BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Events - ECPv6.18.4.1//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Events
X-ORIGINAL-URL:https://events.ucsc.edu
X-WR-CALDESC:Events for Events
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:20250309T100000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:20251102T090000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:20260308T100000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:20261101T090000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:20270314T100000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:20271107T090000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:20280312T100000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:20281105T090000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261123T080000
DTEND;TZID=America/Los_Angeles:20261123T090000
DTSTAMP:20261119T173641Z
CREATED:20261013T161821Z
LAST-MODIFIED:20261119T173641Z
UID:10017791-1795420800-1795424400@events.ucsc.edu
SUMMARY:Bondar\, G. (AM) - Geometric and Graph Structured Optimal Transport for Learning and Control
DESCRIPTION:We motivate\, formulate\, and solve several structured optimal transportation (OT) problems for learning and control. The field of OT had its inception in 1781\, when Gaspard Monge sought the optimal map to transport mass from a given initial distribution to a desired terminal distribution. Both OT and its entropy-regularized variant\, the Schrödinger bridge problem (SBP)\, have catalyzed significant developments across mathematics. More recently\, theoretical and algorithmic developments in OT and SBP have found myriad applications in stochastic optimal control and generative AI. \nThe connecting theme across the problems studied in this dissertation is the presence of structure which can be exploited in their solution — for our work\, this structure arises from geometry underlying a solution space or from a graph capturing a correlation structure between distributional data. Our original contributions begin with automated verification and discovery of OT regularity\, followed by stochastic learning and control for resource allocation and scheduling in multi-core real-time systems\, where the problem has a natural graph structure. We also formulate and solve the problem of finding an optimal graph structure in a distribution-valued dataset where none is present. \nWhile graph-structured OT is a natural fit for data-driven stochastic learning problems\, the knowledge of system dynamics can impose geometric structure to the problem. In this dissertation\, we establish the equivalence between the stochastic optimal control and path space formulations of SBP over a compact connected Lie group. Additionally\, we develop\, analyze\, and validate a novel numerical solution scheme for the SBP with control-affine dynamics. Finally\, we investigate the application of Gromov-Wasserstein (GW) OT — a variant of the problem allowing for transport between incomparable spaces — for spatial proteomics. We show throughout our work how careful considerations of the underlying geometry and graph-structures can enable bespoke solutions for practical learning and control problems through optimal transport. \nEvent Host: Georgiy Bondar\, Ph.D. Candidate\, Applied Mathematics  \nAdvisor: Abhishek Halder \nZoom: https://ucsc.zoom.us/j/96361886043?pwd=U73TbafhzI7kgS7MYAFko3YEPVIxTb.1 \nPasscode: 564490
URL:https://events.ucsc.edu/event/bondar-g-am-geometric-and-graph-structured-optimal-transport-for-learning-and-control/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-1.jpg
GEO:37.0009723;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Engineering 2 Engineering 2 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Engineering 2 1156 High Street:geo:-122.0632371,37.0009723
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261123T104000
DTEND;TZID=America/Los_Angeles:20261123T114500
DTSTAMP:20261116T212258Z
CREATED:20261116T212258Z
LAST-MODIFIED:20261116T212258Z
UID:10018115-1795430400-1795434300@events.ucsc.edu
SUMMARY:ECE 290 Seminar: "Advancing Medical Imaging Through Artificial Intelligence and Deep Learning"
DESCRIPTION:Presented by: Praveen Bharathi\, Assistant Project Scientist @ Radiological Instrumentation Lab\, UCSC \nDescription: “Artificial intelligence (AI) and deep learning are transforming medical imaging by enabling more accurate image analysis\, efficient reconstruction\, and automated clinical decision support. This seminar will explore the application of AI-driven computational methods across medical imaging modalities\, including magnetic resonance imaging (MRI)\, positron emission tomography (PET)\, and other biomedical imaging techniques. Drawing from research experiences in medical image segmentation\, deep learning-based image enhancement\, real-time image quality assessment\, and PET image reconstruction\, the presentation will highlight how machine learning can address challenges in image quality\, computational efficiency\, and quantitative analysis.” \nBio: Dr. Praveen Gurunath Bharathi is an Assistant Project Scientist in Radiological Instrumentation Lab headed by Prof. Shiva Abbaszadeh at the University of California\, Santa Cruz\, specializing in artificial intelligence\, machine learning\, and computational medical imaging. He earned his Ph.D. in Electrical and Electronics Engineering from BITS Pilani\, India. \nHis research experience spans medical image analysis\, deep learning\, computer vision\, and AI-driven image reconstruction. He has previously conducted research at Stanford Medicine and the University of Manchester\, working on advanced medical imaging and intelligent computational methods. His current research focuses on AI-assisted PET imaging\, image reconstruction\, and developing computational techniques to improve medical imaging performance. \nHosted by: Professor Zhicong Yu\, Electrical & Computer Engineering Department \nWhen: Monday\, November 23rd from 10:40 to 11:45 AM PST \nLocation: Classroom E2-194
URL:https://events.ucsc.edu/event/ece-290-seminar-advancing-medical-imaging-through-artificial-intelligence-and-deep-learning/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Seminars
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/11/20231207211-scaled.jpg
GEO:37.0009723;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Engineering 2 Engineering 2 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Engineering 2 1156 High Street:geo:-122.0632371,37.0009723
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261123T160000
DTEND;TZID=America/Los_Angeles:20261123T170000
DTSTAMP:20261007T203353Z
CREATED:20261007T203353Z
LAST-MODIFIED:20261007T203353Z
UID:10017756-1795449600-1795453200@events.ucsc.edu
SUMMARY:Statistics Seminar: Randomization Posteriors and Generalized Bayesian Inference
DESCRIPTION:Presenter: Edric Tam\, Warren Alpert Fellow in AI and Computational Biology\, Department of Biomedical Data Science\, Stanford University \nDescription: A longstanding question in Bayesian analysis concerns the role of randomization. We introduce randomization posteriors\, a class of generalized Bayesian updates obtained by averaging over an auxiliary randomization law. A particularly relevant instance assigns random weights to the log-likelihood contributions of individual observations\, thereby averaging inference over random perturbations of the dataset. The resulting posterior updates are governed by a fundamental design choice: the stage at which the randomization is averaged out. Averaging at the log-likelihood level recovers the power posterior for a broad class of weight distributions\, while averaging at the posterior level\, under suitable choices of weights\, recovers the bagged posterior. Averaging at the likelihood level yields a new family of generalized Bayesian updates\, which we term contamination posteriors. These are indexed by the cumulant transform of the randomization law and require a reference density for well-posedness. For Bernoulli weights\, they recover classical Huber-type contamination models\, while Beta and other weight distributions produce new generalized posteriors. We characterize the theoretical properties of randomization posteriors and illustrate their empirical robustness through simulations and experiments. We also discuss connections with bootstrap methods\, coarsened posteriors\, and other generalized Bayesian updating procedures. \nAbout the speaker: Edric (Ed) Tam is a postdoctoral fellow at the Department of Biomedical Data Science in Stanford University. His research lies at the intersection of Bayesian statistics\, artificial intelligence\, and biomedical applications. His postdoctoral work has been supported by a Warren Alpert Fellowship and a Croucher Postdoctoral Fellowship. He earned his PhD in Statistical Science from Duke University\, where he worked with David Dunson\, and holds degrees in computer science and biomedical engineering from the University of Chicago\, Yale University\, and Johns Hopkins University. He has also held research and engineering roles at Google\, Meta\, Amazon\, and Apple AIML. \nThis seminar is hosted by Professor Yunyi Shen.
