BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Events - ECPv6.17.2//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-ORIGINAL-URL:https://live-events-ucsc.pantheonsite.io
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
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260911T150000
DTEND;TZID=America/Los_Angeles:20260911T160000
DTSTAMP:20260908T164853Z
CREATED:20260908T164853Z
LAST-MODIFIED:20260908T164853Z
UID:10016388-1789138800-1789142400@live-events-ucsc.pantheonsite.io
SUMMARY:Valderrama\, S. (CSE) - Closing Data Gaps for Cybersecurity AI Training and Evaluation
DESCRIPTION:Artificial intelligence and machine learning models depend on large and diverse datasets to learn patterns across different situations and generalize to cases they have not seen before. In cybersecurity\, suitable data is difficult to find for two main reasons: first\, sensitive information limits what organizations can share; second\, public datasets often lack the coverage or context in specific use cases. My research addresses this problem by developing methods that synthetically create and curate data\, where we propose data-generation algorithms that provide mechanisms to generate data to close these gaps. \nSo far\, we have started this research by developing algorithms to curate data for specific cybersecurity needs. We collect and label a collection of 64\,132 command blocks from different data sources trying to describe the security behavior they represent. Measuring this collection reveals important differences in coverage and provides a basis for directing command generation toward these rare behaviors. We then develop a generation and validation process that fills these gaps\, creating new examples that are correct\, relevant\, and diverse. Finally\, we automate the process of creating applications that could serve different cybersecurity purposes\, which together with a set of instructions\, a way of grading them\, and reference solutions became a tasks\, a valuable training unit for machine learning models. \nThese contributions show that we can reliably guide synthetic data generation that produces useful results. As future work\, we will extend this approach from individual applications to complex environments involving multiple systems. Executions in these environments will produce records of activity linked to known actions and their outcomes\, whether benign or malicious\, enabling analysis and modeling of adversarial behavior. The broader goal is to make cybersecurity data generation more targeted\, automatic\, and useful for artificial intelligence research. \nEvent Host: Sergio Valderrama\, Ph.D. Student\, Computer Science & Engineering  \nAdvisor: Alvaro Cardenas  \nZoom: https://ucsc.zoom.us/j/92072651510?pwd=UB55FkjUfMi0vssfcqIbYsPWjDr9xw.1 \nPasscode: 026956 \n 
URL:https://live-events-ucsc.pantheonsite.io/event/valderrama-s-cse-closing-data-gaps-for-cybersecurity-ai-training-and-evaluation/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260825T110000
DTEND;TZID=America/Los_Angeles:20260825T130000
DTSTAMP:20260812T161153Z
CREATED:20260812T161153Z
LAST-MODIFIED:20260812T161153Z
UID:10015333-1787655600-1787662800@live-events-ucsc.pantheonsite.io
SUMMARY:Gomez\, J. (CSE) - Toward Sustainable and Secure Open Source Software: Discovery\, Measurement\, and Defense
DESCRIPTION:In March 2024\, a backdoor was discovered in xz Utils\, a widely used open source data compression library present in nearly every major Linux distribution. The attack was discovered days before merging into major distributions\, and if this had happened\, it would have allowed attackers to execute arbitrary code on millions of systems worldwide via SSH. \nThe success of this backdoor was enabled by two failures. The first was technical: weaknesses in the software supply chain allowed a malicious actor to inject code into a widely trusted release. The second was human: the project’s only maintainer\, overwhelmed and burned out after years of maintaining critical infrastructure alone\, was the target of a multi-year social engineering campaign\, in which a malicious actor built trust under a false identity and gradually obtained commit access to the project. This incident shows that software security failures and sustainability failures are not independent: an overburdened\, unsupported maintainer is itself an attack surface. \nAcademic and scientific open source software (OSS) faces both of these crises simultaneously. Projects that critical infrastructure depends on are maintained by researchers\, students\, and faculty who contribute in their spare time\, without dedicated security training or institutional support. Existing security frameworks including NIST’s SSDF\, OWASP’s SCVS\, and SLSA were not designed with these communities in mind\, and policy efforts such as the EU Cyber Resilience Act have shown that mandates developed without community input risk harming the ecosystems they are meant to protect. \nThis dissertation addresses the sustainability and security of academic open source software through two parallel empirical research tracks. The sustainability track combines GitHub’s REST API with LLM-based filtering to discover and characterize over 216\,000 institutionally affiliated repositories across 32 academic and research institutions\, finding that while 84\% include a README\, only 23.4% carry a detectable license and fewer than 2% include a Contributing Guide. Building on this dataset\, we develop a maturity-staged sustainability framework that classifies projects into four lifecycle stages and generates targeted recommendations for Open Source Program Offices (OSPOs). \nThe security track examines whether post-9/11 trade security programs offer a workable model for OSS supply-chain policy\, finding that effective frameworks require voluntary incentives and direct community engagement rather than top-down mandates. We further evaluate five large language models on the OWASP Benchmark for vulnerability triage\, finding that o1-mini reduces false positives by 20% over the Semgrep baseline\, demonstrating the potential for automation to reduce the security burden on individual maintainers. \nTogether\, these contributions treat sustainability and security as interconnected problems. A project that cannot sustain itself cannot secure its code\, and this dissertation takes steps toward closing both gaps. \n  \nEvent Host: Juanita Gomez\, Ph.D. Candidate\, Computer Science & Engineering \nAdvisor: Alvaro Cardenas  \nZoom: https://ucsc.zoom.us/j/91057980344?pwd=XMMjHZVgbbLXfwxKehrTEbat18066o.1 \nPasscode: 292091
URL:https://live-events-ucsc.pantheonsite.io/event/gomez-j-cse-toward-sustainable-and-secure-open-source-software-discovery-measurement-and-defense/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260824T140000
DTEND;TZID=America/Los_Angeles:20260824T160000
DTSTAMP:20260818T161442Z
CREATED:20260818T161442Z
LAST-MODIFIED:20260818T161442Z
UID:10015343-1787580000-1787587200@live-events-ucsc.pantheonsite.io
SUMMARY:Pawar\, M. (CSE) - Understanding Representations\, Reasoning\, and Decision-Making in Autonomous Driving Models
DESCRIPTION:Modern autonomous-driving models increasingly rely on learned representations and generated reasoning to interpret complex scenes and produce predictions or actions. However\, it remains unclear what information these models encode\, how that information is exposed through common interpretation methods\, and whether their stated reasoning meaningfully influences their behavior. This research investigates these questions across motion-forecasting and vision-language-action models. \nEvent Host: Manasi Pawar\, Ph.D. Student\, Computer Science & Engineering  \nAdvisor: Leilani Gilpin \nZoom: https://ucsc.zoom.us/j/92653015420?pwd=xEyOcy5ZuTcHN04La9w2KP1cmVY1ao.1 \nPasscode: 309299
URL:https://live-events-ucsc.pantheonsite.io/event/pawar-m-cse-understanding-representations-reasoning-and-decision-making-in-autonomous-driving-models/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.png
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260821T150000
DTEND;TZID=America/Los_Angeles:20260821T160000
DTSTAMP:20260817T160248Z
CREATED:20260817T160020Z
LAST-MODIFIED:20260817T160248Z
UID:10015342-1787324400-1787328000@live-events-ucsc.pantheonsite.io
SUMMARY:Huang\, X. (CSE) - Scalable and Verifiable Reasoning for Medical Foundation Models
DESCRIPTION:This PhD research focuses on developing reliable medical foundation models capable of reasoning across textual\, visual\, and interactive clinical information. The work investigates three complementary directions: improving medical reasoning through test-time scaling\, training multimodal medical models with verifiable rewards\, and synthesizing high-quality visual question-answering data from biomedical literature using generator-verifier frameworks. Building on these efforts\, the proposed research will extend medical language and multimodal models toward agentic systems that can gather evidence\, use external tools\, integrate multimodal information\, and verify decisions over sequential interactions. Overall\, this research aims to improve the reliability\, efficiency\, and transparency of medical AI reasoning while supporting reproducible and human-supervised applications in healthcare. \nEvent Host: Xiaoke Huang\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Yuyin Zhou \nZoom: https://ucsc.zoom.us/j/8855787311 \nPasscode: 197379
