BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Events - ECPv6.15.20//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:20240310T100000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0700 TZOFFSETTO:-0800 TZNAME:PST DTSTART:20241103T090000 END:STANDARD 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:20251121T123000 DTEND;TZID=America/Los_Angeles:20251121T140000 DTSTAMP:20260417T115121 CREATED:20251118T163526Z LAST-MODIFIED:20251118T163526Z UID:10005179-1763728200-1763733600@events.ucsc.edu SUMMARY:Ramollari\, H. (ECE) - An Optofluidic Spectrometer and Applications in Biosensing DESCRIPTION:Miniaturized spectrometers have the potential to replace bulky and expensive benchtop models. We have previously demonstrated a multimode interference (MMI) waveguide-based spectrometer that achieves high performance while minimizing its footprint. \nIn this talk\, the integration of the MMI spectrometer into an optofluidic device is proposed. This integration opens up applications such as the detection of single particle fluorescence spectra and absorption spectra. \nMoreover\, adding a metasurface to the spectrometer waveguide is expected to enhance the sensitivity of single particle detection and simplify the analysis methods. \nFinally\, to improve the MMI waveguide spectrometer a new nanophotonic platform is proposed. \nEvent Host: Helio Ramollari\, Ph.D. Student\, Electrical Engineering  \nAdvisor: Holger Schmidt  \nZoom- https://ucsc.zoom.us/j/99623652977?pwd=j2hy77fV9jdGuEzI0iGa5JVAa35W1b.1 \nPasscode- 576057 URL:https://events.ucsc.edu/event/ramollari-h-ece-an-optofluidic-spectrometer-and-applications-in-biosensing/ 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/2025/11/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:20251121T140000 DTEND;TZID=America/Los_Angeles:20251121T160000 DTSTAMP:20260417T115121 CREATED:20251021T182427Z LAST-MODIFIED:20251022T181942Z UID:10004960-1763733600-1763740800@events.ucsc.edu SUMMARY:Torres\, S. (ECE) - An Integrated Platform for Real-time Monitoring and Support of 3D Tissue Growth DESCRIPTION:Organoids are three-dimensional tissue cultures that model real organs and serve as valuable tools for studying development\, disease\, and treatment response. Traditional methods\, which rely on manual handling and incubators\, limit consistency and real-time monitoring. To address these issues\, we developed a modular microfluidic platform that integrates automated feeding\, live fluorescence imaging\, and environmental control without the need for a standard incubator. The core of the system is a vertically oriented PDMS-glass chip that enables precise media delivery and continuous imaging of small 3D structures such as organoids. Using fluorescent dyes to mimic molecules\, such as nutrients or drugs\, we tracked their movement through tissue in real time without invasive sensors. This setup maintains metabolic stability and provides detailed insight into molecular transport\, which improves applications in disease modeling\, drug testing\, and personalized medicine. \n  \nEvent Host- Sebastián Torres\, Ph.D. Candidate\, Electrical & Computer Engineering  \nAdvisor: Mircea Teodorescu \n  \nZoom- https://ucsc.zoom.us/j/2333595627?pwd=aWtwL3V2QnFTMkNDSWowZnRNS0xSQT09 \nPasscode- 579836 URL:https://events.ucsc.edu/event/torres-s-ece-an-integrated-platform-for-real-time-monitoring-and-support-of-3d-tissue-growth/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2025/10/option-3-1.png END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20251124T093000 DTEND;TZID=America/Los_Angeles:20251124T113000 DTSTAMP:20260417T115121 CREATED:20251112T181924Z LAST-MODIFIED:20251112T181924Z UID:10005132-1763976600-1763983800@events.ucsc.edu SUMMARY:Chen\, Q. (CSE) - New Approximation and Online Algorithms using Novel Combinatorial Structures DESCRIPTION:Most optimization problems face the challenge of computing an optimum solution requiring superpolynomial time. In particular\, they are classified as NP-hard problems that have no polynomial-time algorithm to date. Instead\, computer scientists turn to find an approximate solution and create numerous elegant algorithms. However\, in the modern era\, computational environments have changed drastically\, and we are not able to afford to design new algorithms for each new problem via repeated trial and error. Therefore\, systematic ways to understand the possibilities and limitations of these problems are desired. This dissertation studies several central combinatorial optimization problems\, focusing on understanding the key structural obstacles and developing unified frameworks. Mainly\, we study two types of combinatorial optimization problems:\n(1) Scheduling. The problem is associated with limited resources\, and our target is to find an allocation method to complete all jobs over time that minimizes the overall budget cost.\n(2) Network Design. Different from scheduling problems. In this problem\, we aim to find a minimum-cost topological network that supports routing for demanding communications. \nOur first work is focused on a group-to-group survivable network design problem that generalizes the classic point-to-point network to support routing between any pair of subsets of nodes. Previous research stops at limited faults\, and the difficulty comes from the way to compress the graph into a tree. We propose a new framework via capacitated tree embeddings against arbitrary faults in the network\, which gives the first polylogarithmic approximation algorithm. Further\, this framework captures nearly all the recent models proposed in the area. \nIn contrast to the offline optimization problems mentioned above\, online algorithms are natural adaptations that have been found in tremendous real applications. In online algorithms\, the algorithm wants to compete against arbitrary uncertainty\, which means the instance is unknown at first and revealed over time. We study various scheduling problems and focus on some important metrics – average flow time\, which measures the average time a job stays in the system from its arrival to completion. Real-world demands give online scheduling problems enormously different settings. Computer scientists need to repeat errors and trials to find a provably good solution. We find the key required combinatorial property is supermodularity for the residual objective\, which measures the average completion time for all alive jobs assuming they have the same arrival time. Further\, we relate supermodularity with gross-substitute/linear-substitute (GS/LS)\, which is a well-studied definition in economics. Finally\, we propose a meta-algorithm that solves all captured problems in one shot. \nEvent Host: Qingyun Chen\, Ph.D. Student\, Computer Science and Engineering \nAdvisor: Sungjin Im \nZoom-  https://ucsc.zoom.us/j/94376536164?pwd=cPloEcyKuQg1C9reIbuh6rejrOaRfR.1 URL:https://events.ucsc.edu/event/chen-q-cse-new-approximation-and-online-algorithms-using-novel-combinatorial-structures/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2025/11/ph.d.-presentation-graphic-option-1.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20251205T090000 DTEND;TZID=America/Los_Angeles:20251205T110000 DTSTAMP:20260417T115121 CREATED:20251118T165217Z LAST-MODIFIED:20251119T192149Z UID:10005180-1764925200-1764932400@events.ucsc.edu SUMMARY:Littschwager\, N. (CSE) - A Proposal for Characterizing Replicated Systems and Emulators DESCRIPTION:Simulation is a coinductive proof technique to assert the behavioral equivalence of computing systems that has seen fruitful application in distributed systems\, concurrent process calculi\, and programming languages\, since the 1970’s. We have also utilized simulation in our prior work\, where we formalized and proved a folklore claim that the state-based and operation-based approaches to Conflict-free Replicated Data Types (CRDTs) are ‘equivalent’ since they can ‘emulate each other’. More specifically\, a CRDT system consists of a collection of nodes called replicas. Clients interact with individual replicas by querying or updating their state\, and replicas interact by message passing over a network to eventually reach a convergent state. There are two main approaches to implementing a CRDT: operation-based\, and state-based. We showed that the main state-based and operation-based approaches to CRDTs do indeed ‘emulate each other’ since one can exhibit a pair of weak simulations between the original type of CRDT\, and its corresponding translation into the other type. We then leveraged the existence of these weak simulations to formally prove a ‘representation independence’ result\, in the sense that when access to the CRDTs is mediated by an imperative programming language\, the programmer cannot discern the underlying CRDT implementation by producing a program that terminates when run using one type of CRDT implementation\, but not when run with the other. \n Unfortunately\, our results are impractical for the purpose of being reapplied to asserting the equivalence of other replicated systems\, since the simulation relations (that one needs to exhibit in order to prove the necessary representation-independence) are non-modular\, requiring the user to reason about the potential executions of their entire replicated system. Additionally\, we observed that behavioral equivalence of state-based and operation-based CRDTs is a specific instance of the more general paradigm of ‘emulation’\, which is the process by which an ‘emulator’ translates the behavior of one system into the behavior of a different system. \nWe propose to generalize the techniques of our prior work to be applicable for any pair of replicated    systems\, and correct the ‘non-modularity’ issue by decomposing the overall proof structure into compositional simulation proofs about the local behavior of a replica\, and the behavior of the communication medium. Our second proposal comes from the observation that\, to our knowledge\, ‘emulation’ has not been given a formal and general mathematical semantic model that adequately captures the practical nuances faced by researchers and practitioners working on emulators. With that in mind\, we propose a notion of a faithful emulator\, inspired by the concept of a faithful functor 𝐹 ∶ C → D which lets us regard objects in C as ‘the same as’ the objects in D\, but with additional structure. \nHost: Nathan Littschwager\, Ph.D. Student\, Computer Science and Engineering  \nAdvisor: Lindsey Kuper  \n  URL:https://events.ucsc.edu/event/littschwager-n-cse-a-proposal-for-characterizing-replicated-systems-and-emulators/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2025/11/ph.d.