URL:https://events.ucsc.edu/event/statistics-seminar-randomization-posteriors-and-generalized-bayesian-inference/
LOCATION:Jack Baskin Engineering\, Baskin Engineering 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Lectures & Presentations,Seminars
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/10/picture.jpeg
GEO:37.000369;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Jack Baskin Engineering Baskin Engineering 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Baskin Engineering 1156 High Street:geo:-122.0632371,37.000369
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261123T160000
DTEND;TZID=America/Los_Angeles:20261123T170000
DTSTAMP:20261112T184235Z
CREATED:20261112T184235Z
LAST-MODIFIED:20261112T184235Z
UID:10018103-1795449600-1795453200@events.ucsc.edu
SUMMARY:AM Seminar: Physics-Informed Learning for Nonlinear Constrained Optimization: Methods and Applications to Power Systems
DESCRIPTION:Presenter: Yu Zhang\, Associate Professor\, ECE Department of UC\, Santa Cruz \nDescription: Nonlinear constrained optimization problems arise in many scientific and engineering applications\, but solving them repeatedly can be computationally expensive\, particularly for large-scale systems. Recent advances in machine learning offer promising opportunities to accelerate optimization by learning mappings from problem parameters to optimal solutions. However\, conventional supervised learning approaches often require substantial amounts of labeled training data generated by numerical solvers and may struggle to satisfy the underlying physical constraints. In this talk\, I will present two complementary physics-informed learning frameworks for accelerating AC optimal power flow (AC-OPF)\, a fundamental nonlinear and nonconvex optimization problem in power systems. First\, I will introduce an unsupervised learning approach that integrates physical constraints into neural network training through Lagrangian duality and adaptive multiplier updates\, eliminating the need for labeled optimal solutions. Second\, I will discuss a semi-supervised learning framework that combines limited labeled data with pseudo-labeling and physics-informed gradient estimation. By exploiting the structure of the governing equations and approximating the associated Jacobian matrices\, the proposed approach substantially reduces the computational and memory requirements of training. Numerical experiments on power networks with up to several thousand buses demonstrate significant improvements in computational efficiency while maintaining near-optimal solutions and small constraint violations. These results illustrate the potential of combining mathematical optimization\, numerical methods\, and machine learning to develop scalable approaches for solving complex constrained optimization problems. \nAbout the speaker: Yu Zhang is an Associate Professor in the Department of Electrical and Computer Engineering at the University of California\, Santa Cruz. He received his Ph.D. from the University of Minnesota in 2015. Prior to joining UC Santa Cruz\, he was a postdoctoral researcher at the University of California\, Berkeley\, and Lawrence Berkeley National Laboratory. His research interests lie at the intersection of optimization\, machine learning\, and power and energy systems\, with a focus on physics-informed learning\, stochastic and robust optimization\, smart grid operations\, and power system resilience. His work develops computationally efficient and data-driven methods for addressing complex decision-making problems in modern energy systems. Dr. Zhang is a recipient of the 2019 Hellman Fellowship\, the 2021 INFORMS Energy\, Natural Resources\, and the Environment Section Early-Career Best Paper Award\, and the 2025 IISE Energy Systems Division Outstanding Young Investigator Award. His team won first place in the 2023 Learning to Run a Power Network (L2RPN) competition and the 2025 INFORMS Data Mining Society Data Challenge. \nThis seminar is hosted by Professor Qi Gong.