URL:https://live-events-ucsc.pantheonsite.io/event/huang-x-cse-scalable-and-verifiable-reasoning-for-medical-foundation-models/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.png
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260814T083000
DTEND;TZID=America/Los_Angeles:20260814T103000
DTSTAMP:20260811T162729Z
CREATED:20260811T162729Z
LAST-MODIFIED:20260811T162729Z
UID:10015331-1786696200-1786703400@live-events-ucsc.pantheonsite.io
SUMMARY:Krishnaswamy\, L. (CSE) - Network Load Balancing for Geographically Distributed Datacenters
DESCRIPTION:As datacenters scale up and become more geographically distributed\, wide-area network inter-datacenter traffic\, which typically consists of data-heavy tasks\, has become increasingly prevalent. Some of the noteworthy challenges raised by the coexistence and interaction between inter- and intra-datacenter traffic are the differences in their QoS requirements\, link utilization\, and round-trip times. To the best of our knowledge\, these challenges have not yet been addressed by current datacenter load balancers. To highlight this gap\, we conducted a comparative performance study of state-of-the-art datacenter load balancers. Through extensive simulations\, we study how they perform under different network topologies and workloads\, including intra-datacenter\, inter-datacenter\, and mixed intra- and inter-datacenter workloads that reflect how datacenters have evolved to keep up with their continuously changing driving application landscape. Our study shows that current load balancers are not able to adequately distribute load under inter-DC workloads as well as mixed intra- and inter-datacenter traffic coexistence.\nMotivated by our observations\, we introduce Balancia\, a transport agnostic\, lightweight network load balancer that dynamically switches between per-flow and per-packet control in order to provide adequate performance for both intra- and inter-DC traffic given their different characteristics and quality-of-service (QoS) requirements. We show that\, when compared against state-of-the-art load balancers\, Balancia achieves close to 80% reduction in the 99% tail flow completion times for inter-datacenter traffic in the presence of intra- and inter-datacenter workload coexistence. Further in this work\, we explore proactively monitoring for congestion with the help of phantom queues and rerouting flows in a timely manner. Through Balancia2.0 we decouple congestion control and load balancing signaling\, and examine its effects on DC and WAN traffic. \nEvent Host: Lakshmi Krishnaswamy\, Ph.D. Candidate\, Computer Science & Engineering \nAdvisor: Katia Obraczka \nZoom: https://ucsc.zoom.us/j/94414934371?pwd=7P3Umt0QQ930ESV02jMvCVHVbkIp9r.1 \nPasscode: 905786
URL:https://live-events-ucsc.pantheonsite.io/event/krishnaswamy-l-cse-network-load-balancing-for-geographically-distributed-datacenters/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260810T103000
DTEND;TZID=America/Los_Angeles:20260810T123000
DTSTAMP:20260721T182603Z
CREATED:20260721T182603Z
LAST-MODIFIED:20260721T182603Z
UID:10015110-1786357800-1786365000@live-events-ucsc.pantheonsite.io
SUMMARY:Zhao\, Z. (CSE) - TOWARD VERIFIABLE REASONING IN LLMS
DESCRIPTION:Chain-of-thought (CoT) prompting can improve final-answer performance\, but it does not guarantee that intermediate reasoning steps are faithful\, valid\, or checkable. This proposal studies how formal methods can make natural-language reasoning more reliable by translating CoT rationales into Lean artifacts\, checking the resulting theorem statements and proofs\, and using compiler feedback to diagnose and repair failures. The completed work evaluates direct zero-shot and few-shot auto-formalization pipelines for quantity- and logic-focused reasoning problems\, measuring proof type-check rate\, theorem-statement validity\, assumption faithfulness\, and repair behavior. The ongoing work extends this pipeline with AMR-guided semantic representations and altered-rationale stress tests. The planned work proposes methods of modeling LLM Agent thinking in Lean. Together\, these components separate two questions that are often conflated: whether a model translated the reasoning into the right formal goal\, and whether that goal can be proved once translated. The research goal is to develop an evaluation framework and a tool-supported workflow to improve the reliability\, auditability\, and semantic faithfulness of LLM reasoning. \nEvent Host: Zekun Zhao\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Jeffrey Flanigan \nZoom: https://ucsc.zoom.us/j/96068207641?pwd=anTpLhBXhIdaIhXKDTB32HlDMA6uIO.1 \nPasscode: 656037
URL:https://live-events-ucsc.pantheonsite.io/event/zhao-z-cse-toward-verifiable-reasoning-in-llms/
LOCATION:Silicon Valley Campus\, 3175 Bowers Avenue\, Santa Clara\, CA\, 95054\, United States
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option2.jpg
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
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260713T100000
DTEND;TZID=America/Los_Angeles:20260713T120000
DTSTAMP:20260707T160215Z
CREATED:20260707T160215Z
LAST-MODIFIED:20260707T160215Z
UID:10015010-1783936800-1783944000@live-events-ucsc.pantheonsite.io
SUMMARY:Scott\, J. (CSE) - Mechanistic Specialization Does Not Guarantee Performance: Evidence from Dual AttentionTransformers
DESCRIPTION:Dual Attention Transformers (DATs) extend decoder-only Transformers with a dedicated relational-attention stream\, making them a natural architecture for abstract identity rules such asABA and ABB. Surprisingly\, we find that comparably sized GPT-2 models outperform DATs on these tasks. We investigate this gap with two complementary mechanistic analyses. First\, causal mediation analysis shows that DATs exhibit stronger evidence of hypothesized symbolic mechanisms: symbol abstraction\, symbol induction\, and retrieval\, than GPT-2. Second\, a routing analysis shows why this specialization does not translate into better behavior: DATs make more wrong-copy errors\, can attend to the correct source token while still predicting the wrong token\, and show weak direct contribution from relational attention to the correct-versus-wrong outputmargin. Ablating positive-routing heads hurts performance\, while amplifying those headsimproves DAT more than matched controls. These results show that explicit relational attentioncan shape internal organization without guaranteeing task success. For identity-rule tasks\, performance depends not only on whether relational information is represented\, but whether it is routed to the final output position in a form that affects the next-token prediction. Because pretrained DAT and GPT-2 models differ in training data\, tokenizer\, and other implementation details\, these findings should be interpreted as evidence about the mechanisms used by existing models rather than as a definitive architectural comparison. Follow-up experiments will address these confounders through controlled training comparisons that match data\, scale\, and evaluation conditions across architectures. \nEvent Host: Jonathan Scott\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Leilani Gilpin \nZoom: https://ucsc.zoom.us/j/95404396322?pwd=0e0AegKSxhcFDDKrn08muHcqfHs6WW.1 \nPasscode: 985103
URL:https://live-events-ucsc.pantheonsite.io/event/scott-j-cse-mechanistic-specialization-does-not-guarantee-performance-evidence-from-dual-attentiontransformers/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-1.jpg
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260710T110000
DTEND;TZID=America/Los_Angeles:20260710T123000
DTSTAMP:20260626T170310Z
CREATED:20260626T170310Z
LAST-MODIFIED:20260626T170310Z
UID:10014993-1783681200-1783686600@live-events-ucsc.pantheonsite.io
SUMMARY:Levine\, R. (CSE) - Validating GPU Memory Consistency and Safety at Scale
