BEGIN:VCALENDAR PRODID:-//Trumba Corporation//Trumba Calendar Services 0.11.25916//EN VERSION:2.0 CALSCALE:GREGORIAN X-WR-CALNAME:CMU School of Computer Science Events X-WR-TIMEZONE:America/New_York METHOD:PUBLISH BEGIN:VEVENT SUMMARY:Welcome HONDA DTSTART:20260720T123000Z DTEND:20260720T210000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:By Invitat ion Only X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science DTSTAMP:20260715T162857Z DESCRIPTION:Meetings and participation by invitation only. UID:https://www.cs.cmu.edu/calendar/205688278 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205688278 END:VEVENT BEGIN:VEVENT SUMMARY:Pittsburgh Quantum Institute Workshop DTSTART:20260720T130000Z DTEND:20260721T200000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Wean Hall 7500 / 7316 X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website and Registration X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.pqi.org/workshop-efficient-high-frequency-all-electrical-control-spi n-dynamics-quantum-magnetic-topological DTSTAMP:20260626T145011Z DESCRIPTION:This workshop brings together leading researchers in condens ed matter physics\, materials science\, electrical engineering\, and ult rafast spectroscopy to explore emerging approaches for the efficient\, h igh-frequency\, all-electrical control of magnetic order in quantum mate rials.\n\nA central focus of the meeting is the development of new mecha nisms for controlling spin dynamics without charge transport or Joule he ating\, particularly through adiabatic spin-orbit torques in intrinsic m agnetic topological insulators. These advances open a fundamentally new paradigm for spin-charge coupling and create exciting opportunities for ultrafast\, energy-efficient spintronic technologies.\n\nThe workshop wi ll foster interdisciplinary discussions on electrically driven magnetic dynamics\, symmetry and topology in quantum materials\, high-frequency a nd THz spin phenomena\, nanoscale device integration\, and advanced expe rimental probes including THz spectroscopy and time-resolved magneto-opt ics. By connecting theory\, materials development\, device engineering\, and advanced characterization\, the meeting aims to define key scientif ic challenges and catalyze future collaborations in this rapidly evolvin g field.\n\nChairs: Enrique del Barco (UCF) and Simranjeet Singh (CMU)\n \nThis workshop is sponsored by the Army Research Laboratory\, the Pitts burgh Quantum Institute\, The University of Central Florida\, The Univer sity of South Florida\, and the Center for Quantum Phenomena at New York University.\n\nREGISTER: \nhttps://panthercentralpitt.wufoo.com/forms/p qi-workshop-july-20th-21st-2026/ UID:https://www.cs.cmu.edu/calendar/205026480 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205026480 END:VEVENT BEGIN:VEVENT SUMMARY:Combinatorics at the Confluence DTSTART:20260720T133000Z DTEND:20260722T163000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Carnegie M ellon Campus X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science\, Department of Mathematical Sciences X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website and Registration X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// combcon.github.io/ DTSTAMP:20260717T085138Z DESCRIPTION:CMU will host the International Congress of Mathematicians ( ICM) satellite conference Combinatorics at the Confluence\, which will t ake place July 20-22\, 2026\, immediately preceding the ICM. The confere nce will feature an outstanding lineup of speakers spanning the breadth of combinatorics\, as well as the Institute for Computer-Aided Reasoning in Mathematics (ICARM: https://icarm.io/)-sponsored events on the inter face of AI and combinatorics.\n\nREGISTER: https://combcon.github.io/\n→ Please register by May 15 to apply for travel funding\, a poster presen tation\, or to receive the discounted rate. UID:https://www.cs.cmu.edu/calendar/201707118 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D201707118 END:VEVENT BEGIN:VEVENT SUMMARY:Computer Science Ph.D. Thesis Oral - YIGE HONG DTSTART:20260720T140000Z DTEND:20260720T153000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Traffic21 Classroom\, Gates Hillman 6501 and Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// csd.cmu.edu/calendar/2026-07-20/doctoral-thesis-oral-defense-yige-hong DTSTAMP:20260713T150858Z DESCRIPTION:Decision making in large stochastic systems is a central and challenging problem across computer systems\, machine learning\, and op erations research. This thesis studies a wide variety of such systems th at\, despite their diverse problem definitions\, share similar underlyin g structures and admit similar fundamental techniques for performance an alysis and policy design. We identify two such structures\, each groupin g a family of seemingly distinct problems. The two structures share a co mmon spirit: an intractable system can be approximated by a simpler\, we ll-understood one.