URL:https://events.ucsc.edu/event/am-seminar-physics-informed-learning-for-nonlinear-constrained-optimization-methods-and-applications-to-power-systems/
LOCATION:Jack Baskin Engineering\, Baskin Engineering 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Lectures & Presentations,Seminars
ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/11/myPhoto.png
GEO:37.000369;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Jack Baskin Engineering Baskin Engineering 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Baskin Engineering 1156 High Street:geo:-122.0632371,37.000369
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261130T100000
DTEND;TZID=America/Los_Angeles:20261130T123000
DTSTAMP:20261116T211006Z
CREATED:20261116T211006Z
LAST-MODIFIED:20261116T211006Z
UID:10018114-1796032800-1796041800@events.ucsc.edu
SUMMARY:Aguirre Pérez de León\, J. (STAT) - Statistical Inference for Conditional Quantiles and Fixed-Content Intervals: Models\, Computation\, and Applications
DESCRIPTION:This dissertation develops Bayesian and generalized Bayesian methods for forecasting conditional quantiles and estimating intervals with specified probability content. \nOur river-flow model combines observations and hydrologic forecasts through a shared quantile process that accounts for systematic differences between sources. Synthesizing the fitted quantile models produces distributional forecasts with the lowest average 28-day continuous ranked probability score among the various models and benchmarks considered. \nWe extend the exdqlm R package with variational updates that preserve dependence between scale and skewness\, more efficient posterior sampling\, and static regression with shrinkage priors. Numerical examples show faster variational fitting and quantify differences from Markov chain Monte Carlo in posterior and predictive accuracy. \nQuantile deep echo state networks use fixed random recurrent networks to construct nonlinear predictors for Bayesian quantile regression. Sharing information across quantile levels substantially reduces crossings\, while separate fits often recover the true quantiles more accurately. In electricity-price forecasting\, our models\, selected by validation\, achieve lower average quantile loss than a foundation-model benchmark. \nFor interval estimation\, we introduce a mean-tilted loss whose population minimizers have specified probability content and a conditional mean controlled by the tilt. We establish identification under regularity conditions; zero tilt recovers intervals that preserve the population mean. A common augmentation yields conditionally conjugate Gibbs sampling for generalized Bayesian inference and expectation/conditional maximization (ECM) for mode estimation. We use this development to select interval content and tilt in a calibrated tolerance procedure\, with a separate Dirichlet-process check of posterior content. Simulations assess content attainment and interval width\, and pharmaceutical batch data illustrate the procedure. \n  \nEvent Host: José Antonio Aguirre Pérez de León\, PhD Candidate\, Statistical Science  \nFaculty Advisors: Raquel Prado and Bruno Sanso \nZoom: https://ucsc.zoom.us/j/8168737980?pwd=RGRlNjVZbkdyaWF4S3FIVGhLQ08ydz09&omn=97370467422 \nPasscode: 4WQk7u
URL:https://events.ucsc.edu/event/aguirre-perez-de-leon-j-stat-statistical-inference-for-conditional-quantiles-and-fixed-content-intervals-models-computation-and-applications/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option2.jpg
GEO:37.0009723;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Engineering 2 Engineering 2 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Engineering 2 1156 High Street:geo:-122.0632371,37.0009723
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261130T110000
DTEND;TZID=America/Los_Angeles:20261130T130000
DTSTAMP:20261117T184341Z
CREATED:20261117T184341Z
LAST-MODIFIED:20261117T184341Z
UID:10018122-1796036400-1796043600@events.ucsc.edu
SUMMARY:Khan\, M. (CSE) - Clock Mesh Analysis and Optimization using Surrogate Models with Graph Neural Networks
DESCRIPTION:Clock Distribution Networks (CDNs) are an essential part of every modern chip that\ndistributes a clock to all the synchronous elements. Among various CDN architectures\,\nclock meshes are among the most robust architectures to On-Chip Variation (OCV) and\nprovide low skew due to its redundant paths\, thereby allowing high frequency opera-\ntion. However\, the trade-offs are: more routing resources\, increased power consumption\,\nand the inability of Static Timing Analysis (STA) to annotate delays due to no single\nsource-to-sink path. Therefore\, a full SPICE simulation is the only option for timing\nanalysis of clock meshes. Existing clock-mesh synthesis frameworks incorporate heuris-\ntics to avoid SPICE simulations\, however\, they are inaccurate especially under OCV.\nIn this thesis\, a solution for faster and accurate analysis of clock meshes under OCV\nis proposed. This is the first work on representing a clock mesh in graph domain and\ntraining a GNN-based surrogate model to predict timing and power metrics of the clock\nmesh. Furthermore\, this surrogate model is used inside an optimization loop to buffer\na given clock mesh considering OCV which has never been feasible due to SPICE being\nexpensive and heuristics being inaccurate. The proposed surrogate model achieves aver-\nage errors of 3.40 ps in delay\, 4.14 ps in slew and 4.46 ps in skew as compared to SPICE.\nWhereas\, the heuristic method has 74.37 ps delay error and 27.51 ps skew error\, while\nit cannot annotate slews. The surrogate model predicts power within 3.4% of SPICE\non average\, while delivering 18×-1706× speed-ups over the approximate method under\nOCV modeling\, enabling faster and more accurate clock-mesh design-space exploration.\nWhen used inside the optimization loop the surrogate model finds buffering candidates\nthat\, at a balanced tradeoff\, improves 99th-percentile skew by 49.1%\, reduces power\nby 25.7%\, and reduces area by 37.3% at the cost of a 6.8% increase in 99th-percentile\nmaximum slew as compared to the state-of-the-art work. \nEvent Host: Muhammad Hadir Khan\, PhD Candidate\, Computer Science & Engineering  \nAdvisor: Matthew Guthaus 
URL:https://events.ucsc.edu/event/khan-m-cse-clock-mesh-analysis-and-optimization-using-surrogate-models-with-graph-neural-networks/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-1.jpg
GEO:37.0009723;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Engineering 2 Engineering 2 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Engineering 2 1156 High Street:geo:-122.0632371,37.0009723
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261130T123000
DTEND;TZID=America/Los_Angeles:20261130T133000
DTSTAMP:20261116T183332Z
CREATED:20261116T183332Z
LAST-MODIFIED:20261116T183332Z
UID:10018112-1796041800-1796045400@events.ucsc.edu
SUMMARY:CM Seminar: Dr. Tyler Munger