-presentation-graphic-option2.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20251205T100000 DTEND;TZID=America/Los_Angeles:20251205T123000 DTSTAMP:20260417T115121 CREATED:20251125T212206Z LAST-MODIFIED:20251125T212206Z UID:10005646-1764928800-1764937800@events.ucsc.edu SUMMARY:DeGrendele\, C. (AM) - Learning-Augmented and Structure-Preserving Methods for Conservation Law Solvers DESCRIPTION:In this work\, we develop numerical methods for conservation laws that explore statistical\, structure-preserving\, and machine-learning-based approaches\, each built on top of traditional numerical solvers. First\, we develop a general Gaussian-process-based “recipe’’ for constructing high-order linear operators such as interpolation\, reconstruction\, and derivative approximations. Building on this recipe\, we derive a kernel-agnostic convergence theory for GP-based operators that interprets them as generalized finite-difference schemes\, defines an effective order-of-accuracy proxy that captures non-ideal truncation-error structure\, and uses this metric to select stencil geometries and kernel hyperparameters analytically. We then introduce a new second-order kernel\, Discontinuous Arcsin (DAS)\, that is stationary and prevents oscillations. DAS is integrated into a shock-capturing framework called the Multidimensional Optimal Order Detection (MOOD) method and shows an increase in efficiency by admitting less first order cascades. Next\, we address the long-standing problem of spurious pressure oscillations in compressible multi-component and real-fluid simulations by introducing a fully conservative pressure-equilibrium-preserving scheme and a high-order fully conservative approximate variant that apply to arbitrary equations of state. Unlike existing approaches\, these methods avoid non-conservative updates or EOS-specific constructions\, and on smooth interface advection tests with ideal-gas\, stiffened-gas\, and van der Waals fluids they reduce spurious pressure oscillations by orders of magnitude relative to current schemes. We then propose a hybrid numerical–machine learning framework for mixed hyperbolic–parabolic systems in which only the diffusive contribution is learned while the hyperbolic fluxes are advanced with standard shock-capturing methods\, enabling timesteps at a hyperbolic CFL. Within this framework\, we compare several neural architectures and loss designs on viscous Burgers tests and on the one-dimensional Euler equations with heat conduction\, showing that U-shaped neural operators combined with multi-step and TVD-style regularization improve long-time stability and spectral behavior\, and we analyze the resulting coupled schemes via eigenvalue-based stability diagnostics. Finally\, we apply high-order\, shock-capturing finite-difference methods within NASA’s Launch Ascent and Vehicle Aerodynamics (LAVA) framework to quantify acoustic and pressure loads on the Artemis Mobile Launcher\, including multiphase simulations of water-suppression systems and comparisons to flight data that inform hardware design for future missions. Collectively\, this work offers a set of targeted advances in kernel-based numerical operators\, conservative schemes and learning-augmented solvers each aimed at improving accuracy\, stability\, or efficiency in complex multiphysics flow simulation. \nEvent Host: Chris DeGrendele\, Ph.D. Candidate\, Applied Mathematics \nAdvisor: Dongwook Lee  \nZoom- https://ucsc.zoom.us/j/96308438100?pwd=9El4idgPoaVnAd9m8M6As6uaSbcojp.1 \nPasscode-  123456 URL:https://events.ucsc.edu/event/degrendele-c-am-learning-augmented-and-structure-preserving-methods-for-conservation-law-solvers/ 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/2025/11/ph.d.-presentation-graphic-option-1-1.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:20251205T130000 DTEND;TZID=America/Los_Angeles:20251205T140000 DTSTAMP:20260417T115121 CREATED:20251203T234430Z LAST-MODIFIED:20251203T234430Z UID:10005731-1764939600-1764943200@events.ucsc.edu SUMMARY:Garg\, S. (CSE) - MAPPING ANNOTATIONS FROM NETLIST TO SOURCE CODE DESCRIPTION:Hardware design flows have become increasingly complex as modern chips integrate billions\nof transistors and rely on aggressive synthesis optimizations to meet performance\,\narea\, and power targets. While these transformations improve circuit efficiency\, they\nalso erase the correspondence between gate-level netlists and their originating HDL\nsource lines. The loss of traceability makes post-synthesis debugging\, timing backannotation\,\nand root-cause analysis extremely difficult. Existing solutions depend on\ntool-specific metadata or preserved signal names\, which are often lost after flattening\,\nretiming\, or logic restructuring.\nTo address this long-standing problem\, this thesis presents SynAlign\, a structural\nalignment framework that restores the mapping between optimized netlists and\nsource code without relying on synthesis metadata. SynAlign treats both the reference\nRTL and synthesized designs as graphs and iteratively aligns them using shared\nstructural cues—such as sequential boundaries\, fan-in/fan-out relationships\, and partial\nnaming patterns. The algorithm employs anchor-based seeding\, multi-stage neighborhood\nmatching\, and a lightweight scoring function to propagate correspondences\nefficiently across large designs.\nExtensive evaluation demonstrates that SynAlign achieves over 90% line-level\nalignment accuracy across diverse designs\, maintaining robustness even when 60% of\nsignal names are obfuscated or removed. The framework scales linearly with design size\,\ncompleting alignment on multi-million-node circuits within minutes. Controlled tests\nconfirmed structural stability under synthetic noise\, while production-level validation\non real processor and accelerator modules verified industrial applicability.\nBy recovering structural visibility lost during synthesis\, SynAlign bridges a\ncritical gap between front-end design intent and post-synthesis implementation. Its explainable\nalignment enables faster debug cycles\, more accurate timing correlation\, and\nprovides a foundation for next-generation EDA tools that integrate traceability\, optimization\ntransparency\, and source-level introspection into the hardware development\nprocess. \nHost: Sakshi Garg\, Ph.D. Candidate\, Computer Science and Engineering  \nAdvisor: Jose Renau \nZoom- https://ucsc.zoom.us/j/96207792766?pwd=bjBfusfaucoqMGZNgayum2te4tsLc5.1 \nPasscode- 669162 URL:https://events.ucsc.edu/event/garg-s-cse-mapping-annotations-from-netlist-to-source-code/ LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064 CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2025/10/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:20251208T091500 DTEND;TZID=America/Los_Angeles:20251208T103000 DTSTAMP:20260417T115121 CREATED:20251205T173457Z LAST-MODIFIED:20251205T174005Z UID:10005749-1765185300-1765189800@events.ucsc.edu SUMMARY:Jamilan\, S. (CSE) - Profile-guided Compiler Optimizations for Data Center Workloads DESCRIPTION:Modern applications\, such as data center workloads\, have become increasingly complex. These applications primarily operate on massive datasets\, which involve large memory footprints\, irregular access patterns\, and complex control and data flows. The processor-memory speed gap\, combined with these complexities\, can lead to unexpected performance inefficiencies in these applications\, preventing them from achieving optimal performance. Considering the complexity and size of data center applications\, manually identifying and resolving performance issues is often impractical or impossible. Instead\, developing new compiler optimization techniques can be a more effective and scalable solution to boost both performance and energy efficiency. In this thesis\, we focus on identifying the root causes that limit the performance of data center workloads. We analyze the limitations of current profile-guided compiler optimization techniques for addressing these performance gaps. Finally\, we propose two profile-guided optimization techniques\, APT-GET and RIFS\, which can be integrated into the LLVM optimization pipeline to deliver further improvements. To hide the long latency of memory accesses\, we introduce APT-GET\, a profile-guided technique that ensures timely prefetches by leveraging dynamic