DESCRIPTION:Graphics Processing Units (GPUs) have become essential platforms for parallel computing\, supporting applications far beyond graphics. Central to GPU programming models is its memory consistency specification (MCS)\, which defines the semantics of concurrent shared-memory operations and interacts with other language features to determine security guarantees such as memory safety. Understanding whether implementations conform to an MCS\, and whether the MCS provides a sound abstraction of real hardware\, is essential for reasoning about GPU programs and validating implementations. \nThis thesis develops techniques and large-scale studies for validating GPU memory consistency and memory safety. First\, it introduces MC Mutants\, a mutation testing methodology that systematically evaluates GPU MCS test environments. Applied to WebGPU\, MC Mutants generates a suite of conformance tests and uncovers two implementation bugs. Next\, it presents GPUHarbor\, a browser- and Android-based framework for large-scale testing across commodity GPUs. GPUHarbor enables a study of 106 GPUs from seven vendors\, reveals two previously unknown memory consistency bugs\, and provides new insights into GPU behavior that inform subsequent architectural and security studies. Finally\, this thesis presents SafeRace\, a collection of security assessments and specification proposals for preserving WebGPU memory safety in the presence of data races. Evaluated across dozens of GPUs and 21 WebGPU compilation stacks\, SafeRace identifies vulnerabilities in multiple GPU implementations\, including one assigned a CVE\, and proposes a validated path toward stronger memory safety guarantees in WebGPU. \nEvent Host: Reese Levine\, Ph.D. Candidate\, Computer Science & Engineering \nAdvisor: Tyler Sorensen \nZoom: https://ucsc.zoom.us/j/94641390195?pwd=RWXp9aprCMqmaAo8nq7oKwqTt02zwN.1 \nPasscode: 628349
URL:https://live-events-ucsc.pantheonsite.io/event/levine-r-cse-validating-gpu-memory-consistency-and-safety-at-scale/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260709T133000
DTEND;TZID=America/Los_Angeles:20260709T153000
DTSTAMP:20260623T160412Z
CREATED:20260623T160248Z
LAST-MODIFIED:20260623T160412Z
UID:10014929-1783603800-1783611000@live-events-ucsc.pantheonsite.io
SUMMARY:Carrión\, H. (CSE) - Deep Learning Algorithms for Medical Image Representation Learning and Understanding
DESCRIPTION:AI-assisted clinical decisions in medicine\, and particularly in dermatology\, demand fine-grained understanding across diverse skin tones\, body sites\, and disease types\, yet expert-annotated datasets are scarce\, demographically imbalanced\, and almost devoid of rare presentations. This dissertation develops four deep learning systems for this low-label\, low-coverage regime. We introduce HealNet\, which learns wound healing stages from longitudinal photographs without any human labels\, reaching 90.6% downstream stage-classification accuracy on a small longitudinal cohort. The Fair\, Efficient\, and Diverse Diffusion (FEDD) model then leverages powerful diffusion-model embeddings to build a skin-tone-fair\, data-efficient classifier for skin lesions\, matching or exceeding state-of-the-art performance while using only 5-20% of available labels and contributing explicit skin-tone-stratified fairness evaluation of the work. Next\, Controllable Generation of Diverse Dermatological Imagery (cgDDI) re-tasks this diffusion model to controllably synthesize skin-tone-balanced dermatological imagery\, growing a small biopsy-confirmed dataset by over 400x and reaching state-of-the-art 90.9% accuracy and improved fairness in malignancy classification\, with a +13.9% cross-dataset gain on the Fitzpatrick17k benchmark. Finally\, we introduce D-Synth and DermDepth: a synthetic dermoscopic dataset with pixel-perfect 3D ground truth and a metric-scale foundation model that closes the loop into 3D dermatology\, correcting metric scale error from over 16x to under 1.1x on real dermoscopic data and enabling single-photograph measurement of lesion reconstruction: size\, area\, and volume without specialized hardware. All data\, code\, and models are released openly to support reproducibility and ongoing fairness research. \nEvent Host: Héctor Carrión\, Ph.D. Candidate\, Computer Science & Engineering \nAdvisor: Narges Norouzi \nZoom: https://ucsc.zoom.us/j/96678782408?pwd=71f0ObEnUMNgkZ9NYnpbFLMlg1Pdm0.1 \nPasscode: 0FMVtz
URL:https://live-events-ucsc.pantheonsite.io/event/carrion-h-cse-deep-learning-algorithms-for-medical-image-representation-learning-and-understanding-2/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.png
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260630T173000
DTEND;TZID=America/Los_Angeles:20260630T200000
DTSTAMP:20260603T215647Z
CREATED:20260603T215647Z
LAST-MODIFIED:20260603T215647Z
UID:10014896-1782840600-1782849600@live-events-ucsc.pantheonsite.io
SUMMARY:Inaugural PyTorch Santa Cruz Meetup
DESCRIPTION:A community gathering of people interested in PyTorch and the projects that use it – not an official PyTorch organization. Sponsored by Red Hat and University of California Santa Cruz \nLocation: Engineering 2\, Room 180 \n​Food\, Socializing\, and Excellent talks from the PyTorch Ecosystem\n\n5:30 – 6:30 Food and Socializing\n6:30 – 7:00 Talk 1\n​7:00 – 7:30 Talk 2\n7:30 – 8:00 Talk 3\n\nFor detailed agenda and registration – visit the event website.
URL:https://live-events-ucsc.pantheonsite.io/event/inaugural-pytorch-santa-cruz-meetup/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Meetings & Conferences
ATTACH;FMTTYPE=image/avif:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/06/PyTorch_6.30.26_Event.Image_.avif
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:20260625T140000
DTEND;TZID=America/Los_Angeles:20260625T160000
DTSTAMP:20260625T183144Z
CREATED:20260625T183144Z
LAST-MODIFIED:20260625T183144Z
UID:10014992-1782396000-1782403200@live-events-ucsc.pantheonsite.io
SUMMARY:Burbano\, L. (CS) - Security of autonomous decision-making agents: From control systems to embodied AI
DESCRIPTION:Due to their increasing complexity\, autonomous decision-making agents rely on increasingly advanced algorithms\, from classical control theory to reinforcement learning (RL) and\, more recently\, large vision-language models. While these algorithms help automate the decision-making in complex systems\, they bring newer attack vulnerabilities that an adversary can exploit. In this dissertation\, we study the security of autonomous decision agents that use control systems\, RL\, and AI. We focus on the security of cyber-physical and autonomous cyber-defense systems. In particular\, we study how an attacker can compromise decision-making agents. \nFor control systems\, this dissertation studies the existence of backdoor attacks against control systems that rely on data and proposes a defense strategy against the sensors of control systems. \nFor reinforcement learning\, we study the security of autonomous cyber-defense (ACD)) agents that automatically respond to attackers’ actions in a network. While previous works focus on creating agents\, we study an adversary who compromises the agent’s own infrastructure\, manipulating the information it observes to steer the network toward an attacker-chosen state. We also propose a defense strategy that focuses on determining if an attacker is compromising the ACD. \nFinally\, we study the security of embodied AI\, where CPS rely on large vision-language models (LVLMs) for decision-making. We propose a novel attack that can cause an agent to make unsafe decisions by presenting a well-designed textual sign via the visual modality. While previous attacks against neural network-based algorithms rely on creating adversarial patches without semantic meaning\, in this work\, we exploit the fact that LVLMs can understand text. \n  \nEvent Host: Luis Burbano\, Ph.D. Candidate\, Computer Science  \nAdvisor: Alvaro Cardenas \nZoom: https://ucsc.zoom.us/j/92373119649?pwd=BLFQMrGkOxJVXnjrJhXqudN1iciZAn.1 \nPasscode: 160434\n   
URL:https://live-events-ucsc.pantheonsite.io/event/burbano-l-cs-security-of-autonomous-decision-making-agents-from-control-systems-to-embodied-ai/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260618T100000
DTEND;TZID=America/Los_Angeles:20260618T120000
DTSTAMP:20260609T193755Z
CREATED:20260609T193755Z
LAST-MODIFIED:20260609T193755Z
UID:10014912-1781776800-1781784000@live-events-ucsc.pantheonsite.io
SUMMARY:Wang\, Z. (CSE) - From Static Alignment to Adaptive Safety: Toward Reliable and Capable AI Systems
DESCRIPTION:Modern AI systems are rapidly moving beyond static text generation toward capable models and agents that reason\, use tools\, store memories\, and update persistent state\, yet safety methods still often assume a fixed model whose behavior can be controlled by output-level refusal. This leaves critical gaps in understanding why aligned models fail under adversarial pressure\, how to align reasoning models without suppressing their useful capabilities\, and how to preserve safety once capability and control are externalized into editable agent state. My research proposes a static-to-adaptive safety framework for building reliable and capable AI systems: studying the mechanisms that shape behavior inside models\, using reasoning capability as a substrate for safety alignment\, and governing persistent state as agents learn and adapt over time. We instantiate this agenda through two completed works and three proposed directions. AttnGCG studies adversarial failures in aligned language models\, showing how jailbreak attacks can manipulate model attention and expose limitations of output-level safety analysis. STAR-1 studies safety alignment for large reasoning models\, showing that policy-grounded reasoning data can improve safety while largely preserving general reasoning capability. Building on these foundations\, we further study when editable agent harnesses meaningfully affect future behavior\, how persistent state creates new safety risks\, and how adaptive agents can safely update state while preserving useful learning. Together\, my research aims to move beyond static alignment alone\, toward AI systems whose safety remains reliable as their capabilities expand through reasoning and adaptation. \nEvent Host: Zijun Wang\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Cihang Xie  \nZoom ID:  962 8317 0929 \nPasscode: 687715