DESCRIPTION:CMPM 26-27 Seminar \nWelcome to the last Computational Media seminar of the fall quarter. Open to all\, class enrollment is not required. Food is provided. \nPlease join us in-person at E2-280 or via Zoom to learn more about the works of Dr. Tyler Munger: \nTitle: TBA \nPresenter: Dr. Tyler Munger\, Associate Professor in the Technology and Information Management (TIM) program\, University of California\, Santa Cruz \nDescription: TBA \nBio: Tyler Munger is an associate professor of teaching in the Technology and Information Management (TIM) program\, and a triple Baskin Engineering alumnus. In his undergraduate years at UC Santa Cruz\, Munger joined an industry-sponsored research project that sparked his interest in TIM\, and inspired him to stay at UC Santa Cruz to complete his M.S. and Ph.D.\, while gaining nearly a decade of combined industry experience along the way. After working in industry for the last 10 years\, Munger has returned to campus as a faculty member to prepare students for successful careers in the age of artificial intelligence. \nHosted by: Assistant Professor of Computational Media\, David Lee \nZoom Link: https://ucsc.zoom.us/j/94103955103?pwd=od4ueavRt06ECpwAwqvgofwn5Suyal.1
URL:https://events.ucsc.edu/event/cm-seminar-113026/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Lectures & Presentations,Seminars
ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/03/BElogoWHITE.png
GEO:37.0009723;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Engineering 2 Engineering 2 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Engineering 2 1156 High Street:geo:-122.0632371,37.0009723
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261130T160000
DTEND;TZID=America/Los_Angeles:20261130T170000
DTSTAMP:20261015T171006Z
CREATED:20261015T171006Z
LAST-MODIFIED:20261015T171006Z
UID:10017823-1796054400-1796058000@events.ucsc.edu
SUMMARY:Statistics Seminar: Robustness\, Model Checking and Hierarchical Models
DESCRIPTION:Presenter: Rafael Cabral\, Member of Technical Staff\, Axiom Math \nDescription: Latent Gaussian models (LGMs) are among the most widely used classes of statistical models\, with applications ranging from biostatistics to spatial modeling. Their central assumption\, that the latent variables are Gaussian\, is nevertheless rarely checked. Many datasets contain sudden jumps or local spikes that a Gaussian latent process over-smooths\, impacting both inference and prediction. This talk presents a two-step workflow for hierarchical models\, in the spirit of George Box’s loop: model criticism\, to check whether the latent assumptions hold\, followed by robustification\, to fit a more flexible model when they do not. In the criticism step\, diagnostic statistics are derived rather than chosen ad hoc\, by targeting the departures from Gaussianity induced by a more flexible alternative model. This connects Bayesian predictive checking with robustness analysis and yields a general recipe for assessing latent assumptions\, and their impact on the results. In the robustification step\, latent non-Gaussian models are fitted using variational Bayes and Laplace approximations. The resulting algorithm iteratively refits LGMs that downweight extreme events in the latent process\, giving robust inferences at a fraction of the cost of MCMC. Both steps are implemented in Stan and R-INLA and illustrated with longitudinal\, areal and geostatistical applications. \nAbout the speaker: Rafael Cabral is a Member of Technical Staff at Axiom Math in Palo Alto\, where he develops AI systems for mathematics and verified reasoning. He received his PhD in Statistics from KAUST in 2023\, advised by Håvard Rue and David Bolin\, where he worked on model criticism and robust inference for latent Gaussian models. He was subsequently a Research Fellow at the National University of Singapore and a Staff Engineer at Huawei\, working on AI safety and reliability. His research interests span Bayesian modeling\, robustness\, and the application of language models to mathematics and formal reasoning. \nThis seminar is hosted by Professor Yunyi Shen.
URL:https://events.ucsc.edu/event/statistics-seminar-robustness-model-checking-and-hierarchical-models/
LOCATION:Jack Baskin Engineering\, Baskin Engineering 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Lectures & Presentations,Seminars
ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/10/1760450372199.png
GEO:37.000369;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Jack Baskin Engineering Baskin Engineering 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Baskin Engineering 1156 High Street:geo:-122.0632371,37.000369
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261201T120000
DTEND;TZID=America/Los_Angeles:20261201T130000
DTSTAMP:20261119T173919Z
CREATED:20261119T173919Z
LAST-MODIFIED:20261119T173919Z
UID:10018131-1796126400-1796130000@events.ucsc.edu
SUMMARY:Wang\, Z. (CSE) - Scaling Robust Visual Representation Learning: From Convolutional Architectures to Vision-Language Models
DESCRIPTION:Visual encoders learned at scale\, from convolutional networks and Vision Transformers to CLIP\, now serve as the perceptual front end of Vision-Language Models (VLMs).\nYet they remain brittle: their accuracy degrades under natural distribution shifts\, and imperceptible adversarial perturbations can change their predictions or the text that a VLM generates.\nRobustness has also lagged behind scale.\nIt is commonly attributed to the self-attention architecture of Transformers; its most reliable defense\, adversarial training\, has been too expensive to apply beyond small models and datasets; and VLMs have been defended only by post-hoc fine-tuning that sacrifices zero-shot capability. \nIn this dissertation\, we show that robustness can be designed into architectures\, scaled with efficient training\, and built into vision-language models.\nFirst\, through controlled architecture studies under a fixed training recipe\, we identify three Transformer-inspired designs—patchified inputs\, large convolution kernels\, and fewer activation and normalization layers—that let pure CNNs match or exceed the out-of-distribution robustness of Vision Transformers of similar size.\nSecond\, we introduce AdvXL\, an adversarial training framework that pre-trains at reduced token length with weak attacks\, briefly fine-tunes at full resolution with strong attacks\, and uses a frozen CLIP text encoder to learn from web-crawled image-text data.\nScaling to a one-billion-parameter model trained on more than one billion images sets new ImageNet-1K records under AutoAttack\, improving the previous best $\ell_\infty$-\, $\ell_2$-\, and $\ell_1$-robust accuracy by 11.4\%\, 14.2\%\, and 12.9\%.\nThird\, we propose Double Visual Defense\, which adversarially pre-trains CLIP from scratch on web-scale data and adds adversarial visual instruction tuning to LLaVA.\nEvaluated on more than 20 datasets\, the resulting models improve robustness over prior defenses by about 20\% in zero-shot ImageNet classification\, about 30\% in image captioning\, and about 20\% in visual question answering\, while hallucinating less and staying close to non-robust models on clean benchmarks. \nTogether\, these results show that robustness is not tied to a particular architecture and need not be traded for capability: efficient training at scale makes it possible to build robustness into visual foundation models from the start. \nEvent Host: Zeyu Wang\, PhD Candidate\, Computer Science & Engineering  \nAdvisor: Cihang Xie  \nZoom: https://ucsc.zoom.us/j/94378973643?pwd=GN3DqdWgUTmOfetlQ9xFgccjPQKzex.1 \nPasscode: 778543