execution-time information to build a novel analytical model that finds the optimal prefetch distance and injection site based on the collected profile. We study APT-GET across 10 real-world applications and demonstrate that it achieves a speedup of up to 1.98× and an average of 1.30×. To enable runtime value-invariant function specialization to reduce redundant operations\, we introduce RIFS\, a profile-guided compiler technique that specializes functions based on runtime-invariant call-site-specific argument values. RIFS introduces a novel value-profiling LLVM pass to identify runtime invariant arguments and a subsequent LLVM transformation pass to generate specialized function variants tailored to these value profiles. To efficiently select among potentially thousands of specialization candidates\, we develop a predictive cost model that estimates each candidate’s performance benefit before code generation. RIFS achieves an average speedup of 5.3% and an instruction reduction of 2.5% over the LLVM -O3+PGO baseline across 12 real-world applications. \nHost: Saba Jamilan\, Ph.D. Candidate\, Computer Science and Engineering  \nAdvisor: Heiner Litz  \nZoom- https://ucsc.zoom.us/j/95818759324?pwd=rdaS7G1V7O6faRhNOgFyq1OR50eSLK.1 \nPasscode- 652917 \n  URL:https://events.ucsc.edu/event/jamilan-s-cse-profile-guided-compiler-optimizations-for-data-center-workloads/ 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/2025/10/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:20251208T130000 DTEND;TZID=America/Los_Angeles:20251208T140000 DTSTAMP:20260417T115121 CREATED:20251202T163305Z LAST-MODIFIED:20251202T163305Z UID:10005718-1765198800-1765202400@events.ucsc.edu SUMMARY:de Priester\, J. (ECE) - Hybrid Reinforcement Learning DESCRIPTION:Reinforcement Learning (RL) is a machine learning paradigm that trains a decision maker\, or policy\, by learning from interaction with an environment. The power of RL lies in its ability to learn complex strategies without explicit human instruction\, which can lead to better solutions that human designers overlook in domains ranging from robotics to scientific discovery. Despite these successes\, applying RL to safety-critical control systems remains a significant challenge due to the fragility of black-box policies. Standard RL controllers are prone to “chattering” or indecisiveness\, which is rapid\, detrimental switching between decisions induced by small disturbances\, and lack formal closed-loop safety\, stability\, and robustness guarantees. Furthermore\, existing discrete and continuous-time RL paradigms struggle to model hybrid systems\, where continuous state evolution is intertwined with instantaneous discrete updates. Consequently\, standard RL approaches cannot effectively be applied to safety-critical hybrid dynamical systems\, as such approaches suffer from discretization artifacts\, computational inefficiency\, and a lack of closed-loop safety\, stability\, and robustness guarantees. \nTo bridge the gap between hybrid control theory and RL\, this research proposal is organized into four interconnected thrusts. Thrust 1 addresses the fragility of existing standard RL-based policies by designing RL algorithms to construct robust hybrid supervisors to eliminate chattering. Thrust 2 establishes the theoretical bedrock of a native hybrid RL formulation. By leveraging insights from discounted MPC\, the hybrid RL problem is formulated with intrinsic closed-loop stability\, safety\, and robustness properties. Thrust 3 extends standard RL components to the hybrid domain to create RL algorithms capable of solving the hybrid RL problem defined in Thrust 2. Finally\, Thrust 4 provides comprehensive empirical validation\, confirming the robustness of the supervisors from Thrust 1 and demonstrating the advantages of the native hybrid RL formulation developed in Thrusts 2 and 3 over a standard RL formulation. \nHost: Jan de Priester\, Ph.D. Student\, Electrical and Computer Engineering  \nAdvisor: Ricardo Sanfelice \nZoom- https://ucsc.zoom.us/j/95229790206?pwd=ICevzd4QdEE7ZAlYALZIYbhU2bCU4W.1 \nPasscode-  981137 URL:https://events.ucsc.edu/event/de-priester-j-ece-hybrid-reinforcement-learning/ LOCATION:Jack Baskin Engineering\, Baskin Engineering 1156 High Street\, Santa Cruz\, CA\, 95064 CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2025/10/ph.d.-presentation-graphic-option-1-1.jpg 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:20251208T130000 DTEND;TZID=America/Los_Angeles:20251208T140000 DTSTAMP:20260417T115121 CREATED:20251203T220535Z LAST-MODIFIED:20251203T220535Z UID:10005728-1765198800-1765202400@events.ucsc.edu SUMMARY:Ferdous\, N. (CSE) - SPECSIM : A Simulation Infrastructure Mitigating Transient Timing Attacks DESCRIPTION:   Transient execution attacks are serious security threats in modern-day processors. Out-of-order execution compels the processor to access data that should not be otherwise perceived. Leakage of that secret information creates a covert channel for the attacker for various types of transient and speculative attacks. Transient based execution attacks emanate when the secret information is leaked by the execution of transient instructions which are executed by the processor but never got committed from the processor pipeline. However\, on the microarchitectural level\, the effect of these transient instructions is noticeable. Generally\, microarchitectural state is the state that a processor maintains to improve performance which is transparent to software. The secret data retained in the microarchitectural state are susceptible to create a covert channel and thereby are at higher risk to be observed by the attacker for transient attacks.\nThis research work presents a robust and secure simulation infrastructure that implements multiple strategies to mitigate transient attacks in the timing domain. This work proposes various strategies e.g.\, Reorder Buffer Transient Flushing Technique in Randomized Transient Pipeline\, SpecSCB for making the speculative instructions invisible to the architectural state\, for the mitigation of the timing attack. In this work\, transient instructions are added in the proposed Randomized Transient Pipeline and are flushed effectively\, using Transient Flushing Techniques\, squashing all the transient instruction residues that could remain in the Randomized Transient Pipeline. This flushing strategy also ensures no difference in the execution time of the base simulation and the proposed Randomized Transient Simulation\, leaving no leakage for transient based timing attacks. In addition to the simulation platform\, a novel Transient Verification Framework is also proposed which consists of Global Time Signature Verification Model and Retirement Time Signature Verification Model. The transient verification framework identifies if there is any anomaly in the timing domain\, related to all existing instructions\, which could leave space for covert channel for timing attacks. Overall\, this work has provided an extensive and robust simulation platform infrastructure for the researchers to explore various types of attacks with their respective mitigating solutions. \nHost: Nilufar Ferdous\, Ph.D. Student\, Computer Science and Engineering  \nAdvisor: Jose Renau  \nZoom- https://us06web.zoom.us/j/84111701472?pwd=l3s5sQszKt35paVOWNxxLaE8jphG80.1 \nPasscode- Qi1pAk URL:https://events.ucsc.edu/event/ferdous-n-cse-specsim-a-simulation-infrastructure-mitigating-transient-timing-attacks/ 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/2025/10/ph.d.-presentation-graphic-option-1-1.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:20251208T140000 DTEND;TZID=America/Los_Angeles:20251208T150000 DTSTAMP:20260417T115121 CREATED:20251205T175704Z LAST-MODIFIED:20251205T175952Z UID:10005750-1765202400-1765206000@events.ucsc.edu SUMMARY:Wang\, Y. (CSE) - Toward Practical and Effective Large Language Model Unlearning DESCRIPTION:The growing integration of Large Language Models (LLMs) into real-world applications has heightened concerns about their trustworthiness\, as models may reveal private information\, reproduce copyrighted content\, propagate biases\, or generate harmful instructions. These risks\, alongside emerging privacy regulations\, motivate the need for LLM unlearning\, methods that remove the influence of specific data while preserving overall model capability.\nThis proposal investigates how to design practical and effective unlearning methods that enable LLMs to produce reliable and responsible outputs. We study both training-free and training-based paradigms. On the training-free side\, we introduce ECO\, which achieves unlearning via embedding-corrupted prompts detected by a lightweight classifier\, and DRAGON\, a generalizable black-box framework that combines detection with chain-of-thought guard reasoning for safe in-context intervention. On the training-based side\, we present FLAT\, a forget-data-only loss adjustment method grounded in a variational $f$-divergence formulation.\nTogether\, these approaches provide complementary strategies for aligning LLM behavior with safety and regulatory requirements while maintaining general utility. This proposal outlines their motivation\, design\, empirical performance\, and the broader research plan toward responsible and accountable LLM systems. \nHost: Yaxuan Wang\, Ph.D. Student\, Computer Science and Engineering  \nAdvisor: Yang Liu \nZoom- https://ucsc.zoom.us/j/94186242839?pwd=ubGMNF25W8gABNIl2S7EaIBHEXletV.1 \nPasscode- 786334 URL:https://events.ucsc.edu/event/wang-y-cse-toward-practical-and-effective-large-language-model-unlearning/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2025/10/ph.d.