URL:https://live-events-ucsc.pantheonsite.io/event/wang-z-cse-from-static-alignment-to-adaptive-safety-toward-reliable-and-capable-ai-systems/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-1.jpg
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260618T100000
DTEND;TZID=America/Los_Angeles:20260618T120000
DTSTAMP:20260526T162714Z
CREATED:20260526T162714Z
LAST-MODIFIED:20260526T162714Z
UID:10014867-1781776800-1781784000@live-events-ucsc.pantheonsite.io
SUMMARY:Carrión\, H. (CSE) - Deep Learning Algorithms for Medical Image Representation Learning and Understanding
DESCRIPTION:AI-assisted clinical decisions in medicine\, and particularly in dermatology\, demand fine-grained understanding across diverse skin tones\, body sites\, and disease types\, yet expert-annotated datasets are scarce\, demographically imbalanced\, and almost devoid of rare presentations. This dissertation develops four deep learning systems for this low-label\, low-coverage regime. We introduce HealNet\, which learns wound healing stages from longitudinal photographs without any human labels\, reaching 90.6% downstream stage-classification accuracy on a small longitudinal cohort. The Fair\, Efficient\, and Diverse Diffusion (FEDD) model then leverages powerful diffusion-model embeddings to build a skin-tone-fair\, data-efficient classifier for skin lesions\, matching or exceeding state-of-the-art performance while using only 5-20% of available labels and contributing explicit skin-tone-stratified fairness evaluation of the work. Next\, Controllable Generation of Diverse Dermatological Imagery (cgDDI) re-tasks this diffusion model to controllably synthesize skin-tone-balanced dermatological imagery\, growing a small biopsy-confirmed dataset by over 400x and reaching state-of-the-art 90.9% accuracy and improved fairness in malignancy classification\, with a +13.9% cross-dataset gain on the Fitzpatrick17k benchmark. Finally\, we introduce D-Synth and DermDepth: a synthetic dermoscopic dataset with pixel-perfect 3D ground truth and a metric-scale foundation model that closes the loop into 3D dermatology\, correcting metric scale error from over 16x to under 1.1x on real dermoscopic data and enabling single-photograph measurement of lesion reconstruction: size\, area\, and volume without specialized hardware. All data\, code\, and models are released openly to support reproducibility and ongoing fairness research. \nEvent Host:  Héctor Carrión\, Ph.D. Candidate\, Computer Science & Engineering \nAdvisor: Narges Norouzi \nZoom: https://ucsc.zoom.us/j/96678782408?pwd=71f0ObEnUMNgkZ9NYnpbFLMlg1Pdm0.1 \nPasscode: 0FMVtz
URL:https://live-events-ucsc.pantheonsite.io/event/carrion-h-cse-deep-learning-algorithms-for-medical-image-representation-learning-and-understanding/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.png
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260609T120000
DTEND;TZID=America/Los_Angeles:20260609T130000
DTSTAMP:20260526T161617Z
CREATED:20260526T161617Z
LAST-MODIFIED:20260526T161617Z
UID:10014865-1781006400-1781010000@live-events-ucsc.pantheonsite.io
SUMMARY:Kim\, C. (CSE)- Toward Adaptive Graph Processing and Fault-Tolerant Agentic Inference on Heterogeneous Distributed Systems
DESCRIPTION:Edge computing and distributed AI systems increasingly operate under heterogeneous resources\, dynamic workloads\, and frequent failures\, requiring both adaptivity and fault tolerance for efficient execution. In heterogeneous edge clusters\, nodes differ significantly in CPU throughput\, memory capacity\, and network bandwidth\, while modern distributed GPU clusters supporting agentic LLM inference must recover large amounts of runtime state under routine failures. This dissertation addresses these challenges through two systems: Zsiga\, an adaptive distributed graph processing system for heterogeneous edge clusters\, and Forte\, a fault-tolerant KV cache recovery system for distributed agentic LLM inference. \nZsiga improves connected component computation through capacity-aware graph partitioning and runtime-adaptive boundary migration\, reducing execution time by up to 90.9% while eliminating out-of-memory failures under heterogeneous resource constraints. Forte addresses KV cache recovery for long-running agentic inference workloads\, where failures can erase accumulated reasoning trajectories and tool interaction histories. Forte exploits the observation that not all KV blocks are equally critical\, introducing criticality-aware erasure coding\, domain-diverse placement\, and prioritized foreground recovery to enable efficient recovery under correlated failures. Experimental results show that Forte is the only evaluated scheme that successfully resumes execution under correlated domain failures\, reducing foreground stall by 89.7% and end-to-end recovery latency by 50.6–58.9% at 2.0$\times$ memory overhead. Together\, these systems demonstrate how adaptivity and fault tolerance can improve the efficiency and resilience of distributed systems in heterogeneous and failure-prone environments. \nEvent Host: Chaeeun Kim\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Chen Qian & Liting Hu \nZoom: https://ucsc.zoom.us/j/9863615188?pwd=kTka0aZXJ070tor1EKvrt3X6AveBRp.1 \nPasscode:  cG5SL8 \n  \n 
URL:https://live-events-ucsc.pantheonsite.io/event/kim-c-cse-toward-adaptive-graph-processing-and-fault-tolerant-agentic-inference-on-heterogeneous-distributed-systems/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.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:20260609T103000
DTEND;TZID=America/Los_Angeles:20260609T130000
DTSTAMP:20260526T194445Z
CREATED:20260526T194326Z
LAST-MODIFIED:20260526T194445Z
UID:10014873-1781001000-1781010000@live-events-ucsc.pantheonsite.io
SUMMARY:Shen\, G. (CSE) - Library-Level Choreographic Programming
DESCRIPTION:Modern software increasingly relies on distributed systems to provide accessible\, scalable\,\nand reliable services. Choreographic programming brings a global perspective to distributed\nsystem development: programmers write a single program that describes the behavior of a\nwhole system\, and a compiler projects that global description into local programs run by each\nnode. By making distributed control flow explicit\, choreographic programming can rule out\nimportant classes of errors\, including deadlocks. This dissertation investigates library-level\nchoreographic programming\, an approach that embeds choreographic abstractions in existing\nhost languages rather than implementing them as standalone languages. The central claim\nis that the library approach can retain the safety and global reasoning principles of chore-\nographic programming while taking advantage of the host language’s features\, tools\, and\necosystem. First\, we present HasChor\, a first-of-its-kind library-level choreographic program-\nming language in Haskell\, built using freer monads. Next\, we generalize the design underlying\nHasChor to algebraic effects\, giving library-level implementations in Agda and OCaml. Fi-\nnally\, we present Parkour\, a backward-compatible extension to HasChor that adds a construct\nfor expressing parallel behavior in choreographies. Together\, these systems show that chore-\nographic programming can be implemented\, generalized\, and extended at the library level\,\nmaking global programming techniques available within practical host-language settings. \nEvent Host: Gan Shen\, Ph.D. Candidate\, Computer Science & Engineering  \nAdvisor: Lindsey Kuper  \nZoom: https://ucsc.zoom.us/j/93790633483?pwd=Jg8JlISsrwjLBaQIi1KdHk36bNMIv7.1 \nPasscode: 902041 \n 
URL:https://live-events-ucsc.pantheonsite.io/event/shen-g-cse-library-level-choreographic-programming/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260605T080000
DTEND;TZID=America/Los_Angeles:20260605T100000
DTSTAMP:20260527T160819Z
CREATED:20260527T160819Z
LAST-MODIFIED:20260527T160819Z
UID:10014878-1780646400-1780653600@live-events-ucsc.pantheonsite.io
SUMMARY:Chen\, Z. (CSE) - GPU Subgroup Semantics for Portable High-Performance Kernels