\n\nThe first part studies the weak-coupling structure \, where a system can be approximated by a collection of independent\, l ow-dimensional subsystems. We design a control policy on this simple pro xy and convert it into a near-optimal policy for the original\, coupled system. Applied to stochastic bin packing\, restless bandits\, and weakl y-coupled Markov decision processes (WCMDPs)\, this yields efficiently c omputable policies that are provably near-optimal at scale\, under subst antially weaker and more easily verifiable conditions than were previous ly required.\n\nThe second part studies the one-dimensional structure\, where a key quantity of the system can be approximated by a one-dimensio nal process even when the full state is high- or infinite-dimensional. A pplying this idea to multiserver queues\, we prove new universal bounds for the G/G/n queue\, show that the Gittins policy is near-optimal for G /G/n queues with setup times\, and determine the best achievable delay t ogether with a near-optimal policy for the multiserver-job model.\n\nThe sis Committee\nWeina Wang (Chair)\nMor Harchol-Balter\nAlan Scheller-Wol f\nJim Dai (Cornell University)\nYudong Chen (University of Wisconsin-Ma dison)\nQiaomin Xie (University of Wisconsin-Madison)\n\nIn Person and Z oom: \nhttps://cmu.zoom.us/j/93646101737?pwd=4WkHrvtyowssxErYFFcOrJzL44d 5Hb.1 Participation. See announcement. UID:https://www.cs.cmu.edu/calendar/205609381 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205609381 END:VEVENT BEGIN:VEVENT SUMMARY:Human-Computer Interaction Ph.D. Thesis Defense - Luke Guerdan DTSTART:20260720T140000Z DTEND:20260720T153000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Newell-Sim on 3305 and Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science DTSTAMP:20260708T085838Z DESCRIPTION:As artificial intelligence (AI) systems are introduced in a range of socially consequential settings\, measuring their capabilities \, risks\, and limitations has become crucial. Yet many properties of AI systems—such as the "helpfulness" of a chatbot response\, or the "fairne ss" of a predictive algorithm—are unobservable latent constructs. Obtain ing valid\, reliable\, and informative measurements of such constructs i s challenging in practice\, given both their contested nature and the of ten limited resources available to measure them effectively. In response \, this thesis introduces the concept of a measurement intervention: a t argeted approach designed to help organizations improve the validity of AI system performance measurements under practical constraints.\n\nThe f irst part of this thesis advances measurement interventions in the conte xt of predictive modeling for algorithmic decision support (ADS). I prop ose a conceptual framework that characterizes threats to the validity of predictive models used in ADS. I then report findings from qualitative interviews with data scientists\, which illustrate the nuanced trade-off s they make while balancing the validity of a predictive modeling formul ation with competing desiderata\, such as its resource requirements and portability across institutional contexts. Based on these findings\, I d evise two statistical approaches—"expert anchors" and "uncertainty cance llation"—that help teams more effectively repurpose existing organizatio nal data for new measurement tasks.\n\nThe second part of this thesis ad vances measurement interventions for Generative AI (GenAI) evaluation. I first conduct a formative study with sixteen industry AI practitioners \, to examine their current practices for assessing the validity of GenAI evaluations. I find that\, while Subject Matter Experts (SMEs) often ha ve a rich understanding of latent constructs being measured in GenAI out puts\, AI practitioners currently lack mechanisms for effectively incorp orating this expertise into the design of GenAI measurement instruments. \n\nBased on these findings\, I introduce RubricStudio: an interactive s ystem that supports cross-functional teams of SMEs and AI practitioners in operationalizing validity principles while designing GenAI evaluation s. RubricStudio automatically flags potential validity issues in evaluat ion results\, and provides scaffolding designed to help SMEs and AI prac titioners diagnose and resolve issues together through synchronous co-re finement sessions. I evaluate RubricStudio through a controlled between- subjects study\, which shows that RubricStudio's validity scaffolding he lps teams make more targeted diagnoses of validity-related issues\, and creates opportunities for SMEs to directly shape the course of the evalu ation design process.