-presentation-graphic-option2.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20251209T160000 DTEND;TZID=America/Los_Angeles:20251209T170000 DTSTAMP:20260417T115121 CREATED:20251202T204536Z LAST-MODIFIED:20251209T182652Z UID:10005719-1765296000-1765299600@events.ucsc.edu SUMMARY:Zhu\, R. (ECE) - From Neuromorphic Principles to Efficient Neural Language Architectures DESCRIPTION:While Large Language Models exhibit remarkable capabilities\, their reliance on the standard Transformer architecture imposes prohibitive computational costs and quadratic memory complexity. To bridge the gap between biological efficiency and high-performance AI\, we have established foundational work in linearizing attention and maximizing hardware utilization through architectures such as RWKV and MatMul-Free networks. Addressing the remaining bottlenecks in long-term memory consolidation and optimization stability\, we propose a research roadmap focused on “In-Place Test-Time Training” (TTT) to enable compositional memory via dynamic weight updates\, and the Muon optimizer to stabilize deep reasoning through orthogonal gradient updates. Ultimately\, this work aims to unify neuromorphic principles with scalable deep learning to enable robust performance in resource-efficient environments. \nEvent Host: Ridger Zhu\, Ph.D. Student\, Electrical and Computer Engineering  \nAdvisor: Jason Eshraghian \nZoom- https://ucsc.zoom.us/j/95241268060?pwd=WDMgDWhhSyXNh8NZpBDvgpbcMVbvUz.1 \nPasscode- 256794 URL:https://events.ucsc.edu/event/ridger-z-ece-from-neuromorphic-principles-to-efficient-neural-language-architectures/ LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064 CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2025/10/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:20251210T130000 DTEND;TZID=America/Los_Angeles:20251210T150000 DTSTAMP:20260417T115121 CREATED:20251204T161744Z LAST-MODIFIED:20251205T222136Z UID:10005732-1765371600-1765378800@events.ucsc.edu SUMMARY:Singh\, A. (ECE) - Quantum Key Distribution Using Entangled Pairs with Random Grouping DESCRIPTION:Quantum Key Distribution (QKD) provides information-theoretic security for cryptographic key establishment\, but existing protocols exhibit limited noise tolerance\, restricting their applicability in practical quantum channels with finite resources. This work introduces a QKD protocol based on entanglement swapping that significantly enhances error tolerance and key generation rates. The protocol encodes six-bit classical symbols into six-qubit entangled states organized as three Bell pairs. Key contributions include: (1) maintaining positive secrecy rates under 100% intercept-resend attacks\, unprecedented among existing protocols\, (2) proven security against collective attacks up to 29.29% quantum bit error rate (QBER)\, substantially exceeding BB84’s 11% threshold\, and (3) finite-key security analysis demonstrating viable key generation under practical block size constraints. These results establish that structured multi-qubit encoding fundamentally broadens the operational capabilities of quantum key distribution\, enabling secure communication in high-noise environments such as free-space satellite links and urban channels where conventional protocols fail. \nHost: Archana Jayprakash Singh\, Ph.D. Student\, Electrical and Computer Engineering  \nAdvisor: Zouheir Rezki  \nZoom- https://ucsc.zoom.us/j/92875779810?pwd=xIWhFkOw5WR3vyBvVhBCkd7ueJs2m2.1 \nPasscode- 530049 URL:https://events.ucsc.edu/event/singh-a-ece-quantum-key-distribution-using-entangled-pairs-with-random-grouping/ LOCATION:Jack Baskin Engineering\, Baskin Engineering 1156 High Street\, Santa Cruz\, CA\, 95064 CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2025/11/ph.d.-presentation-graphic-option2.jpg 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:20251211T090000 DTEND;TZID=America/Los_Angeles:20251211T110000 DTSTAMP:20260417T115121 CREATED:20251202T162054Z LAST-MODIFIED:20251209T161343Z UID:10005717-1765443600-1765450800@events.ucsc.edu SUMMARY:Tran\, L. (BMEB) - Polysome Shadowing: A Long-Read Sequencing Approach to Study Translation DESCRIPTION:Translation is a central and highly regulated step of gene expression\, yet there are few quantitative\, high-throughput tools to study translation. Existing methods such as sucrose gradients provide only bulk ribosome counts\, while Ribo-Seq offers positional information in the genome but destroys long-range structure and transcript expression information. Because of these limitations\, many fundamental questions about mRNA translation into protein remain difficult to assay. In this proposal\, I outline my plans to develop a novel technology\, deemed Polysome Shadowing\, that covalently marks ribosome-unprotected regions of RNA with hyperactive base editors. Because ribosomes protect ~21–30 nt regions of mRNAs\, ribosome “shadows” appear as tracts of unedited bases in long-read sequencing. In Aim 1\, I will identify ribosome shadows on single molecules by increasing editing efficiency through optimization of dual cytosine and adenosine base editors and statistical modeling. In Aim 2\, I will maximize the accuracy of information recovered from highly-edited RNAs by developing a multipass library preparation protocol to generate high-confidence reads. In Aim 3\, I will apply the tools I have already developed to examine previously difficult-to-assay paradigms of translational control in the form of viral frameshifting mechanisms. Together\, completion of these aims will build an information-rich sequencing technology capable of positioning ribosomes on intact mRNAs while preserving long-range information and establish feasibility to study nascent paradigms. \nHost: Liam Tran\, Ph.D. Student\, Biomolecular Engineering and Bioinformatics  \nAdvisor: Joshua Arribere  URL:https://events.ucsc.edu/event/tran-l-bmeb-polysome-shadowing-a-long-read-sequencing-approach-to-study-translation/ LOCATION:Biomedical Sciences Building\, 575 McLaughlin Drive CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2025/11/ph.d.-presentation-graphic-option2.jpg 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:20251211T120000 DTEND;TZID=America/Los_Angeles:20251211T140000 DTSTAMP:20260417T115121 CREATED:20251209T224244Z LAST-MODIFIED:20251209T224244Z UID:10005759-1765454400-1765461600@events.ucsc.edu SUMMARY:Chambers\, K. (BMEB) - Using Genomics and Artificial Intelligence to improve prognosis for osteosarcoma patients DESCRIPTION:Transcriptomic profiling has been transformative in pediatric oncology. Pediatric cancers arise from disrupted developmental programs. Their impaired transcriptional states reflect cell lineage infidelity\, aberrant differentiation\, and immune-microenvironment interactions distinct from those of adult tumors(Gröbner et al.\, 2018; X. Ma et al.\, 2018). Within the osteosarcoma (OS) landscape\, despite being the most common bone tumor of childhood\, it remains one of the least genomically characterized pediatric cancers. Advancements in survival for localized disease\, outcomes for metastatic or recurrent OS have remained stagnant for decades. Transcriptomics characterization of OS has facilitated the exposure of the unique chromothripsis patterns associated with the disease (Sayles et al.\, 2019; Schott et al.\, 2023). Largely\, progress in OS genomics is still limited by the lack of harmonized\, cross-study datasets accessible to researchers. I detail my contributions to OS research\, beginning with the curation of the largest publicly available and harmonized RNA-sequencing osteosarcoma dataset (Chapter 2). A continuous part of my research involved the systematic democratization\, aggregation\, harmonization\, and open sharing of pediatric cancer transcriptomic datasets within the Treehouse Childhood Cancer Initiative (Beale et al.\, 2025). This dataset provided a foundation for the analyses and discoveries presented in this dissertation. I utilize the multi-cohort and transcriptomic multi-omic public OS dataset to discover and define biologically meaningful subtypes that may explain differences in progression and treatment response (Chapter 3). Finally\, I expand these advanced computational approaches into the realm of diagnostic pathology by evaluating strategies for integrating generative AI into rare cancer classification. I leverage both general and domain-specific diffusion models alongside GPT-4o–generated pathology prompts to guide histologic image synthesis (Chapter 4). In summary\, my work advances transcriptional subtyping in OS by leveraging transcriptomic data to identify molecular subtypes of OS that could inform treatment strategies. \nHost: Krizia Chambers\, Ph.D. Candidate\, Biomolecular Engineering & Bioinformatics  \nAdvisor: Olena Vaske \nZoom- https://ucsc.zoom.us/j/93569812001?pwd=RWBuZUdQq2Yo1K4kQ75WRmP0uKjYAH.1&jst=3 \nPasscode- 915392 URL:https://events.ucsc.edu/event/chambers-k-bmeb-using-genomics-and-artificial-intelligence-to-improve-prognosis-for-osteosarcoma-patients/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2025/10/ph.d.-presentation-graphic-option2.