DESCRIPTION:Modern high-performance GPU kernels increasingly rely on subgroup-level execution\, including subgroup-level communication\, subgroup operations\, and matrix operations. These features are essential for workloads such as matrix multiplication and FlashAttention\, but their language-level guarantees remain difficult to reason about. Existing programming models often leave unclear which threads participate in subgroup operations\, when subgroup threads are required to execute together\, and what synchronization is implied by subgroup-level operations. This ambiguity becomes especially important in portable GPU programming\, where the same kernel may run across devices with different subgroup sizes\, compiler stacks\, browser backends\, and hardware execution behavior. \nMy research studies how precise subgroup semantics can support portable and correct high-performance GPU kernels. SIMT-Step\, my main completed work\, develops a formal and flexible operational semantics for GPU subgroup execution. It introduces dynamic blocks to specify converged subgroup execution and subgroup-operation participation\, classifies instructions as independent\, synchronous\, or collective to express a spectrum of candidate subgroup semantics\, and validates these models through a TLA+ implementation and an empirical fuzzing study across real GPUs. My systems work studies how subgroup-dependent kernels behave in practice\, including WebGPU FlashAttention kernels for LLM inference\, tunable WebGPU kernels for performance portability\, and Vulkan-based execution for heterogeneous SoCs. Building on these foundations\, my proposed verification work develops data-race-free checking techniques for ML kernels that rely on subgroup operations and matrix operations. Together\, these projects aim to clarify the execution guarantees that optimized GPU kernels can rely on and to support portable GPU programming systems whose performance and correctness can be reasoned about across diverse hardware. \nEvent Host: Zheyuan Chen\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Tyler Sorensen \nZoom: https://ucsc.zoom.us/j/92175288480?pwd=jGajtqerVbKuW1FPNr3awqOYoxATsp.1&jst=3 \nPasscode: 693354
URL:https://live-events-ucsc.pantheonsite.io/event/chen-z-cse-gpu-subgroup-semantics-for-portable-high-performance-kernels/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260604T140000
DTEND;TZID=America/Los_Angeles:20260604T153000
DTSTAMP:20260527T164116Z
CREATED:20260527T164116Z
LAST-MODIFIED:20260527T164116Z
UID:10014879-1780581600-1780587000@live-events-ucsc.pantheonsite.io
SUMMARY:Imlau Dagostini\, J. (CSE) - Intent-Driven Orchestration for Scientific Computing
DESCRIPTION:The growing complexity of high-performance computing (HPC) systems poses a fundamental challenge for domain scientists\, whose primary objective is to obtain scientifically valid results rather than to optimize resource utilization. Modern leadership-class facilities combine heterogeneous CPUs\, GPUs\, and specialized accelerators across systems that simultaneously support traditional scientific simulations and AI-driven workloads. This creates a vast\, machine-dependent configuration space that even experienced systems researchers find difficult to navigate. In practice\, users must explicitly specify resources\, node counts\, and walltime estimates before submitting jobs to an orchestrator\, resulting in iterative trial-and-error that wastes both human effort and compute resources. \nThis thesis proposes an intent-driven orchestration middleware for scientific computing\, in which domain scientists express high-level computational goals rather than low-level resource parameters\, and the system assumes responsibility for identifying configurations that satisfy those goals efficiently. This thesis proposal builds on a completed study of the computational performance of pangenome mapping\, a representative workload of data-intensive pipelines increasingly common in modern science. We demonstrate that tailoring tuning parameters to specific inputs and architectures yields significant performance improvements while exposing the depth of the configuration search problem that motivates this thesis. We then present an in-progress user-aware\, intent-driven middleware that uses surrogate models to aid this exploration and map high-level goals to suitable configurations. We end this presentation by proposing a cluster-aware orchestrator that enables existing HPC resource managers to support intent-aware decision-making. \nEvent Host: Jessica Imlau Dagostini\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Abel Souza \nZoom: https://ucsc.zoom.us/j/93851280425?pwd=v4ONi9N5UlfZmsMqiI4gSkxFXe0oaX.1 \nPasscode: 835985 \n 
URL:https://live-events-ucsc.pantheonsite.io/event/imlau-dagostini-j-cse-intent-driven-orchestration-for-scientific-computing/
LOCATION:Jack Baskin Engineering\, Baskin Engineering 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.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:20260604T100000
DTEND;TZID=America/Los_Angeles:20260604T120000
DTSTAMP:20260512T171434Z
CREATED:20260512T161057Z
LAST-MODIFIED:20260512T171434Z
UID:10014625-1780567200-1780574400@live-events-ucsc.pantheonsite.io
SUMMARY:Kordonowy\, S. (CS) - The Role of Circuits in Near-Term Quantum Computation
DESCRIPTION:As quantum computing transitions from theory to practice\, understanding which algorithms suit near-term devices becomes critical. Current quantum computers are severely constrained by limited qubit counts\, short coherence times\, and high error rates that quickly degrade computation into noise. This thesis addresses two interconnected questions: what non-trivial computational tasks can near-term devices execute and how should algorithms be implemented to exploit available hardware? We examine circuit design as the bridge between these concerns\, analyzing how gate choices determine algorithmic efficiency and computational hardness. By deriving explicit circuit constructions\, we obtain tangible cost estimates for practical quantum computation\, enabling precise comparisons to classical approaches and identification of break-even points in system size and error rates. Understanding these trade-offs is essential for near-term quantum computing\, where experiments are expensive and error-prone. \nWe apply these ideas to three domains:\n1. Streaming: we provide circuit implementations for the Boolean Hidden Matching problem\, a combinatorial problem which exhibits exponential space separation compared to classical algorithms. We give explicit resource estimates and experimentally validate on Quantinuum’s trapped-ion hardware. We demonstrate that quantum advantage persists even when accounting for error correction overhead. \n2. Variational eigensolving: We examine how gate set choices influence trainability of variational quantum eigensolvers and provide Lie algebraic decompositions for differing gate sets. These decompositions are in turn used as a warm-starting heuristic to overcome barren plateaus\, a common problem in quantum machine learning tasks\, and improve convergence. We apply this technique to three combinatorial problems with primary focus on portfolio optimization. \n3. Cryptography: We develop a digital signature scheme based on circuit learning hardness and classical shadows. Error detection plays a direct role in the circuits considered\, with a focus on practical implementation for near-term devices. \nThese case studies demonstrate how careful circuit design can either mitigate near-term\nconstraints or expose where error correction becomes necessary to achieve quantum\nadvantage. \n  \nEvent Host: Steven Kordonowy\, Ph.D. Candidate\, Computer Science  \nAdvisor: Alexandra Kolla  \nZoom: https://ucsc.zoom.us/j/9524731001?pwd=MzdrNmhidVBsTXNFbktBcjEvNmZIQT09&omn=96338496668  \nPasscode: J29XGi \n  \n 
URL:https://live-events-ucsc.pantheonsite.io/event/kordonowy-s-cs-the-role-of-circuits-in-near-term-quantum-computation/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.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:20260603T110000
DTEND;TZID=America/Los_Angeles:20260603T121500
DTSTAMP:20260529T172740Z
CREATED:20260529T172740Z
LAST-MODIFIED:20260529T172740Z
UID:10014889-1780484400-1780488900@live-events-ucsc.pantheonsite.io
SUMMARY:
DESCRIPTION:Presenter: Sai Teja Peddinti\, Google \nAbstract: As the digital landscape expands\, traditional models of threat mitigation and user support are failing to keep pace with the unprecedented security\, privacy\, and safety challenges. Fortunately\, the rise of large language models (LLMs) offers a powerful new paradigm for defense. This talk explores how LLMs are being leveraged to improve digital privacy\, security\, and safety from the network layer down to the individual user. We will examine how LLMs are opening new frontiers in cybersecurity and solving complex challenges\, such as: inferring device identities through semantic analysis of network traffic\, mapping global privacy trends by distilling over a decade of app reviews\, and analyzing user help-seeking behaviors across millions of social media interactions. Ultimately\, this talk will demonstrate how AI is evolving from a technological novelty into an essential foundation for scalable\, proactive\, and human-centric digital defense. \nBio: Sai Teja Peddinti (https://www.saitejapeddinti.com) is a Staff Research Scientist at Google\, where his research focuses on the intersection of Privacy\, Security\, Artificial Intelligence\, and Data Mining. His research employs a multidisciplinary approach\, blending qualitative and quantitative methods to investigate user and developer privacy preferences and translate those insights into scalable privacy/security features using LLMs and large-scale data analysis. Sai Teja holds a Ph.D. in Computer Science from the NYU Tandon School of Engineering (2014). His research has garnered industry recognition\, including the IAPP SOUPS Privacy Award and finalist placements in major applied research competitions. Throughout his education\, he has been honored with numerous accolades. \nHosted by: Professor Ram Sundara Raman \nDate and Time: Wednesday\, June 3\, from 11:00 am – 12:15 pm \nLocation: Engineering 2\, Room E2-180 (Refreshments such as fruit\, pastries\, coffee\, and tea will be provided.) \nZoom Option: https://ucsc.zoom.us/j/93445911992?pwd=YkJ2TQtF79h0PcNXbEcpZLbpK0coiY.1&jst=3