\n\nFinally\, I propose two statistical measurement interventions\, designed to help teams scale up GenAI evaluations via a n LLM-as-a-judge—i.e.\, a second GenAI system that acts as an automated evaluator. The first intervention proposes a multi-label rating elicitat ion\, aggregation\, and performance measurement framework for validating LLM-as-a-judge systems on subjective measurement tasks\, where more tha n one rating may be reasonably considered "correct." The second leverage s a doubly-robust estimation approach to yield informative evaluation re sults when human ratings are scarce but LLM-as-a-judge ratings cheap and abundant. Based on these overall findings\, I conclude by reflecting on the opportunities and challenges for measurement interventions to advan ce the evaluation of future human-AI systems.\n\nThesis Committee\nKen H olstein (Co-Chair)\nZhiwei Steven Wu (Co-Chair)\nJeffrey Bigham\nAlexand ra Chouldechova (Abridge)\n\nAdditional Information: \nhttps://drive.goo gle.com/file/d/1EBnEOCBjEQ0SFiLyWyqzqDrVDPTjRmMS/view?usp=sharing\n\nIn Person and Zoom: \nhttps://cmu.zoom.us/j/97135016726?pwd=QYmeIq4TNDXqyoa HmVoCrjaFIMLjrf.1&jst=2 Participation. See announcement. UID:https://www.cs.cmu.edu/calendar/205415048 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205415048 END:VEVENT BEGIN:VEVENT SUMMARY:Vision and Autonomous Systems Seminar DTSTART:20260720T193000Z DTEND:20260720T203000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Newell-Sim on 4305 X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.ri.cmu.edu/event/cutting-the-skip-training-residual-free-transformer s/ DTSTAMP:20260714T114013Z DESCRIPTION:Transformers are ubiquitous. They influence nearly every asp ect of modern AI. However\, the mechanics of their training remain poorl y understood. This poses a problem for the field due to the immense amou nts of data\, computational power\, and energy being invested in the tra ining of these networks. I highlight a recent intriguing empirical resul t from our group. Specifically\, although self-attention catastrophicall y fails to train without a skip connection under standard conditions\, d eep transformers can in fact be trained successfully without them. In th is talk\, I explore what makes this possible and what it reveals about t he fundamental training dynamics of modern transformers. I also speculat e on why truly deep networks may be important for improving generalizati on and efficiency. \n\n—\nSimon Lucey: \nhttps://simon-lucey.github.io /index.html Ph.D. is the Director of the Australian Institute for Machin e Learning (AIML) and a professor in the School of Computer Science\, at Adelaide University. He is also Director of the CommBank Foundational A I Research Centre. Prior to this he was an associate research professor at Carnegie Mellon University's Robotics Institute (RI) in Pittsburgh US A\; where he spent over 10 years as an academic. He was also Principal R esearch Scientist at the autonomous vehicle company Argo AI from 2017-20 22. He has received various career awards\, notably the AmCham AI Scient ist of the year in 2024. He was also a member of the Australian Governme nt’s AI Expert Group\, and their National Robotics Strategy committee. S imon’s research interests span AI\, machine learning\, computer vision a nd robotics. \n\nThe VASC seminar is generously sponsored by HeyGen: h ttps://www.heygen.com UID:https://www.cs.cmu.edu/calendar/205645839 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205645839 END:VEVENT BEGIN:VEVENT SUMMARY:Master of Science in Robotics Thesis Presentation - Shantanu Jai swal DTSTART:20260720T200000Z DTEND:20260720T213000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Blelloch-S kees Conference Room\, Gates Hillman 8115 and Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.ri.cmu.edu/event/towards-smarter-and-safer-self-improving-ai/ DTSTAMP:20260716T113250Z DESCRIPTION:As AI systems become more capable\, further progress may dep end not only on scaling models and training data\, but also on enabling systems to evaluate and improve their own behavior and development. This raises a dual challenge: how can we make self-improvement more effectiv e while ensuring increasingly autonomous systems remain trustworthy?