runner up. She works in the area of networked systems. Her current work consists of two threads: (1) significantly improving Quality of Experience of Internet applications, and (2) improving the usability, accessibility, and privacy of mobile systems. She is the recipient of the SIGMobile Rockstar award, a Ubicomp best paper award, a Computing Innovation Fellowship, a VMWare Early Career award, several Google research awards, an
University Libraries Presents:
Join librarian Christine Fena for an interactive workshop that invites you to explore AI tools first hand, not just as users, but as critical investigators.
Through playful experimentation and collaborative discovery, you'll uncover inherent biases, probe algorithmic flaws, and gain a deeper understanding of AI's limitations and societal impacts.

RSVP on SBEngaged

Location: Melville Library, Central Reading Room, Lab B

Abstract: Traditional questionnaires remain the primary method for assessing psychological outcomes and beliefs, capturing individuals' and populations' inner states. This dissertation presents an alternative computational method that overcomes key limitations in current mental health monitoring, particularly in spatiotemporal resolution, responses to major events, and automatic belief identification. By analyzing ∼1 billion Tweets from 2 million geo-located users, we created a big data pipeline for estimating depression and anxiety at the county-week level. These Language-Based Mental Health Assessments (LBMHA) demonstrated higher reliability and validity than traditional survey measures. Our approach effectively captured mental health trends and highlighted significant increases in mental illness following major events. Using the LBMHA pipeline, we conducted quasi-experiments, research designs that simulate randomized control trials, to generate explanations for mental health changes due to COVID-19 incidence/death. Utilizing these time-series analyses, we conducted discontinuity forecasting for community-specific anxiety shifts using statistical learning via ensemble and contextual models. To likewise investigate individual internal states, we created a novel task and annotated dataset for self belief language identification. Our fine-tuned language model for self-belief classification, despite its relatively small scale, outperformed GPT-4o. The self belief topics identified by our model successfully predicted depression, anxiety, and stress, offering insights into the relationship between self-conceptualization and mental health. The adoption of scalable language-based assessments with modern distributed computation presents a promising avenue for advancing community and individual mental health research.

Speaker: Siddharth Mangalik

https://stonybrook.zoom.us/j/91251321639?pwd=faggV5jZ7ByFDCFmnLXD3HiYxjQ1Eb.1&jst=2

Abstract: Computational pathology has revolutionized cancer diagnosis and research through the analysis of digitized whole slide images (WSIs). However, the giga-pixel size of these images presents profound technical challenges, creating two intertwined bottlenecks: computational inefficiency and label inefficiency. The immense data scale makes standard end-to-end (E2E) training of deep neural networks infeasible due to prohibitive GPU memory requirements, while the reliance on expert pathologists for annotations makes obtaining high-quality labeled data a tedious and expensive process. This proposal confronts these dual challenges by developing a series of novel model architectures, training paradigms, and self-supervised learning methods designed to create a more efficient and effective framework for WSI analysis.

To improve computational efficiency, this proposal first introduces a locally supervised learning paradigm that enables E2E training on entire WSIs by partitioning a network into gradient-isolated modules, circumventing the memory bottleneck of backpropagation. Second, it presents Prompt-MIL, a parameter-efficient fine-tuning framework that reduces the number of trainable parameters, memory consumption, and training time by fine-tuning only few prompts to guide large pre-trained models. Third, this work advances the efficient architecture on WSIs by developing novel State-Space Models (SSMs). It proposes 2DMamba, the first intrinsic Mamba architecture that preserves the crucial 2D spatial structure of images, overcoming the spatial discrepancy inherent in 1D models. Fourth, to address the inefficiency of multi-directional scans in Mamba models, including 2DMamba, it presents Locally Bi-directional Mamba (LBMamba), which introduces a novel, hardware-aware local backward scan that integrates bi-directional scan into a single forward pass, significantly improving throughput performance trade-off. Lastly, it proposes an extension to the LBMamba, warp-level Bi-directional Mamba (WLBMamba) that extends the thread-level bidirectional scan to warp-level bidirectional scan that further improves the throughput performance trade-off.

To improve label efficiency, this proposal proposes a Precise Location-based Matching strategy for self-supervised dense contrastive learning. By allowing a local patch in one augmented view to match multiple overlapping patches in another, creates a more accurate correspondence, leading to superior feature representations for dense prediction tasks like segmentation and detection.

In summary, this proposal presents a holistic investigation into the efficiency bottlenecks in computational pathology. Through these combined contributions in model architecture, training paradigms, and self-supervised learning, this work establishes a more scalable, efficient, and powerful computational framework for analyzing giga-pixel pathology images.

Speaker: Jingwei Zhang

Location: Old Computer Science Room 2114

Zoom: https://stonybrook.zoom.us/j/95187903649?pwd=tV0CNxLu1QKqw7hGmcE1h0rJ2C6n1b.1
Meeting ID: 951 8790 3649 | Passcode: 488916