URL:https://live-events-ucsc.pantheonsite.io/event/12348/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Lectures & Presentations,Seminars
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/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:20260602T130000
DTEND;TZID=America/Los_Angeles:20260602T150000
DTSTAMP:20260526T162137Z
CREATED:20260526T162137Z
LAST-MODIFIED:20260526T162137Z
UID:10014866-1780405200-1780412400@live-events-ucsc.pantheonsite.io
SUMMARY:Sheaves\, T. (CSE) - Timing Side-Channels in Commercial ReRAM: Toward ReRAM Pentimenti
DESCRIPTION:Recently\, a class of non-invasive hardware side-channel attacks has been discovered in field-programmable gate arrays (FPGAs). These attacks extract remnants of prior users’ activity that persist as transistor defect states within reconfigurable routing resources. These remnants are known as FPGA Pentimenti. Resistive random-access memory (ReRAM) is a compelling candidate for pentimenti-like attacks beyond FPGAs. However\, unlike FPGAs\, where sophisticated on-chip sensors capable of detecting pentimenti have been well-studied\, non-invasive pentimenti recovery in commercial ReRAM must rely on measurements of observable write latency. These measurements are dominated by data-dependent structural biases that obscure any underlying defect-dynamics signal. In this dissertation\, we demonstrate that the structural and stochastic components of commercial ReRAM write latency can be decoupled and recovered through non-invasive timing analysis alone. Our results provide the reverse engineering and measurement infrastructure for future study of ReRAM pentimenti by isolating the component of programming latency sensitive to defect dynamics. \nEvent Host: Tyler Sheaves\, Ph.D. Candidate\, Computer Science & Engineering  \nAdvisor: Dustin Richmond  \nZoom: https://ucsc.zoom.us/j/92729427179?pwd=BpYLqft18YdOU0mDdQWs8erID2VcHi.1 \nPasscode: 939530
URL:https://live-events-ucsc.pantheonsite.io/event/sheaves-t-cse-timing-side-channels-in-commercial-reram-toward-reram-pentimenti/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260529T110000
DTEND;TZID=America/Los_Angeles:20260529T123000
DTSTAMP:20260515T164420Z
CREATED:20260515T164420Z
LAST-MODIFIED:20260515T164420Z
UID:10014643-1780052400-1780057800@live-events-ucsc.pantheonsite.io
SUMMARY:Zhou\, K. (CSE) - Toward Safer Frontier AI: From Evaluation and Red-Teaming to Alignment and Oversight
DESCRIPTION:This dissertation investigates how to make modern AI systems safer as they grow more capable. It addresses two central sources of risk: malicious misuse\, in which adversarial users coerce models into harmful behavior\, and internal misalignment\, in which models themselves pursue goals that diverge from human intent through deception\, sandbagging\, or other covert behaviors. The dissertation identifies novel safety risks in frontier multimodal large language models and AI agents\, introduces a black-box red-teaming framework for AI agents\, proposes new safety alignment algorithms\, and builds the first probe-based misalignment monitoring system\, developing practical approaches for evaluating\, red-teaming\, aligning\, and overseeing frontier language models and agents. The central conclusion is that responsible AI cannot rest on any single guardrail: capability-scaled evaluation\, active red-teaming\, training-time alignment\, and scalable monitoring together form a coordinated stack for frontier AI safety. \nEvent Host: Kaiwen Zhou\, Ph.D. Candidate\, Computer Science & Engineering  \nAdvisor: Xin Wang \nZoom: https://ucsc.zoom.us/j/94196702062?pwd=b9LJMfL232ixG2THMab8XuJ32a4FVD.1 \nPasscode:  584794
URL:https://live-events-ucsc.pantheonsite.io/event/zhou-k-cse-toward-safer-frontier-ai-from-evaluation-and-red-teaming-to-alignment-and-oversight/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option2.jpg
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260528T130000
DTEND;TZID=America/Los_Angeles:20260528T150000
DTSTAMP:20260514T160625Z
CREATED:20260514T160341Z
LAST-MODIFIED:20260514T160625Z
UID:10014635-1779973200-1779980400@live-events-ucsc.pantheonsite.io
SUMMARY:Yang\, D. (CSE) - Inner Monologue: a Pathway to Human-Like Reasoning for Complex Tasks
DESCRIPTION:A central goal on the path toward general AI is to build systems capable of deliberative reasoning before action. Such systems should inspect what they know\, identify what they need\, seek or construct useful information\, and revise their reasoning through intermediate cognitive states. This dissertation studies this goal through the lens of Inner Monologue (IM)\, a mechanism that enables AI systems to coordinate internal components\, acquire external information\, and reason through structured intermediate states. \nI will first introduce IM as a mechanism for internal coordination in static information systems\, where multiple models collaborate within one AI system to solve reasoning tasks. I will then extend IM to dynamic information systems\, where AI system is learned to retrieve external information. Finally\, I will present how IM can move beyond verbal reasoning toward multimodal thinking\, where generated visual states represent the system’s current understanding and support iterative refinement. \nTogether\, this dissertation demonstrates the success and potential of human-inspired Inner Monologue mechanisms for improving complex multi-step reasoning in AI systems. \nEvent Host: Diji Yang\, Ph.D. Candidate\, Computer Science & Engineering \nAdvisor: Yi Zhang \nZoom: https://ucsc.zoom.us/j/99915235963?pwd=7Jqo6fc83LWobTEYRZCUzbrWbeov3Y.1 \nPasscode: 7Jqo6fc83LWobTEYRZCUzbrWbeov3Y.1
URL:https://live-events-ucsc.pantheonsite.io/event/yang-d-cse-inner-monologue-a-pathway-to-human-like-reasoning-for-complex-tasks/
LOCATION:Silicon Valley Campus\, 3175 Bowers Avenue\, Santa Clara\, CA\, 95054\, United States
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.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
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260528T120000
DTEND;TZID=America/Los_Angeles:20260528T140000
DTSTAMP:20260526T163353Z
CREATED:20260526T163353Z
LAST-MODIFIED:20260526T163353Z
UID:10014868-1779969600-1779976800@live-events-ucsc.pantheonsite.io
SUMMARY:Ortiz Barbosa\, D. (CSE) - HARDENING AUTONOMOUS CYBER-PHYSICAL SYSTEMS AGAINST ADVERSARIAL CONDITIONS
DESCRIPTION:Autonomous systems\, such as Autonomous Vehicles (AVs) and drones\, are increasingly\ndeployed across a wider array of contexts for both civilian and military use. As these\nsystems become more common\, they may be targeted by malicious actors seeking to\nexploit and abuse them\, compromising safety-critical operations. Among the ways to\nprotect these systems simulation based testing frameworks have been developed. How-\never\, existing testing frameworks primarily focus on identifying logical flaws or system\nvulnerabilities\, often emphasizing static scenarios and paying less attention to an adap-\ntive intelligent adversary.\nTo help reduce this gap\, this dissertation develops and applies adaptive\, adversary-\naware methodologies to discover\, formalize\, and mitigate security vulnerabilities in au-\ntonomous systems spanning vehicle platooning\, drone swarms\, and vision-based drone\nrecovery. We first apply NLP techniques to discover and formalize driving rules across\nNorth American and Australian jurisdictions\, identifying possible restriction that an\nadversary can exploit. Likewise\, these rules can be used to test the adaptability of AVs\nto new contexts and to establish a formal basis for context-aware AV testing. Next\,\nwe apply optimization-based adversarial search to both ACC-controlled vehicle pla-\ntoons and obstacle-avoiding drone swarms. We uncover maneuvers that an adversary\ncan use against the system that range from crash-inducing patterns against platooning\ncontrollers to herding strategies that divert swarms from their objectives. Finally\, to\naddress the gap regarding the possible solutions to an adversarial attack we explore how\na drone can recover from it by using LVLMs to understand its context and select a safe\nlanding surface. \nEvent Host: Diego Ortiz Barbosa\, Ph.D. Candidate\, Computer Science & Engineering  \nAdvisor: Alvaro A Cardenas