URL:https://events.ucsc.edu/event/wang-z-cse-scaling-robust-visual-representation-learning-from-convolutional-architectures-to-vision-language-models/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.png
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261201T140000
DTEND;TZID=America/Los_Angeles:20261201T153000
DTSTAMP:20261106T210933Z
CREATED:20261106T210933Z
LAST-MODIFIED:20261106T210933Z
UID:10018063-1796133600-1796139000@events.ucsc.edu
SUMMARY:Gao\, X. (ECE) - Engineering Wireless Systems Where Infrastructure Fails: Low-Power Through-Wall Communication for Nuclear-Facility Monitoring\, UAV-Assisted Localization for Post-Disaster Search and Rescue\, and Timely Status Delivery via UAV-Assisted Data Collection
DESCRIPTION:This dissertation presents wireless systems for three operating environments without the usual infrastructure: the containment building of a nuclear facility\, the area affected by a natural disaster\, and a field of sensors with no network of their own. Each system is engineered\nto an operational requirement and tested against it by measurement or simulation. \nThe through-wall communication system carries monitoring data through a nuclear containment wall\, replacing cable wiring that radiation damages. A battery-powered sub-GHz ultra-wideband (UWB) multicarrier transmitter inside containment reaches a receive array\noutside; a water-filling algorithm is reformulated to minimize transmit power for a required throughput\, and a single-ray through-wall channel model matches measurements on a one-meter wall to about 0.1dB. At the 0.2W of an earlier UWB system\, the simulated link supports about 780Mbps\, whereas that system delivered under 10Mbps. \nThe search-and-rescue localization framework locates survivors of earthquakes\, wildfires\, or floods from their phones’ beacons\, using unmanned aerial vehicles (UAVs) with antenna arrays\, when transmit power\, path loss\, and line-of-sight (LOS) or non-line-of-sight (NLOS)\nstate are unknown. A deep-learning classifier labels each link\, and the labels select the estimator: weighted least squares\, an iterative second-order cone program with a bounded bias\, or a label-aware projection for LOS\, NLOS\, and mixed environments\, respectively. With\nthe classifier in the loop on simulated NYUSIM channels\, the end-to-end error is 46.8m\, against 66.3m for the strongest baseline; the labeled data are public. \nThe UAV-assisted status-update system maintains fresh status information at a ground station (GS) from unconnected sensors with an energy-limited\, half-duplex UAV. The objective is the age of information (AoI) at the GS\, because a collected but undelivered update does not reduce it. A hierarchical actor–critic learns only return-feasible movement\, while a deterministic scheduler constructs feasible collection and burst forwarding. For 100 sensors\, the controller lowers the average GS-side AoI by 22.5% relative to a heuristic that shares its scheduler\, and\nretains a 21.3% advantage on an unseen clustered deployment without retraining. \nThe dissertation concludes by outlining a future integrated post-disaster system: one UAV fleet for locating\, monitoring\, and supplying survivors. \n  \nEvent Host: Xiangjian Gao\, PhD Candidate\, Electrical and Computer Engineering  \nAdvisor: Hamid R. Sadjadpour \nZoom: https://ucsc.zoom.us/j/94712429320?pwd=XYXXTge4U6A3VL20wMZjzZx2bqUQX9.1 \nPasscode: 743822
URL:https://events.ucsc.edu/event/gao-x-ece-engineering-wireless-systems-where-infrastructure-fails-low-power-through-wall-communication-for-nuclear-facility-monitoring-uav-assisted-localization-for-post-disaster-search-and-re/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-1.jpg
GEO:37.0009723;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Engineering 2 Engineering 2 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Engineering 2 1156 High Street:geo:-122.0632371,37.0009723
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261202T100000
DTEND;TZID=America/Los_Angeles:20261202T110000
DTSTAMP:20261020T164621Z
CREATED:20261020T164621Z
LAST-MODIFIED:20261020T164621Z
UID:10017862-1796205600-1796209200@events.ucsc.edu
SUMMARY:Kobenova\, A. (CM) - Inclusive Joyful Learning: Designing and Evaluating Playful Immersive Systems for Neurodivergent Youth
DESCRIPTION:Neurodivergent children and young adults are frequently subjected to therapeutic and educational interventions anchored in “behavior deficits” inherited from the Medical Model of Disability. These paradigms prioritize normative performance\, forcing disabled individuals into cognitive and emotional “masking” at the expense of their autonomy and well-being. Resisting these ableist connotations\, this dissertation advances a philosophy of inclusive joyful learning\, grounded in the Affirmative Model of Disability and Flourishing Human-Computer Interaction (HCI)\, which position play\, self-directed exploration\, and bodily agency as acts of resistance against coercive instructional design. \nThis work introduces\, implements\, and evaluates the Joyful Extended Reality (XR) framework\, a set of inclusive design principles centered on neurodiverse populations. Operating through Research through Design (RtD) and adapted Participatory Design (PD) approaches\, the dissertation evaluates three immersive systems combining child-centered and expert evaluations. First\, a room-scale motion-capture implementation of the “Joyful XR” framework investigates how spatial mixed reality supports cognitive engagement and motor imitation through physical-virtual object mapping\, evaluated through a study with 21 autistic children from special education classrooms. Second\, “The Road to Acceptance”\, a narrative VR environment for emotional literacy and grief processing\, was evaluated through expert workshops with 12 mental health professionals\, including behavior analysts and bereavement support workers. Third\, “Coastal Resilience”\, an immersive game for environmental problem-solving and community safety\, was evaluated with 12 caregivers\, special education teachers\, and psychosocial professionals. \nAcross the three systems\, the findings demonstrate that instructional structures\, such as Discrete Trial Teaching and bereavement therapy\, can be successfully decoupled from adult coercion and embedded into self-directed\, collaborative game mechanics. The empirical evaluations reveal that these systems shift the facilitator’s role in educational and therapy settings. The participating children engaged more freely when a familiar adult provided relational support\, and practitioners emphasized the need for human-guided debriefing for emotionally intense concepts and for longitudinal transfer. This dissertation contributes: (1) the theoretical and structural Joyful XR design framework; (2) three documented\, playable immersive systems; and (3) transferable design requirements for multisensory facilitation\, relational mediation\, and non-coercive play in accessible extended reality environments. \nEvent Host: Amina Kobenova\, Ph.D. Candidate\, Computational Media  \nAdvisor: Sri Kurniawan \nZoom: https://ucsc.zoom.us/j/94504075063?pwd=PifJ9OXtqvFlZTYbdJECwJfGhPQi8s.1 \nPasscode: 831245