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20251211T130000 DTEND;TZID=America/Los_Angeles:20251211T150000 DTSTAMP:20260417T115121 CREATED:20251202T232256Z LAST-MODIFIED:20251202T232256Z UID:10005722-1765458000-1765465200@events.ucsc.edu SUMMARY:Laffan\, N. (CM) - Digital Memory Tools and Their Impact On Collective Remembering DESCRIPTION:Today\, both individual and collective memories are increasingly mediated by digital platforms. Both are fundamentally enmeshed in platform ecosystems that orient around commercial imperatives very much at odds with community cohesion. The digital archive where our mediated memories are stored does not merely store information but actively inscribes it\, often privileging narratives aligned with commercial incentives rather than community cohesion. This invisibility is a problem: as we offload our personal memories onto commercial tools\, we unwittingly subject our shared past to algorithmic curation and “algo-time\,” which raises serious questions about how the use of our personal devices is quietly restructuring the way societies remember. \nDuring this presentation\, I will propose a three-pronged method of investigating and engaging in this conceptual space. All three prongs revolve around a shared question : how do the technologies that extend our personal memories affect what we remember collectively? The research first establishes a conceptual ecology around the question by tracing the lifecycle of a single image from individual capture to platform archive. Second\, it employs Research through Design (RtD) and speculative design methods to prototype tools explicitly built for collective remembrance rather than commercial extraction. Finally\, it utilizes artistic practice to “diffract” these concepts\, creating interactive installations that expose the distortions and contradictions inherent in digital memory. Together\, these projects aim to make visible the hidden dynamics that shape the memories we construct together. \nHost: Nate Laffan\, Ph.D. Student\, Computational Media  \nAdvisor: Nathan Altice  \nZoom- https://ucsc.zoom.us/j/93762016105?pwd=RBXDHnuleAECZdVghEaAz9L4KK4p1d.1 \nPasscode- 668969 URL:https://events.ucsc.edu/event/laffan-n-cm-digital-memory-tools-and-their-impact-on-collective-remembering/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2025/11/ph.d.-presentation-graphic-option2.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260122T080000 DTEND;TZID=America/Los_Angeles:20260122T170000 DTSTAMP:20260417T115121 CREATED:20260122T184550Z LAST-MODIFIED:20260122T184550Z UID:10009092-1769068800-1769101200@events.ucsc.edu SUMMARY:HSI Equity Talk DESCRIPTION:Title: Understanding the advising praxes central to student success at a four-year Hispanic-Serving Research Institution \nPresenter: Dr. Lydia Iyeczohua Zendejas \nLocation: Via Zoom (link provided via RSVP) \nAbstract: Higher education scholars increasingly recognize academic advising as a critical strategy for supporting the persistence of systemically marginalized students. Since the 1990s\, UC Santa Cruz has undergone significant growth and demographic shifts—undergraduate enrollment grew from 10\,269 in 1999 to 17\,517 in 2019\, with sharp increases in underrepresented\, first-generation\, and Hispanic students—creating both challenges and opportunities for advancing equitable outcomes. \nDr. Zendejas’s interview-based qualitative study examines how UCSC’s decentralized\, dual shared advising model shapes advisors’ ability to provide holistic\, culturally responsive advising. In this HSI equity talk\, she will share how advising structures\, practices\, and policies impact advisors’ capacity to support students\, how the current model can act as a structural barrier to collaboration\, and the advising praxis advisors identify as essential to student success\, persistence\, and retention. \nPlease complete this RSVP form if you plan to attend. The Zoom information and a calendar invitation will be sent to those who RSVP.  URL:https://events.ucsc.edu/event/hsi-equity-talk/ LOCATION:CA CATEGORIES:Lectures & Presentations,Ph.D. Presentations ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/01/Equity-Talk-Feb.-4th.png END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260123T093000 DTEND;TZID=America/Los_Angeles:20260123T110000 DTSTAMP:20260417T115121 CREATED:20260120T223725Z LAST-MODIFIED:20260120T223725Z UID:10008684-1769160600-1769166000@events.ucsc.edu SUMMARY:Sharma\, R. (CSE) - Automatically Evolving GPU Libraries for Performance Portable AI Kernels DESCRIPTION:GPUs are the workhorses of modern AI\, widely deployed and developed by many vendors including Apple\, Qualcomm\, Intel\, AMD\, and NVIDIA. While these GPUs all offer high compute potential\, programming them effectively is difficult because they differ in performance-critical features like SIMT width\, cache capacity\, and memory bandwidth\, demanding different optimization strategies. Tunable kernels address this by exposing parameters such as tiling dimensions and workgroup sizes\, enabling per-device specialization. Yet this produces static libraries: tuned once\, then frozen\, degrading as new hardware emerges. We propose automatically evolving libraries that expand their tuning knowledge as new hardware emerges\, with minimal impact on user experience. \nTo build such libraries\, we first need to understand the tuning landscape. We address this through GPU Goldmines\, a WebGPU-based framework for exhaustively collecting tuning data across diverse devices. Our tuned matrix multiplication kernels outperform an optimized baseline by 8.4x on average\, while matrix-vector kernels achieve 93% of platform bandwidth. We find that hyper-tuning for a single GPU causes 50% performance degradation on other devices\, whereas data-driven portability methods recover 88% of peak performance. These kernels are fundamental to the prefill and decode phases of LLM inference. We integrate them into llama.cpp as our evaluation platform\, where they outperform CPU and Vulkan backends. \nBuilding on this data\, we are developing Living Libraries to improve performance continuously without disrupting users. This means choosing good parameters upfront\, learning from real-world execution\, and knowing when to keep searching versus when to stop\, though hand-designed parameter spaces remain inherently bounded. To move beyond this\, we extend toward LLM-based kernel evolution\, where language models propose entirely new kernel variants\, opening a less structured but higher potential search space. \nEvent Host: Rithik Sharma\, Ph.D. Student\, Computer Science and Engineering \nAdvisor: Tyler Sorensen & Yuanchao Xu   \n  \nZoom: https://ucsc.zoom.us/j/92739836317?pwd=0ydDzimUFIoaLDUKst96dk27th4lvW.1 \nPasscode: 089560 URL:https://events.ucsc.edu/event/sharma-r-cse-automatically-evolving-gpu-libraries-for-performance-portable-ai-kernels/ LOCATION:Engineering 2\, Engineering 2 1156 High Street\, Santa Cruz\, CA\, 95064 CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/png:https://events.ucsc.edu/wp-content/uploads/2026/01/option-3-1.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:20260206T110000 DTEND;TZID=America/Los_Angeles:20260206T120000 DTSTAMP:20260417T115121 CREATED:20260127T193801Z LAST-MODIFIED:20260127T193801Z UID:10009119-1770375600-1770379200@events.ucsc.edu SUMMARY:Johnstone\, J. (AM) - The Effects of Asymmetry on Overshooting and Magnetic Pumping from Compressible Convection Zones DESCRIPTION:We present a comprehensive numerical investigation examining how vertical asymmetry in compressible convection affects overshooting and the transport of large-scale magnetic fields from convective to stably stratified regions. Using three-dimensional direct numerical simulations\, we systematically vary the superadiabaticity and stratification of a convective layer to control the vertical asymmetry of the flow and analyze its influence on overshooting depth and magnetic pumping efficiency. We extend previous work by Tobias et al. (2001) and draw guidance from the asymmetry regimes identified by John & Schumacher (2023)\, investigating whether similar asymmetric convecting regimes emerge in our overshooting model that incorporates a stably stratified region below. We find that vertical asymmetry increases significantly with stratification at a moderate\, fixed Rayleigh number\, while superadiabaticity contributes primarily through enhanced downflow velocities\, with both combined leading to increasing overshooting depths reaching approximately 0.46 − 0.7 pressure scale heights. Magnetic pumping efficiency initially increases with stratification but unexpectedly decreases at higher stratification\, despite increasing overshooting depths. We find that this behavior arises from the increasing thermal and magnetic diffusivities that result from increasing stratification at fixed Ra. When instead either holding these diffusivities constant or increasing Ra sufficiently\, we find that then both overshooting and magnetic pumping depths both decrease with increasing stratification. This behavior is explained by a change of dynamical state from one of laminar downflows to one of turbulent downflowing plumes leading to a high degree of turbulent mixing and entrainment. We thus find two distinct regimes that might be described as a microscopically diffusive regime and a turbulently diffusive one. These results suggest that\, in the highly turbulent regime expected in the Sun\, magnetic pumping efficiency may decrease with increasing stratification due to enhanced turbulent entrainment\, with important implications for solar dynamo theory and the transport of large-scale magnetic fields in the solar interior. \n  \nEvent Host: Jason Johnstone\, Ph.D. Student\, Applied Mathematics \nAdvisor: Nic Brummell \nZoom- https://ucsc.zoom.us/j/5428987373?pwd=JSmNz3ZZby5ZnVBYbSoakjjQb2qQj6.1&omn=98571815542 \nPasscode- 778899 URL:https://events.ucsc.edu/event/johnstone-j-am-the-effects-of-asymmetry-on-overshooting-and-magnetic-pumping-from-compressible-convection-zones/ 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/2026/01/ph.d.