URL:https://live-events-ucsc.pantheonsite.io/event/ortiz-barbosa-d-cse-hardening-autonomous-cyber-physical-systems-against-adversarial-conditions/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260528T110000
DTEND;TZID=America/Los_Angeles:20260528T120000
DTSTAMP:20260522T165248Z
CREATED:20260522T165248Z
LAST-MODIFIED:20260522T165248Z
UID:10014863-1779966000-1779969600@live-events-ucsc.pantheonsite.io
SUMMARY:Oh\, S. (CSE) - Efficient Instruction Supply for Datacenter Processors
DESCRIPTION:Modern datacenter CPUs lose 25–66% of execution cycles to instruction-delivery stalls. This bottleneck persists\, despite the recent trend towards accelerators and GPUs\, as there is continuing demand by applications that only execute on CPUs. Two workload classes dominate today’s datacenter execution cycles: hyperscale server software (databases\, build systems\, and content stores)\, whose large instruction footprints create severe frontend pathologies; and agentic AI systems\, in which large-language-model agents plan\, dispatch tools\, and maintain growing conversational contexts\, causing CPUs to account for up to 88% of end-to-end agent latency. Reflecting this shift\, major CPU vendors have publicly repositioned the CPU as the orchestration layer of the AI stack and have begun shipping processors optimized for agent-centric workloads. \nThis dissertation argues that instruction delivery is the dominant CPU bottleneck across both workload classes and that the recent trend towards agentic AI further exacerbates this challenge. In hyperscale server binaries\, the primary pathologies are wrong-path prefetch pollution and post-recovery instruction-delivery gaps across large\, irregular call graphs. In agentic AI systems\, the bottleneck shifts to an orchestration substrate composed of protocol stacks\, dynamic-runtime dispatch\, and agent-specific extensions that is even more frontend-bound than traditional warehouse-scale workloads. \nTo address these bottlenecks\, this dissertation presents three technical contributions\, together with a companion infrastructure contribution. First\, Utility-Driven Prefetching (UDP) extends fetch-directed instruction prefetching (FDIP) with a learned per-prefetch utility model that admits candidates based on their historical contribution to demand-fetch hits\, including those reached along wrong-path execution. Second\, Junction-based Unified Miss-point Prefetching (JUMP) addresses the post-recovery instruction-delivery gap that UDP and prior FDIP optimizations cannot reach by launching a lightweight secondary FDIP thread at a learned miss point following each branch-prediction failure. Across a suite of datacenter workloads\, UDP improves IPC by 3.6% on average (up to 16.1%) over a state-of-the-art FDIP baseline\, while JUMP improves IPC by 2.0% on average (up to 14.9%). Combined\, the two mechanisms substantially close the gap between FDIP and a perfect L1 instruction cache at a storage cost of only a few tens of kilobytes.\nThird\, this dissertation introduces the Agentic Tax\, the first CPU characterization study of agentic AI workloads across three runtime families. The study is packaged as a deterministic-replay benchmark infrastructure that enables repeatable\, cycle-level evaluation under controlled conditions. The characterization shows that the orchestration substrate of agentic AI workloads is significantly more frontend-bound than the hyperscale datacenter workloads examined in prior work\, and that it introduces new dominant function families with no analog in traditional warehouse-scale systems. These findings motivate two architectural directions proposed as future work: extending UDP and JUMP to optimize the orchestration substrate itself\, and designing heterogeneous CPU cores that allocate frontend resources according to the execution phase. \nEvent Host: Surim Oh\, Ph.D. Candidate\, Computer Science & Engineering  \nAdvisor: Heiner Litz \nZoom: https://ucsc.zoom.us/j/94753352649?pwd=7vQxlnSJkUb0KfG3t6STo639LhRv7j.1 \nPasscode: 205162
URL:https://live-events-ucsc.pantheonsite.io/event/oh-s-cse-efficient-instruction-supply-for-datacenter-processors/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260527T120000
DTEND;TZID=America/Los_Angeles:20260527T140000
DTSTAMP:20260518T162624Z
CREATED:20260518T162624Z
LAST-MODIFIED:20260518T162624Z
UID:10014651-1779883200-1779890400@live-events-ucsc.pantheonsite.io
SUMMARY:Zheng\, Y. (CSE) - Extending eBPF Beyond Kernel Extensions: Verified Interfaces for Runtime System Extensibility
DESCRIPTION:Modern system software increasingly needs runtime extensibility: userspace applications need safe ways to expose domain-specific extension points\, GPU resource management needs workload-specific memory and scheduling policies\, and kernel eBPF JIT compilers need different runtime optimizations as workloads and hardware vary. However\, built-in policies are safe but difficult to specialize across rapidly changing workloads and hardware environments\, limiting efficiency\, while code modifications are flexible but difficult to deploy safely. This dissertation argues that verified eBPF interfaces can turn eBPF from a kernel-extension mechanism into a general substrate for safe runtime extensibility. In this model\, trusted mechanisms expose narrow\, constrained programmable hooks; extensions declare their requirements; verifier-enforced checks preserve safety; and execution remains low-overhead. \nI develop this thesis through three systems spanning userspace applications\, heterogeneous GPU subsystems\, and the kernel eBPF compiler itself. EIM\, implemented in bpftime\, applies verified eBPF interfaces to userspace applications\, allowing application behavior to be extended through explicit constraints and efficient userspace eBPF execution. gpu_ext extends the same idea to heterogeneous systems by exposing programmable resource management hooks for GPU memory and scheduling policy across driver and device. BpfReJIT with kinsn makes the eBPF JIT compiler itself extensible: it enables runtime-guided optimization through dynamic recompilation and extends eBPF bytecode to express diverse hardware capabilities. Together\, these systems show how verified eBPF interfaces can support safe programmability\, separation of policy and mechanisms\, and runtime specialization across applications\, GPU subsystems\, and the kernel JIT infrastructure. \nEvent Host: Yusheng Zheng\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Andi Quinn \nZoom: 504 350 0245 \nPasscode: 521336
URL:https://live-events-ucsc.pantheonsite.io/event/zheng-y-cse-extending-ebpf-beyond-kernel-extensions-verified-interfaces-for-runtime-system-extensibility/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option-3.png
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260527T110000
DTEND;TZID=America/Los_Angeles:20260527T123000
DTSTAMP:20260330T203942Z
CREATED:20260330T203942Z
LAST-MODIFIED:20260330T203942Z
UID:10011815-1779879600-1779885000@live-events-ucsc.pantheonsite.io
SUMMARY:CSE Colloquium - Learning to Image: Computational Microscopy for Dynamic Systems
DESCRIPTION:Presenter: Laura Waller\, UC Berkeley \nAbstract: \nComputational imaging jointly designs hardware and algorithms to push beyond the classical limits of imaging\, enabling measurement of new quantities (e.g. 3D\, phase\, and super-resolution) with simple\, inexpensive hardware. These approaches have already transformed consumer photography; our goal is to achieve a similar transformation in scientific microscopy. \nIn this talk\, I will show how end-to-end learning is reshaping the design of imaging systems\, from programmable illumination with LED arrays to compact\, lensless cameras built from Scotch tape. By combining physical models with neural networks\, we can jointly learn how to capture data\, reconstruct images\, and self-calibrate systems that would otherwise be too complex to model. However\, many computational methods rely on multiple measurements\, limiting their use for live\, dynamic samples. I will introduce new space-time algorithms based on implicit neural representations (INRs) that jointly recover structure and motion\, correct artifacts\, and enable high-resolution imaging in regimes where traditional approaches fail. \nBio: \nLaura Waller is the Charles A. Desoer Professor of Electrical Engineering and Computer Sciences at UC Berkeley. She received B.S.\, M.Eng. and Ph.D. degrees from the Massachusetts Institute of Technology in 2004\, 2005 and 2010. After that\, she was a Postdoctoral Researcher and Lecturer of Physics at Princeton University from 2010-2012. She is a Packard Fellow for Science & Engineering\, Moore Foundation Data-driven Investigator\, OSA Fellow\, and Chan-Zuckerberg Biohub Investigator. She has received the Carol D. Soc Distinguished Graduate Mentoring Award\, OSA Adolph Lomb Medal\, the SPIE Early Career Award and the Max Planck-Humboldt Medal. \nHosted by: Professor Alvaro Cardenas \nLocation: Engineering 2\, Room E2-180 (Refreshments such as fruit\, pastries\, coffee\, and tea will be provided.) \nZoom Option: https://ucsc.zoom.us/j/93445911992?pwd=YkJ2TQtF79h0PcNXbEcpZLbpK0coiY.1&jst=3