URL:https://events.ucsc.edu/event/kobenova-a-cm-inclusive-joyful-learning-designing-and-evaluating-playful-immersive-systems-for-neurodivergent-youth/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-1.jpg
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261202T100000
DTEND;TZID=America/Los_Angeles:20261202T110000
DTSTAMP:20261117T185325Z
CREATED:20261117T185156Z
LAST-MODIFIED:20261117T185325Z
UID:10018123-1796205600-1796209200@events.ucsc.edu
SUMMARY:Wang\, Y. (CSE) - Controlling Undesirable Knowledge and Behaviors in Large Language Models: Unlearning\, Evaluation\, and Feedback Dynamics
DESCRIPTION:Large language models (LLMs) can acquire and reproduce information and behaviors that later become undesirable\, such as private or copyrighted content\, hazardous knowledge\, and social bias. Controlling them is important but difficult in practice.\nLLM unlearning aims to remove the influence of such content from a model while preserving its overall capability. Retraining LLMs from scratch is often prohibitively expensive\, motivating practical interventions that may need to operate without retain data or access to model weights.\nThis dissertation studies how to unlearn or intervene selectively under practical constraints\, and what evidence is needed to assess whether an intervention works. It makes four main contributions.\nFirst\, we propose FLAT\, a loss adjustment method that unlearns using only forget data to control the knowledge.\nSecond\, for models whose weights cannot be updated\, we study training-free unlearning at inference time\, from embedding-corrupted prompts (ECO) to DRAGON\, which detects forget-worthy prompts and guards black-box LLMs through in-context chain-of-thought reasoning.\nThird\, we introduce UnlearnBench\, a clean and controllable benchmark with explicit knowledge provenance\, and show that the evaluated methods fail at entity-level unlearning even when retraining succeeds\, that multi-stage memorization changes unlearning behavior\, and that reasoning can recover supposedly forgotten information.\nFinally\, we extend our investigation from interventions on individual models to the evolution of undesirable behaviors across successive model generations. We therefore study how social bias evolves when models are repeatedly trained on their own outputs under performative feedback and propose reward-guided selection with reweighting to mitigate bias.\nUltimately\, this dissertation advances the understanding of how to control undesirable knowledge and behaviors in LLMs\, through methods designed for practical constraints and evaluations that test whether unlearning succeeds\, contributing to the responsible development and deployment of LLMs. \nEvent Host: Yaxuan Wang\, PhD Candidate\, Computer Science & Engineering \nZoom: https://ucsc.zoom.us/j/7186149767?pwd=y129sEV0UQmywrma4wQO5ffbBG7rnm.1&omn=98767260141 \nPasscode: 625461 \n 
URL:https://events.ucsc.edu/event/wang-y-cse-controlling-undesirable-knowledge-and-behaviors-in-large-language-models-unlearning-evaluation-and-feedback-dynamics/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.png
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261202T130000
DTEND;TZID=America/Los_Angeles:20261202T150000
DTSTAMP:20261119T174746Z
CREATED:20261119T174746Z
LAST-MODIFIED:20261119T174746Z
UID:10018132-1796216400-1796223600@events.ucsc.edu
SUMMARY:Luo\, Y. (CSE) - Rethinking Contrastive Learning: From Fundamental Limits to Capability Transfer Across Foundation Models
DESCRIPTION:Contrastive learning has become one of the central recipes of modern AI. By learning from comparisons\, telling which examples belong together and which do not\, it produces the embeddings behind search\, retrieval\, and the multimodal systems that connect foundation models to the world. Its success rests on assumptions that are often taken for granted: that the contrastive objective can be optimized efficiently\, that an embedding of modest dimension can faithfully capture the comparisons\, and that a new capability must be trained into an embedding model through more contrastive data. My research revisits each of these assumptions. My completed work shows that the first two can fail at a fundamental level: even finding a locally optimal embedding can be intractable\, and an embedding with too few dimensions can do no better than a trivial one. Rethinking the third\, new abilities need not come from more contrastive training: we show that an embedding model can borrow some of an LLM’s instruction-following ability without any training. Going forward\, we propose to extend this perspective to foundation models at large\, with implications for lowering the cost of building and updating AI systems\, and connections to model merging\, knowledge distillation\, multimodal alignment\, and interpretability. \nEvent Host: Yiyuan Luo\, PhD Student\, Computer Science & Engineering  \nAdvisor: Vaggos Chatziafratis \nZoom: https://ucsc.zoom.us/j/99506265345?pwd=YabUybEHGfZdtXZam8bXekqaTtz4B9.1 \nPasscode: 443041
URL:https://events.ucsc.edu/event/luo-y-cse-rethinking-contrastive-learning-from-fundamental-limits-to-capability-transfer-across-foundation-models/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option2.jpg
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261203T114000
DTEND;TZID=America/Los_Angeles:20261203T131500
DTSTAMP:20261116T195805Z
CREATED:20261116T195805Z
LAST-MODIFIED:20261116T195805Z
UID:10018113-1796298000-1796303700@events.ucsc.edu
SUMMARY:BME 280B Seminar: Dr. Colette Felton
DESCRIPTION:Welcome to the last BME 280B seminar of the fall quarter. \nPlease join us this week to learn more about the works of Dr. Colette Felton: \nTitle of Talk: Developing new tools for understanding driver alterations in lung cancer from long-read RNA sequencing \nPresenter: Colette Felton\, Postdoctoral Researcher and Bioinformaticist in Brooks lab\, University of California\, Santa Cruz \nDescription: TBA \nBio: Colette Felton is a Postdoctoral Scholar at the University of California\, Santa Cruz in the Angela Brooks lab. She primarily develops tools to work with long-read sequencing data to better understand gene alterations in cancer. Her PhD work focused on identifying alternative splicing of gene fusions in CLL and Osteosarcoma\, as well as correlating changes in chromatin accessibility with changes in RNA processing. She is currently using long read RNA sequencing to integrate variant and isoform detection for the functional prediction of cancer-related alterations. She is particularly interested in implementing these tools to identify mechanisms of drug resistance in lung cancer. \nHosted by: Professor of Biomolecular Engineering and Bioinformatics\, Angela Brooks