-presentation-graphic-option-1-2.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:20260206T160000 DTEND;TZID=America/Los_Angeles:20260206T180000 DTSTAMP:20260417T115121 CREATED:20260128T172826Z LAST-MODIFIED:20260128T172826Z UID:10009125-1770393600-1770400800@events.ucsc.edu SUMMARY:Yang\, J. (CSE) - Towards Controllable and Compositional Generative Vision DESCRIPTION:Diffusion-based text-to-image models can generate impressive images\, but they largely treat an image as a single\, flat output\, which makes precise editing of individual elements difficult. This proposal studies layered generative representations that align with professional editing workflows\, enabling users to manipulate foreground objects while preserving the rest of the scene. A central focus is visual effects such as shadows and reflections\, which are essential for realistic composition yet are often missing or inconsistent in current generative pipelines. This proposal outlines a research program toward controllable\, compositional image generation that supports practical\, edit-ready content creation. \nEvent Host: Jinrui Yang\, Ph.D. Student\, Computer Science and Engineering \nAdvisor: Yuyin Zhou \nZoom- https://ucsc.zoom.us/j/91510964517?pwd=NG5Urv2li9HxlcUKrybg6Z5ZtYj9e6.1 \nPasscode- 544143 URL:https://events.ucsc.edu/event/yang-j-cse-towards-controllable-and-compositional-generative-vision/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/01/ph.d.-presentation-graphic-option-1-2.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260209T130000 DTEND;TZID=America/Los_Angeles:20260209T143000 DTSTAMP:20260417T115121 CREATED:20260127T195054Z LAST-MODIFIED:20260127T195054Z UID:10009120-1770642000-1770647400@events.ucsc.edu SUMMARY:Li\, X. (CSE) - Compute-Efficient Scaling of Fully-Open Visual Encoders DESCRIPTION:Vision encoders have demonstrated significant performance gains in visual generation and multimodal reasoning. These improvements are primarily attributed to the scaling of data\, model capacity\, and compute. However\, this progress is becoming less accessible due to a lack of transparency in data curation and training recipes. In combination with the high compute requirements of foundation-scale pre-training\, these factors hinder independent reproducibility. \nIn this dissertation\, we democratize large-scale visual encoder training by developing compute-efficient\, reproducible training recipes for video encoders\, vision-language models (VLMs)\, and multimodal large language models (MLLMs). First\, we challenge the common belief that scaling necessarily requires proportionally more resources. Specifically\, we show that decoupled pre-training separates key factors such as space/time and token length\, and learns strong priors first. This design yields dramatic efficiency gains across image\, video\, and generative modeling. Next\, we address the challenge of undisclosed or inaccessible training data by releasing and systematically studying the curation of high-quality\, large-scale datasets. We demonstrate that high-quality synthetic captions at scale enable vision-language models to learn stronger visual representations\, especially when paired with training frameworks that unify contrastive and generative objectives. Lastly\, building on these findings\, we develop fully open vision encoders with complete training data\, recipes\, and checkpoints\, and show that transparency can enable rather than hinder state-of-the-art performance as an MLLMs’ visual backbone. \nTogether\, these contributions establish that openness and efficiency are mutually reinforcing\, providing a reproducible foundation for the next generation of visual intelligence. \nEvent Host: Xianhang Li\, Ph.D. Candidate\, Computer Science and Engineering \nAdvisor: Cihang Xie  \nZoom- https://ucsc.zoom.us/j/95801462664?pwd=koENnyV65jyPnkJYTbiYr1jaNsV5BE.1 \nPasscode- 782017 URL:https://events.ucsc.edu/event/li-x-cse-compute-efficient-scaling-of-fully-open-visual-encoders/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/01/ph.d.-presentation-graphic-option2-1.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260211T120000 DTEND;TZID=America/Los_Angeles:20260211T130000 DTSTAMP:20260417T115121 CREATED:20260203T232101Z LAST-MODIFIED:20260203T232101Z UID:10009136-1770811200-1770814800@events.ucsc.edu SUMMARY:Centering the Experiences of Undocumented Transfer Students at HSIs: A Brown Bag Presentation by Valeria Alonso Blanco DESCRIPTION:  \nThe Huerta Center is proud to present a brown bag presentation by Graduate Student Research Awardee Valeria Alonso Blanco. She will present on a qualitative study that explores how undocumented Latinx transfer students navigate institutional support\, belonging\, and barriers at a four-year Hispanic Serving Institution (HSI). Findings reveal gaps between institutional commitments and student realities\, and she offers actionable recommendations for more equitable\, transfer-receptive practices. URL:https://events.ucsc.edu/event/centering-the-experiences-of-undocumented-transfer-students-at-hsis-a-brown-bag-presentation-by-valeria-alonso-blanco/ LOCATION:Huerta Center Conference Room (Casa Latina)\, 641 Merrill Rd\, Santa Cruz\,\, CA\, 95064 CATEGORIES:Lectures & Presentations,Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/01/Picture_AlonsoBlanco-Valeria-J-Alonso-Blanco.jpg GEO:37.0003908;-122.0534175 X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Huerta Center Conference Room (Casa Latina) 641 Merrill Rd Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=641 Merrill Rd:geo:-122.0534175,37.0003908 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260212T163000 DTEND;TZID=America/Los_Angeles:20260212T173000 DTSTAMP:20260417T115121 CREATED:20260203T172912Z LAST-MODIFIED:20260203T173017Z UID:10009149-1770913800-1770917400@events.ucsc.edu SUMMARY:Sambamurthy\, A. (AM) - Lazy Diffusion: Resolving Spectral Collapse in Generative Models for Turbulence DESCRIPTION:Diffusion-based generative models offer a principled framework for probabilistic forecasting\, but we show they suffer from a fundamental spectral collapse when applied to turbulent flows. A Fourier-space analysis of the forward SDE reveals that the mode-wise signal-to-noise ratio decays monotonically in wavenumber for power-law spectra\, rendering high-wavenumber content indistinguishable from noise. We reinterpret the noise schedule as a spectral regularizer and introduce power-law schedules that preserve fine-scale structure deeper into diffusion time. We further propose Lazy Diffusion\, a one-step distillation method that leverages the learned score geometry to bypass long reverse trajectories and prevent high-wavenumber degradation. Applied to high-Reynolds-number 2D Kolmogorov turbulence and ocean reanalysis data\, these methods resolve spectral collapse and enable stable long-horizon autoregressive emulation. \nEvent Host: Anish Sambamurthy\, Ph.D. Student\, Applied Mathematics  \nAdvisor: Ashesh Chattopadhyay \nZoom- https://ucsc.zoom.us/j/5144530307?pwd=TllaWnNDc01tcVNpa1NNeVVIMnp5QT09 \nPasscode- 55555 URL:https://events.ucsc.edu/event/sambamurthy-a-am-lazy-diffusion-resolving-spectral-collapse-in-generative-models-for-turbulence/ 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/2026/02/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:20260220T140000 DTEND;TZID=America/Los_Angeles:20260220T160000 DTSTAMP:20260417T115121 CREATED:20260210T193542Z LAST-MODIFIED:20260210T193542Z UID:10009193-1771596000-1771603200@events.ucsc.edu SUMMARY:Fredrickson\, K. (CSE) - Practical Anonymity with Formal Resistance to Traffic Analysis DESCRIPTION:Anonymous communication systems hide who is talking to whom\, not just what is said. However\, existing systems are either vulnerable to traffic analysis attacks–attacks where adversaries observe and correlate the network traffic of users–or are forced to rely on unrealistic and unenforceable assumptions about how users behave. Worse\, existing theory lacks tools to rigorously model traffic analysis attacks\, much less inform whether if a system is secure against traffic analysis or how to design systems that are. \nWe make several contributions toward our goal of practical anonymity systems that resist traffic analysis. First\, we develop the first formal framework for describing the security of systems against traffic analysis attacks\, allowing us to quantitatively describe and compare the security of all existing works. Second\, leveraging this framework\, we develop a security definition that distinguishes between systems that are and are not susceptible to traffic analysis. We call this property input/output independence. We use this definition to prove that the dominant model of systems–synchronous systems–cannot practically provide input/output independence. We then design a new asynchronous anonymity functionality\, deferred retrieval\, that achieves input/output independence with far more flexible user assumptions and up to 3400 times less traffic overhead for the same latency compared to prior methods. Finally\, we design and implement Sparta\, a family of high-throughput\, scalable instantiations