URL:https://live-events-ucsc.pantheonsite.io/event/cse-colloquium-learning-to-image-computational-microscopy-for-dynamic-systems/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Lectures & Presentations,Seminars
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/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:20260527T090000
DTEND;TZID=America/Los_Angeles:20260527T110000
DTSTAMP:20260518T163634Z
CREATED:20260518T163634Z
LAST-MODIFIED:20260518T163634Z
UID:10014652-1779872400-1779879600@live-events-ucsc.pantheonsite.io
SUMMARY:Tu\, H. (CSE) - From Evaluation to Adaptation: Building Reliable Multimodal Intelligence
DESCRIPTION:Multimodal large language models (MLLMs) are rapidly becoming general-purpose AI systems\, yet their capabilities are advancing faster than our ability to evaluate\, improve\, and validate their reliability in realistic use. Standard benchmarks mainly measure in-distribution final-answer accuracy\, leaving critical gaps in safety\, robustness\, fine-grained reasoning evaluation\, and reliability in real-world agentic settings. My research proposes an evaluation-to-adaptation framework for building reliable multimodal intelligence: developing rigorous evaluations that expose failures beyond conventional benchmarks\, learning feedback models that guide inference-time reasoning\, and studying how multimodal systems can adapt through experience. We instantiate this agenda through two completed works and two proposed directions. Unicorn evaluates safety and robustness under out-of-distribution and adversarial conditions\, revealing substantial vulnerabilities across 22 vision-language models. ViLBench studies vision-language process reward modeling as both an evaluation challenge and a mechanism for inference-time improvement\, showing that process-guided reasoning selection can improve reliability. Building on these foundations\, we further study test-time experience accumulation and explore reliable multimodal agents for GUI and computer-use tasks. Together\, my research aims to move beyond capability-driven progress alone\, toward multimodal AI systems whose reliability can be evaluated\, improved\, and tested in realistic deployment settings. \nEvent Host: Haoqin Tu\, Ph.D. Student\, Computer Science & Engineering \nAdvisor: Cihang Xie \nZoom: 964 1355 0550 \nPasscode: zWxU8A
URL:https://live-events-ucsc.pantheonsite.io/event/tu-h-cse-from-evaluation-to-adaptation-building-reliable-multimodal-intelligence/
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/04/ph.d.-presentation-graphic-option2.jpg
LOCATION:
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Los_Angeles:20260526T103000
DTEND;TZID=America/Los_Angeles:20260526T123000
DTSTAMP:20260512T164007Z
CREATED:20260512T164007Z
LAST-MODIFIED:20260512T164007Z
UID:10014630-1779791400-1779798600@live-events-ucsc.pantheonsite.io
SUMMARY:Castro\, S. (CSE) - Agentic AI for Security: Adversarial Foundations for Autonomous Cyber Operations
DESCRIPTION:Autonomous Cyber Operations (ACO) agents promise effective security automation with minimal human intervention\, yet their deployment raises three interconnected challenges: agents must be realistic (reproducing diverse attacker sophistication)\, secure (preventing autonomy from becoming an attack surface)\, and feasible (safely replicating human behavior at full autonomy). \nWe argue that these three properties are requirements for ACO agents. Existing approaches do not address them together and lack diverse adversarial coverage\, formal threat models for attacks against the agents themselves\, and systematic evaluation of multi-agent topologies. \nWe advance all three ACO properties: (1) For realism\, we establish adversarial foundations by discovering Windows OS vulnerabilities and releasing two exploits reliable across XP through 11. (2) For security\, we formalize ACO meta-attacks and meta-defenses\, propose the first invariant-based Meta-IDS detecting both sensor and actuator meta-attacks\, and introduce the first hybrid LLM–RL ACO integration for defense with a novel inter-agent communication protocol. (3) For feasibility\, we present MaLO\, the first dynamic-topology multi-agent ACO system\, achieving a 78.6\% success rate across a new 42-task security benchmark and solving operations up to 40× faster than human experts. We further propose the Security Operation Complexity Index (SOCX) classification and the T×V×O taxonomy as the first objective-driven evaluation methodology for coding-agent attacks. \nTogether\, these contributions demonstrate that ACO agents can match real-world adversarial sophistication\, resist meta-attacks\, and outperform human operators on complex security tasks. Open challenges remain in adaptive adversaries\, LLM–RL co-training\, dynamic topology selection\, and deployment beyond simulated environments. \n  \nEvent Host:  Sebastián R. Castro\, PhD Candidate\, Computer Science & Engineering \nAdvisor: Alvaro A. Cárdenas \nZoom: https://ucsc.zoom.us/j/2267557290?pwd=S0dNTTV3emZGUzlqV3dLbTg3a0NFUT09&omn=92791061627 \nPasscode: G20c06
URL:https://live-events-ucsc.pantheonsite.io/event/castro-s-cse-agentic-ai-for-security-adversarial-foundations-for-autonomous-cyber-operations/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260526T100000
DTEND;TZID=America/Los_Angeles:20260526T110000
DTSTAMP:20260518T190031Z
CREATED:20260518T185313Z
LAST-MODIFIED:20260518T190031Z
UID:10014653-1779789600-1779793200@live-events-ucsc.pantheonsite.io
SUMMARY:Harsh\, B. (CSE) - SUPERSCALAR\, MULTIPLE TAKEN BRANCH PREDICTOR
DESCRIPTION:This work addresses improvements in branch prediction mechanism to support high perfor-\nmance processors. The state of the art aims to balance the prediction latency and prediction\naccuracy using multi level correcting predictors [27]. Prior published work focusses on scalar\ndesigns and prediction accuracy improvement for hard to predict branches employing tailor\nmade\, non generic and non transferrable solutions [8]. Recent work also proposes ahead pre-\ndiction [42–44] to solve the problem of low accuracy of L0 predictor. \nThis work proposes efficent\, generic and transferrable solutions to reduce mispredic-\ntions and to use the fetch bandwidth more efficiently. This includes a biased overriding multi-\nlevel hierarchy with three predictor levels (L0\, L1\, L2). L0 uses a High-Confidence-Only Taken\n(HOTP) predictor that only predicts high-confidence taken control-flow instructions. This work\nfurther uses L1-L2 biased training to decrease mispredictions by L2 while it trains on branches\non which L1 has reached high confidence. This work proposes a superscalar predictor built\nusing the state of the art scalar predictor. Superscalar predictor is implemented by sizing a su-\nperscalar TAGE variant (BATAGE) using Optuna-based search. with varying table sizes and\naspect ratios. The work further proposes a branch predictor frontend design (nTakenBP) to de-\nliver multiple taken branch predictions per cycle. Unlike prior work\, nTakenBP achieves this by\nextending the existing BTB and TAGE tag-comparison logic rather than deepening lookahead. \n  \nEvent Host: Bhawandeep Singh Harsh\, Ph.D. Candidate\, Computer Science & Engineering \nAdvisor: Jose Renau \nZoom: https://ucsc.zoom.us/j/4166778865?pwd=cS9NcnVjRjArYlRRcDcrY3d5N0ZKQT09
URL:https://live-events-ucsc.pantheonsite.io/event/harsh-b-cse-superscalar-multiple-taken-branch-predictor/
LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064
CATEGORIES:Ph.D. Presentations
ATTACH;FMTTYPE=image/jpeg:https://live-events-ucsc.pantheonsite.io/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:20260521T090000
DTEND;TZID=America/Los_Angeles:20260521T143000
DTSTAMP:20260326T204610Z
CREATED:20260326T204610Z
LAST-MODIFIED:20260326T204610Z
UID:10011802-1779354000-1779373800@live-events-ucsc.pantheonsite.io
SUMMARY:Annual BE Student Project Showcase
DESCRIPTION:Join Baskin Engineering for our annual Student Project Showcase to celebrate the innovative work and accomplishments of undergraduate engineers in capstone courses and research pathways. The broader campus community\, parents\, and industry partners are invited to view the culmination of student work. \nThe day begins with oral presentations from nominated “best-in-class” teams and those working on industry-sponsored projects. Following this\, all students will participate in a comprehensive Poster Session featuring project outcomes with some teams including table-top demonstrations of functional hardware. \nEvent Details: \n\nDate: May 21\, 2026\nOral Presentations (Nominated/Industry Teams): 9:00 AM to 11:00 AM\, Engineering 2\, Room 180\nPoster Session (All Student Teams): 11:30 AM to 2:30 PM\, Engineering Courtyard
URL:https://live-events-ucsc.pantheonsite.io/event/be-student-project-showcase-2026/
CATEGORIES:Lectures & Presentations,Undergraduate
ATTACH;FMTTYPE=image/png:https://live-events-ucsc.pantheonsite.io/wp-content/uploads/2026/03/BE-ug-project-showcase.png
END:VEVENT
END:VCALENDAR