\nTh is thesis investigates self-improvement across three complementary direc tions: \n\n\n - Improve individual outputs. We develop iterative refinem ent methods for compositional visual generation\, enabling models to pro gressively refine their outputs using feedback from vision-language mode l critics. We study how different forms of test-time scaling -- depth\, breadth\, and hybrid strategies -- trade off accuracy\, quality\, and co mputational cost. \n - Improve the research process. We extend the feedb ack loop from individual generations to automated experimentation. Using LLM agents for machine-learning and robotic policy-learning tasks\, we investigate whether 'autoresearch' agents can propose improvements\, run experiments\, learn from their outcomes\, and accumulate experience tha t transfers across tasks. \n - Improve safety and oversight. As agents b ecome increasingly autonomous within self-improvement loops\, they may l earn to mislead evaluators in pursuit of their objectives. We investigat e lying in LLMs\, identify internal mechanisms and representations assoc iated with deception\, and evaluate interventions to mitigate it. \nToge ther\, these directions frame self-improvement as a feedback loop involv ing generation\, evaluation\, revision\, and learning. The thesis presen ts work toward making such loops more capable and trustworthy as they sc ale toward increasingly autonomous scientific discovery and recursive AI development\, as envisioned in the AI 2027 scenario: https://ai-2027.co m/.\n\nThesis Committee\nDeepak Pathak (Advisor)\nShubham Tulsiani\nMihi r Prabhudesai\n\nIn Person and Zoom: \nhttps://cmu.zoom.us/j/94511426003 ?pwd=bAQQpKylKtJ6lfolzflkea6yi0GVLu.1 Participation. See announcement. UID:https://www.cs.cmu.edu/calendar/205718090 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205718090 END:VEVENT BEGIN:VEVENT SUMMARY:ARM Institute Tech Day DTSTART:20260721T120000Z DTEND:20260721T200000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Mill 19\, 4501 Lytle Street - Building A X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website and Registration X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.eventcreate.com/e/techday27 DTSTAMP:20260520T091257Z DESCRIPTION:Interested in participating in ARM Institute technology proj ects? \nJoin in for the member-exclusive Tech Day: \nhttps://us.list-man age.com/RNnnJBBAbvn?e=77bf4269e2&c2id=516f5374bc7d478fae48e25d8fe4b80e e vent taking place fully in-person on July 21 at the Pittsburgh headquart ers.\n\nThis Tech Day will support their forthcoming 27-01 Core Technolo gy Project Call — expected to release in the June timeframe.\n\nTech Day events are open only to ARM Members and\, as a benefit of membership\, to help their member organizations craft competitive responses to their Project Call. \n\nThe detailed agenda is forthcoming\, but what to expe ct:\n\n - Project teaming and networking activities \n - Presentations b y subject matter experts on the Project Call's Special Topic Areas\n - O pportunities to get private feedback on your project ideas as a benefit of your membership\n - Dedicated time to ask questions and learn more ab out the strategy behind our Project Call\nREGISTER: \nhttps://www.eventc reate.com/e/techday27 | Learn More: \nhttps://www.eventcreate.com/e/te chday27\nNote: No virtual attendance option. UID:https://www.cs.cmu.edu/calendar/203946202 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D203946202 END:VEVENT BEGIN:VEVENT SUMMARY:Master of Science in Robotics Thesis Presentation - Jacob Thomps on DTSTART:20260721T163000Z DTEND:20260721T174500Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Newell-Sim on 3305 and Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.ri.cmu.edu/event/evaluating-world-models-in-embodied-question-answer ing-through-computational-primitives-and-difficulty-progressions/ DTSTAMP:20260716T120058Z DESCRIPTION:Language modeling progress is largely evidenced by steadily rising scores on benchmarks of increasing apparent difficulty. From this \, the field infers increasingly general capabilities\, many of which pr esuppose robust world modeling. Interpreting a score\, however\, require s understanding both the task's computational requirements and how the t est-taker generalizes from them. Unlike humans\, who demonstrably genera lize well\, large language models (LLMs) often do not: they instead lear n heuristics fit to the minimal sufficient computational requirements of a task\, which may be far simpler than the task appears to demand. This talk extends this analysis to embodiment\, where multimodal LLMs (MLLMs ) serve as the perception and reasoning core of embodied agents\, whose reliable deployment depends on benchmarked capability including a robust world model of the agent's environment.