URL:https://events.ucsc.edu/event/bme-280b-seminar-120326/
LOCATION:Physical Sciences Building\, Physical Sciences Building\, Santa Cruz\, CA\, 95064
CATEGORIES:Lectures & Presentations,Seminars
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/11/120326-BME-280B-Seminar-Speaker-Headshot.jpg
GEO:36.9996638;-122.0618552
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Physical Sciences Building Physical Sciences Building Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Physical Sciences Building:geo:-122.0618552,36.9996638
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261203T120000
DTEND;TZID=America/Los_Angeles:20261203T140000
DTSTAMP:20261117T183643Z
CREATED:20261117T183643Z
LAST-MODIFIED:20261117T183643Z
UID:10018121-1796299200-1796306400@events.ucsc.edu
SUMMARY:Luo\, J. (CSE)- Toward Accessible 3D Head Creation: From Human Face Reconstruction to Character Animation
DESCRIPTION:Creating faithful digital humans and expressive imagined characters often requires extensive capture\, modeling\, and rigging. This dissertation investigates how observations and model structure can make 3D head creation more accessible\, from human face reconstruction to character animation. \nThe first part studies how input data type\, noise\, missing regions\, viewpoint count\, and image resolution affect facial reconstruction. It shows that visual plausibility does not guarantee geometric accuracy and that passive stereo can provide useful shape information despite noisy output. The second part develops methods for reconstructing faces from a few images. Separating camera pose and shape estimation stabilizes reconstruction\, while joint optimization of Gaussian splats and a morphable surface supports both geometry and appearance. The third part extends to imagined non-human heads. Combining 3D and video generation produces expressive blendshape models\, while learned registration enables efficient construction of consistent expression controls across diverse identities. These models support retargeting from human facial performance. \nTogether\, these studies and methods connect data acquisition\, reconstruction\, and animation to support faithful digital humans and expressive virtual characters. \nEvent Host: Jiahao Luo\, PhD Candidate\, Computer Science & Engineering  \nAdvisor: James Davis \nZoom: https://ucsc.zoom.us/j/6724718567?pwd=fe26hjskbaux3YaoWQt1RJGqucM2Z2.1 \nPasscode: 814150
URL:https://events.ucsc.edu/event/luo-j-cse-toward-accessible-3d-head-creation-from-human-face-reconstruction-to-character-animation/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option2.jpg
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261203T150000
DTEND;TZID=America/Los_Angeles:20261203T170000
DTSTAMP:20261117T190102Z
CREATED:20261117T190102Z
LAST-MODIFIED:20261117T190102Z
UID:10018124-1796310000-1796317200@events.ucsc.edu
SUMMARY:Pan\, D. (ECE) - Femtowatt End-Facet Multimode-Interference Waveguide Spectrometer and Its Applications
DESCRIPTION:This work presents a multimode-interference (MMI) waveguide spectrometer that operates at femtowatt input power. A commercial CMOS camera images the interference pattern at the output facet of the chip\, and a neural network reconstructs spectra at input powers down to the femtowatt level. The device has recovered stellar spectra at the Shane 3m telescope of Lick Observatory. Proposed work targets near-infrared astronomy\, chip-integrated photodetectors\, and polarization-resolved spectroscopy. \nEvent Host: Denghui Pan\, PhD Student\, Electrical & Computer Engineering  \nAdvisor: Holger Schmidt \nZoom: https://ucsc.zoom.us/j/98802836059?pwd=qXT5oI7mXsB3OSO4gXjkInuPfpWNHw.1 \nPasscode: 596556
URL:https://events.ucsc.edu/event/pan-d-ece-femtowatt-end-facet-multimode-interference-waveguide-spectrometer-and-its-applications/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option2.jpg
GEO:37.0009723;-122.0632371
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Engineering 2 Engineering 2 1156 High Street Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Engineering 2 1156 High Street:geo:-122.0632371,37.0009723
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261204T130000
DTEND;TZID=America/Los_Angeles:20261204T140000
DTSTAMP:20261117T173955Z
CREATED:20261117T173955Z
LAST-MODIFIED:20261117T173955Z
UID:10018120-1796389200-1796392800@events.ucsc.edu
SUMMARY:Hebbar\, P. (BMEB) - Vertebrate Gene Annotation in the Era of Telomere-to-Telomere Genomes
DESCRIPTION:Long-read sequencing has made complete genomes attainable\, but tools to interpret them lagged behind\, leaving large\, biologically important regions unreadable. A central part of my work has been rebuilding the Comparative Annotation Toolkit\, which transfers gene annotations from well-characterized references to newly assembled genomes. My changes improved accuracy in repetitive\, structurally-complex regions and made the software broadly usable. With the Telomere-to-Telomere Primates Consortium\, I led gene annotation of complete ape genomes. For the Human Pangenome Reference Consortium’s second release\, I led the annotation and analysis of genes across 460 human haplotypes. I also resolved the evolutionary history and human haplotype diversity of several important gene families\, such as NOTCH2NL\, which is implicated in human brain development. In parallel\, I led a multi-institution effort to assemble and annotate the complete genome of the common marmoset\, a primate model for neurodegenerative and neuropsychiatric disease. It is now the NCBI RefSeq and UCSC Genome Browser reference for the species. These projects also address how the genome’s most repetitive regions evolve. Across the great ape and marmoset work\, I co-developed a model in which chromosomes that exchange short-arm sequences also share closely related centromeric satellites. We propose that these shared satellites enable recombination between non-homologous chromosomes\, providing a mechanism for how ribosomal gene arrays are gained and lost across primates. More broadly\, as complete genomes and pangenomes replace single linear references\, accurate annotation becomes the bottleneck for using them. My tools and annotations extend analysis into segmental duplications and satellite arrays\, regions that drive gene innovation and harbor disease-associated variation. They also reduce reference bias when interpreting variants across diverse human populations. \nEvent Host: Prajna Hebbar\, PhD Candidate\, Biomolecular Engineering and Bioinformatics  \nAdvisor: Benedict Paten  \nZoom: https://ucsc.zoom.us/j/96216744034?pwd=zdSvAo0g6UJfcam4hqBRohYjkru3dy.1 \nPasscode: 670212