of deferred retrieval using trusted execution environments and oblivious algorithms\, yielding the first practical anonymity systems that are formally resistant to long-term traffic analysis. \nEvent Host: Kyle Fredrickson\, Ph.D. Candidate\, Computer Science and Engineering \nAdvisor: Darrell Long \nZoom – https://ucsc.zoom.us/j/98133127429?pwd=QNICsMrQa6bQUKNPo40PthZyQEQCFl.1 \nPasscode – 242206 URL:https://events.ucsc.edu/event/fredrickson-k-cse-practical-anonymity-with-formal-resistance-to-traffic-analysis/ 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/2026/02/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:20260225T090000 DTEND;TZID=America/Los_Angeles:20260225T120000 DTSTAMP:20260417T115121 CREATED:20260210T221905Z LAST-MODIFIED:20260210T221905Z UID:10009196-1772010000-1772020800@events.ucsc.edu SUMMARY:Liu\, C. (CSE) - Enabling LLM Unlearning at Inference Time by Decomposing Detection and Intervention DESCRIPTION:Machine unlearning addresses the “right to be forgotten” under GDPR and enables privacy\, copyright\, and safety compliance in large language models. Training-based unlearning can remove targeted behavior on benchmarks\, but it scales poorly\, can degrade utility\, and can fail under adversarial prompting that recovers supposedly forgotten content. This prospectus proposes inference-time behavioral unlearning: rather than modifying weights to “erase” knowledge\, we detect when a query targets forgotten content and intervene in generation so the system behaves like a model never trained on that content. We formalize this approach as Detect-Intervene Decomposition and instantiate it with three complementary methods operating at the embedding\, token\, and reasoning levels under different access capabilities. Comprehensive experiments across entity unlearning\, hazardous knowledge removal\, and copyright protection demonstrate that our methods match or exceed training-based approaches while being orders of magnitude faster and preserving model utility. As LLMs increasingly operate as services with restricted weight access\, inference-time unlearning provides the only practical path for responsible AI deployment that respects privacy\, safety\, and legal requirements. \nEvent Host: Chris Liu\, Ph.D. Student\, Computer Science and Engineering \nAdvisor: Yang Liu \nZoom – https://ucsc.zoom.us/j/94799852992?pwd=EBFQe4U2lRNro1oJ8F36bgORhT2xSv.1 \nPasscode –  242384 URL:https://events.ucsc.edu/event/liu-c-cse-enabling-llm-unlearning-at-inference-time-by-decomposing-detection-and-intervention/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/01/ph.d.-presentation-graphic-option-1-1.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260304T130000 DTEND;TZID=America/Los_Angeles:20260304T170000 DTSTAMP:20260417T115121 CREATED:20260217T222743Z LAST-MODIFIED:20260219T210101Z UID:10009243-1772629200-1772643600@events.ucsc.edu SUMMARY:Shields\, S. (CM) - Procedural\, Player-Centric Game Balancing DESCRIPTION:Game balance is a term widely used among players\, researchers\, and designers of games. It is a concept that feels vitally important to how we make and play games – but when we try to define it or implement it\, we seldom get the same definition twice. Balance appears differently to whoever is judging it\, but as researchers and designers we still must translate this element of game design into technical practice. It also is an expensive and time-consuming subject\, one that requires a constant loop of playtesting and design iteration through nearly the entirety of the game development process. \nThis work seeks to focus our understanding of balance while offering procedural methods to either increase speed or improve quality when performing balancing tasks in game design and research. It accomplishes this by offering a taxonomy of balance alongside a generic design framework that can be used to apply balancing strategies to any game context. It additionally provides a catalog of balancing methods\, allowing designers to use common patterns to apply procedural balancing to their games. Finally\, I offer three technical examples using the taxonomy and framework\, putting theoretical knowledge of balance into concrete technical systems. \nBalance ultimately helps us design games that make us feel fairness in our play. By sharpening and optimizing our understanding of the term\, we improve the games we make and open new doors in game systems design. \nEvent Host: Sam Shields\, Ph.D. Candidate\, Computational Media  \nAdvisor: Edward F. Melcer \nZoom- https://ucsc.zoom.us/j/98956788669?pwd=ao7DzYQebCeS3SJ4PsGaZeGYhYMVNI.1 \nPasscode- 713173 URL:https://events.ucsc.edu/event/shields-s-cm-procedural-player-centric-game-balancing/ LOCATION:Merrill College\, College Office\, Santa Cruz\, CA\, 95064 CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/01/ph.d.-presentation-graphic-option-1-2.jpg GEO:36.9996837;-122.0530485 X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Merrill College College Office Santa Cruz CA 95064;X-APPLE-RADIUS=500;X-TITLE=College Office:geo:-122.0530485,36.9996837 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260305T130000 DTEND;TZID=America/Los_Angeles:20260305T150000 DTSTAMP:20260417T115121 CREATED:20260217T182432Z LAST-MODIFIED:20260217T182432Z UID:10009238-1772715600-1772722800@events.ucsc.edu SUMMARY:Xu\, Y. (CSE) - Right Place\, Right Time: Accelerating Edge Computation on Modern Heterogeneous SoCs DESCRIPTION:Modern edge computing increasingly relies on heterogeneous System-on-Chip (SoC) architectures. These chips tightly integrate general-purpose CPUs with various specialized accelerators\, including GPUs\, FPGAs\, and AI accelerators\, all under a shared memory architecture. Although these shared-memory SoCs enable more efficient communication and data sharing between different processing units\, they are notoriously difficult to program and tune due to architectural diversity across vendors and asymmetric compute capabilities within each SoC. \nThis dissertation introduces Redwood and BetterTogether\, two frameworks that rethink CPU-accelerator collaboration on heterogeneous SoCs. Redwood targets a class of algorithms termed traverse–compute\, that combine irregular tree traversals with dense leaf-level computation\, e.g.\, Nearest-Neighbor Search and Barnes–Hut algorithm. \nIt addresses the efficient mapping of these algorithms onto heterogeneous systems by exploiting the architectural strengths of CPUs\, GPUs\, and FPGAs. BetterTogether extends this methodology to a different class of edge workloads\, specifically multi-stage pipelines and neural networks commonly used in computer vision tasks. Furthermore\, it introduces interference-aware analysis and scheduling techniques tailored for mobile SoCs. Finally\, to broaden the scope of heterogeneous acceleration\, we evaluated emerging domain-specific accelerators. We provide a preliminary analysis of Tensor Processing Units and Tensor Cores within the context of modern programming abstractions. \nEvent Host: Yanwen Xu\, Ph.D. Candidate\, Computer Science and Engineering \nAdvisor: Tyler Sorensen \nZoom- https://ucsc.zoom.us/j/5354629158?pwd=0CVhbwLuXDMX5fAGZd63tcfNqDWp0t.1 \nPasscode- 114514 URL:https://events.ucsc.edu/event/xu-y-cse-right-place-right-time-accelerating-edge-computation-on-modern-heterogeneous-socs/ 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/2026/02/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:20260309T070000 DTEND;TZID=America/Los_Angeles:20260309T080000 DTSTAMP:20260417T115121 CREATED:20260303T175533Z LAST-MODIFIED:20260303T175533Z UID:10009386-1773039600-1773043200@events.ucsc.edu SUMMARY:Hendawy\, M. (CM) - Autonoming Child Online Safety in the Age of AI: From Control to Digital Co-Agency Across Cultures DESCRIPTION:Children’s lives are now inextricably linked with AI-driven digital systems that shape learning\, social interaction\, and development. This has elevated child online safety to a central concern for families\, policymakers\, and educators. This makes Child online safety a wicked socio-technical problem\, emerging from the complex interplay of social norms\, platform incentives\, cultural expectations\, and rapidly evolving technologies. Dominant control-based paradigms—monitoring\, blocking\, and surveillance—undermine children’s developmental capacity\, erode family trust\, and foreclose the iterative cycles of self regulated learning necessary for digital resilience. This proposal advances digital co-agency as a new paradigm for child online safety. It reframes safety from an outcome of unilateral control to a shared\, relational practice distributed across children\, caregivers\, technologies\, and governance structures. To be effective\, digital co-agency must be grounded in a clear normative standard. I define this standard as ethical safety: protection is legitimate only when it is rights-respecting and developmentally supportive. Within this boundary\, the dissertation proposes autonoming as a design stance for AI-mediated safety systems. Autonoming systems act as developmental mentors that support children’s judgment over time through explanation\, negotiation\, and graduated support that can fade as competence grows. Autonoming is grounded in Self-Regulated Learning (SRL) as the developmental mechanism for durable safety capacity. SRL models learning as cyclical forethought (planning)\, performance (in-the-moment regulation)\, and reflection (evaluating outcomes). The dissertation adopts a socio-technical interpretivist stance and a Design Science Research orientation to produce actionable artifacts that are theoretically grounded and evaluable.. Its core methodological contribution is localization-first comparative design across Cairo and Berlin. This comparative structure helps distinguish between: localized variables (culturally specific norms regarding authority\, privacy\, risk\, norms\, expectations\, and legitimacy conditions that must be adapted to) from ethical invariants (accountability\, contestability\, proportionality that should hold across contexts). \nEvent Host: Mennatullah Hendawy\, Ph.D. Student\, Computational Media  \nAdvisor: Magy Seif El-Nasr \nZoom- https://ucsc.zoom.us/j/93831600031?pwd=hsnX574bcXVQRZa16sKbX0u7OuaMlu.1 \nPasscode-459844 URL:https://events.ucsc.edu/event/hendawy-m-cm-autonoming-child-online-safety-in-the-age-of-ai-from-control-to-digital-co-agency-across-cultures/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/02/ph.d.