URL:https://events.ucsc.edu/event/hebbar-p-bmeb-vertebrate-gene-annotation-in-the-era-of-telomere-to-telomere-genomes/
LOCATION:Biomedical Sciences Building\, 575 McLaughlin Drive
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.png
GEO:46.1226939;-64.7891251
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Biomedical Sciences Building 575 McLaughlin Drive;X-APPLE-RADIUS=500;X-TITLE=575 McLaughlin Drive:geo:-64.7891251,46.1226939
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20261209T100000
DTEND;TZID=America/Los_Angeles:20261209T120000
DTSTAMP:20261119T175629Z
CREATED:20261119T175606Z
LAST-MODIFIED:20261119T175629Z
UID:10018133-1796810400-1796817600@events.ucsc.edu
SUMMARY:Huang\, S. (CSE) - From Hallucination Detection to Self-Improvement: Towards Reliable LLMs through Fine-Grained Feedback
DESCRIPTION:Large language models can produce unsupported claims in short phrases embedded in otherwise accurate responses. This PhD proposal investigates how to pinpoint and reduce these hallucinations using span-level supervision. I introduce SHAM-Bench\, a benchmark spanning six generation tasks\, and present a detector that uses retrieved evidence to identify hallucinated spans. The detector achieved the highest average rank in the SemEval 2025 multilingual hallucination detection challenge across 14 languages. In summarization experiments\, I find that unlikelihood training is the most effective of the mitigation methods tested. In future work\, I will investigate how reinforcement learning with verified reasoning steps can improve knowledge recall and whether jointly training an agent and a trajectory verifier enables self-improvement on agentic task performance. \nEvent Host: Sicong Huang\, PhD Student\, Computer Science & Engineering  \nAdvisor: Ian Lane \nZoom: https://ucsc.zoom.us/j/99455562030?pwd=ztbMAYTYNwcbLoYuTnp9kxEtTovVOI.1 \nPasscode: 012478
URL:https://events.ucsc.edu/event/huang-s-cse-from-hallucination-detection-to-self-improvement-towards-reliable-llms-through-fine-grained-feedback/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option2.jpg
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20270115T000000
DTEND;TZID=America/Los_Angeles:20270117T235959
DTSTAMP:20261120T200858Z
CREATED:20261120T200858Z
LAST-MODIFIED:20261120T200858Z
UID:10018142-1799971200-1800230399@events.ucsc.edu
SUMMARY:Hackathon at UCSC - CruzHacks 2027: Empowering the Next Generation of Tech Innovators
DESCRIPTION:Welcome to the premier hackathon on the West Coast. We gather hundreds of students across the world to develop solutions to real-world problems\, pursue inclusion in tech\, and kindle the spirit of innovation. \n\n\nOne weekend.\n36 hours.\n$40\,000+ in Prizes. \nBuild something real with technology alongside students\, engineers\, and mentors across the Bay Area and Silicon Valley.\nJust bring your laptop and your ambitious ideas. \nWe supply the fuel and handle the rest: mentors\, industry connections\, prizes\, and the stage for your big demo. \nOpen to students curious about building with emerging technology. No prior experience needed! \nIt’s a great place to experiment\, learn\, and build real-world expertise and experience. Applications are now open! \nThe Details: \n\nFood: Fully catered for 36 hours. Energy drinks\, snacks\, and meals are on us.\nMajor League Hacking Official Event\nSpecial Guests*: Opening and closing ceremonies featuring VIP industry leaders (to be announced).\nTitle Sponsors: Baskin School of Engineering\, Vercel\, Fetch.ai\, RedBull\, Unwrap\, VIAM\, Neon Postgres\, Nexos.ai\, Framer\, Mobbin\, Balsamiq\, Autodesk\, Machinations\, and more!\nLocation: Stevenson Event Center\nDates: January 15th – 17th (4:00 PM Check-in on Friday)\nOpen to all collegiate students.
URL:https://events.ucsc.edu/event/hackathon-at-ucsc-cruzhacks-2027-empowering-the-next-generation-of-tech-innovators/
LOCATION:Stevenson Event Center\, Stevenson Service Road\, Santa Cruz\, CA\, 95064
ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/11/Screenshot-2026-11-20-at-11.30.00-AM.png
GEO:36.996897;-122.0512963
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Stevenson Event Center Stevenson Service Road Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=Stevenson Service Road:geo:-122.0512963,36.996897
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20270125T123000
DTEND;TZID=America/Los_Angeles:20270125T133000
DTSTAMP:20261110T213556Z
CREATED:20261110T213535Z
LAST-MODIFIED:20261110T213556Z
UID:10018096-1800880200-1800883800@events.ucsc.edu
SUMMARY:CM Seminar: Dr. Sharon Hsiao
DESCRIPTION:CM 26-27 Seminar \nWelcome to the second Computational Media seminar of the winter quarter. Open to all\, class enrollment is not required. Food will be provided. \nPlease join us in-person at SVC 3212 or via Zoom to learn more about the works of Dr. Sharon Hsiao: \nTitle: TBA \nPresenter: Dr. Sharon Hsiao\, Assistant Professor of Computer Science and Engineering\, Santa Clara University \nDescription: TBA \nBio: Dr. Sharon Hsiao is an Assistant Professor & David Packard endowed junior fellow in the Department of Computer Science & Engineering at Santa Clara University. Before SCU\, Dr. Hsiao was an assistant professor at Arizona State University and Columbia University\, where she established several research projects and taught graduate & undergraduate level courses. Dr. Hsiao’s research focuses on educational technologies\, computational modeling\, visual analytics\, and adaptive technology for personalized learning. Her research methodology involves AI and HCI approaches\, applied data science\, and machine learning techniques in researching effective technology to help people learn. \nDr. Hsiao is actively involved in several international research communities\, including Educational Data Mining (EDM)\, Learning Analytics & Knowledge(LAK)\, Artificial Intelligence in Education (AIED)\, and European Conference on Technology Enhanced Learning (EC-TEL). She serves as the Program Chair of the 14th International Conference on Educational Data Mining\, Organizing Chair of the 9th International Conference on Learning Analytics & Knowledge\, and several other leadership and editing positions. \nHosted by: Associate Professor of Computational Media\, Christina Chung \nZoom Link: https://ucsc.zoom.us/j/94103955103?pwd=od4ueavRt06ECpwAwqvgofwn5Suyal.1
URL:https://events.ucsc.edu/event/cm-seminar-012527/
LOCATION:Silicon Valley Campus\, 3175 Bowers Avenue\, Santa Clara\, CA\, 95054\, United States
CATEGORIES:Lectures & Presentations,Seminars
ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/03/BElogoWHITE.png
GEO:37.3796975;-121.9765484
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Silicon Valley Campus 3175 Bowers Avenue Santa Clara CA 95054 United States;X-APPLE-RADIUS=500;X-TITLE=3175 Bowers Avenue:geo:-121.9765484,37.3796975
END:VEVENT
END:VCALENDAR