-presentation-graphic-option-1.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260309T100000 DTEND;TZID=America/Los_Angeles:20260309T110000 DTSTAMP:20260417T115122 CREATED:20260303T174856Z LAST-MODIFIED:20260303T174856Z UID:10009385-1773050400-1773054000@events.ucsc.edu SUMMARY:Robbins\, A. (ECE) - How to train your organoid: goal-directed learning in biological neural networks DESCRIPTION:Artificial neural networks can now learn to play games\, control robots\, generate language\, and solve complicated reasoning tasks\, yet we still lack a clear understanding of how to directly guide learning in biological neural networks. We show that brain organoids can learn to solve a fundamental control task\, balancing an inverted pendulum\, through closed-loop electrophysiology. Cortical organoids interfaced with high-density microelectrode arrays received sensory input about the pole’s angle and produced motor output through their neural activity. Training signals selected by a reinforcement learning algorithm significantly outperformed random stimulation and no-stimulation controls. Blocking glutamatergic transmission abolished the learning and washout restored it\, confirming the adaptation depends on synaptic plasticity. To support this work and future experiments\, we developed BrainDance\, an open-source framework for running reproducible biological learning experiments\, and contributed to RT-Sort\, a real-time spike sorting algorithm. This dissertation presents the tools\, experiments\, and findings from pursuing goal-directed learning in biological neural networks. BrainDance makes these experiments easy-to-create\, reproducible and shareable\, letting any lab with compatible hardware start training their own organoids. \nEvent Host: Ash Robbins\, Ph.D. Candidate\, Electrical and Computer Engineering  \nAdvisor: Mircea Teodorescu \nZoom- https://ucsc.zoom.us/j/95839863615?pwd=EmqTWPN9RRBYZRW7rcpoaT9kqacfRP.1 \nPasscode- 069118 URL:https://events.ucsc.edu/event/robbins-a-ece-how-to-train-your-organoid-goal-directed-learning-in-biological-neural-networks/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/02/ph.d.-presentation-graphic-option-1.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260309T140000 DTEND;TZID=America/Los_Angeles:20260309T160000 DTSTAMP:20260417T115122 CREATED:20260303T174216Z LAST-MODIFIED:20260303T174216Z UID:10009384-1773064800-1773072000@events.ucsc.edu SUMMARY:Harrison\, D. (CS) - Multi-Level Control in Neural Dialogue Generation: Style\, Semantics\, and Selection through Over-Generation and Ranking DESCRIPTION:End-to-end neural generation models have largely displaced the modular architectures that once gave dialogue system designers explicit control over what is said and how it is said. While these models produce fluent text\, they collapse content planning\, sentence planning\, and surface realization into a single undifferentiated decoding step\, sacrificing the controllable structure that earlier systems provided. This dissertation investigates how that structure can be recovered through the over-generate-and-rank (OGR) paradigm: generating multiple candidate outputs and selecting among them using learned or prompt-based ranking functions that jointly optimize semantic fidelity\, stylistic appropriateness\, and conversational coherence. We instantiate OGR at three levels of natural language generation for dialogue: utterance-level stylistic control\, cross-domain semantic evaluation\, and dialogue-level response selection. \nFirst\, we show that explicit conditioning mechanisms\, specifically decoder-level side constraints for personality variation and discourse contrast\, re-introduce stylistic control into neural sequence-to-sequence models without compromising semantic accuracy. Second\, we demonstrate that prompt-based learning with structured linguistic profiles achieves near-perfect personality accuracy and effectively zero slot error rate when combined with ranking\, establishing that LLM prompting with explicit pragmatic specifications can match or exceed fine-tuning for personality-conditioned generation. Third\, we develop a cross-domain semantic error rate evaluation framework that frames slot error computation as an extraction task\, using a LoRA-adapted language model to extract meaning representations from generated text and a trained ranker to select among candidate extractions\, achieving reliable evaluation across 23 topic domains without domain-specific rules. Fourth\, we build and evaluate a speaker-aware transformer response ranker for Athena\, our Alexa Prize socialbot\, demonstrating that learned ranking over heterogeneous generator pools produces significantly longer conversations and higher user ratings than heuristic rule-based selection in a live A/B study with over 6\,000 conversations. \nA unifying finding emerges across all four contributions: the pragmatic features that control personality style in generation—acknowledgements\, engagement questions\, hedges\, exclamations—are the same features that distinguish high-quality from mediocre responses in open-domain dialogue. This parallel reveals that stylistic control and response ranking are complementary mechanisms for achieving the same goal: making dialogue systems sound more natural and engaging. Together\, these results support the dissertation’s central hypothesis that over-generate-and-rank provides a general\, extensible mechanism for controllable neural language generation\, restoring explicit decision points where competing communicative objectives can be weighed. The ranking function serves a role analogous to the sentence planner in classical NLG architectures\, but operates on the outputs of modern neural and LLM-based generators. \n  \nEvent Host: Davan Harrison\, Ph.D. Candidate\, Computer Science \nAdvisor: Marilyn Walker \n  URL:https://events.ucsc.edu/event/harrison-d-cs-multi-level-control-in-neural-dialogue-generation-style-semantics-and-selection-through-over-generation-and-ranking/ LOCATION:CA CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/02/ph.d.-presentation-graphic-option2.jpg END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Los_Angeles:20260310T163000 DTEND;TZID=America/Los_Angeles:20260310T173000 DTSTAMP:20260417T115122 CREATED:20260217T184921Z LAST-MODIFIED:20260217T184921Z UID:10009240-1773160200-1773163800@events.ucsc.edu SUMMARY:Mashhadi\, N. (CSE) - Compositional\, Clinically Conditioned\, and Confound-Aware Deep Learning for Alzheimer’s Disease Neuroimaging DESCRIPTION:Alzheimer’s disease (AD) is a progressive neurodegenerative disorder and a leading cause of dementia. Neuroimaging and clinical biomarkers can reveal early disease changes\, but building reliable machine learning models is difficult because data come from different scanners and sites\, some modalities are missing\, labeled cohorts are limited\, and factors such as age and scanner/site effects can bias results. \nThis dissertation develops machine learning methods for robust\, interpretable\, and controllable analysis of AD-related neuroimaging data. First\, I introduce a modular\, graph-based framework for multimodal AD detection that treats datasets and models as nodes and directed edges that can be combined to build more complex predictors. Second\, I propose a clinically conditioned 3D VAE-GAN to synthesize brain MRI\, enhanced with diffusion-driven sampling in clinical feature space to improve realism and control\, supporting data augmentation. Third\, I present a disentangled 3D masked autoencoder (MAE) that learns separated representations for age\, pathology\, and scanner effects\, making it possible to isolate and adjust age\, pathology\, or scanner effects\, while remaining reliable across sites. \nTogether\, these contributions advance practical methods for modular prediction\, controllable image generation\, and confound-aware representation learning in neuroimaging\, with an emphasis on generalization and interpretability for clinically relevant applications. \nEvent Host: Najmeh Mashhadi\, Ph.D. Candidate\, Computer Science and Engineering \nAdvisor: Razvan Marinescu \nZoom- https://ucsc.zoom.us/j/98195204428?pwd=nyfvbmd9t81Xj5Z3yPPVtu4R58CXHq.1 \nPasscode- 688069 URL:https://events.ucsc.edu/event/mashhadi-n-cse-compositional-clinically-conditioned-and-confound-aware-deep-learning-for-alzheimers-disease-neuroimaging/ LOCATION: CATEGORIES:Ph.D. Presentations ATTACH;FMTTYPE=image/jpeg:https://events.ucsc.edu/wp-content/uploads/2026/02/ph.d.-presentation-graphic-option2.jpg END:VEVENT END:VCALENDAR