\n\nWe evaluate world model robu stness by independently manipulating object count\, duplication\, trajec tory length\, and viewpoint change in a controlled synthetic benchmark o f multi-frame egocentric trajectories\, and find that frontier MLLMs deg rade sharply as difficulty increases\, while humans remain near ceiling. We then characterize embodied question answering (EQA) task demands thr ough three nested paradigms: selecting relevant observations\, clusterin g nearby observations into local spatial-semantic units\, and propagatin g semantic state across experience. Stratifying questions by the weakest sufficient paradigm yields a difficulty progression. Prominent EQA benc hmarks predominantly test only the weakest paradigm\, single-frame selec tion\, and so we introduce Campus-Bench\, three multi-hour\, campus-scal e episode histories with questions stratified into selection and propaga tion regimes\, directly comparing model performance on the two over the same episodes. We additionally develop a diagnostic method in which an M LLM incrementally constructs and traverses a hierarchical spatial-semant ic memory\, executing propagation explicitly. A frontier long-context ML LM performs strongly on selection but collapses on propagation\; the sam e model leveraging our method dramatically improves it while remaining c ompetitive on selection\, suggesting models struggle to maintain state i nternally.\n\nFrom this\, we argue that frontier MLLMs do not yet mainta in the robust world models their benchmark scores suggest\, and that the field needs to make progress in formalizing tasks’ computational requir ements and evaluating along their progressions of difficulty.\n\nThesis Committee\nYonatan Bisk (Advisor)\nWennie Tabib\nHaochen Zhang\n\nIn Per son and Zoom: \nhttps://cmu.zoom.us/j/91625695469?pwd=ufHvaoLs1EPxgEQccC KDSb5ddlcwBf.1 Participation. See announcement. UID:https://www.cs.cmu.edu/calendar/205719045 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205719045 END:VEVENT BEGIN:VEVENT SUMMARY:Machine Learning Ph.D. Thesis Defense - Amrith Setlur DTSTART:20260722T170000Z DTEND:20260722T181500Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Gates Hill man 8102 and Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science DTSTAMP:20260716T115238Z DESCRIPTION:Traditional AI systems rely on static predictors that execut e a fixed set of operations at test time\, such as neural networks. But many of the hardest problems we care about require test-time adaptation: models must perform variable computation at test time\, adapting what t hey do to the specific instance in order to make progress on open-ended tasks. For example\, given a conjecture in math research\, a model may n eed to decompose the problem into lemmas\, reason about intermediate cla ims\, revise failed proof attempts\, and allocate more computation to pr omising search directions. Enabling this behavior requires rethinking ho w we train models: rather than learning static predictors\, we must trai n models to function as effective algorithms that improve their performa nce by scaling computation at test time. This is now possible because pr e-trained models\, such as large language models (LLMs)\, come with powe rful priors that make such adaptation feasible.\n\nIn this talk\, I intr oduce a framework that casts algorithm learning for test-time adaptation as meta-reinforcement learning\, yielding a principled basis for both f ormal analysis and the practical design of objectives and methods that t rain models to adapt efficiently at test time. I conclude by showing how these ideas enable us to train a 4B LLM for theorem proving that outper forms models 30-50× larger by learning to scale test-time computation.\n \n\nThesis Committee\nVirginia Smith (Chair)\nRuslan Salakhutdinov\nSerg ey Levine (University of California\, Berkeley)\nYejin Choi (Stanford Un iversity)\n\nAdditional Information: \nhttps://drive.google.com/file/d/1 YXrCge_0K9enCWz8KjTSks5l03UNJ3eW/view?usp=sharing\n\nIn Person and Zoom: \nhttps://cmu.zoom.us/j/91060368068?pwd=5RXpMYWlliLGNqaht1mxwm47yqlHta. 1 Participation. See announcement. UID:https://www.cs.cmu.edu/calendar/205714616 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205714616 END:VEVENT BEGIN:VEVENT SUMMARY:Master of Science in Robotics Thesis Presentation - Nate Ludlow DTSTART:20260722T170000Z DTEND:20260722T181500Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Newell-Sim on 4305 and Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.ri.cmu.edu/event/knowledge-graph-augmented-reinforcement-learning-in jecting-structured-task-knowledge-into-arbitrary-policy-architectures/ DTSTAMP:20260716T115417Z DESCRIPTION:Reinforcement learning agents in complex tasks often require extensive exploration of large state spaces before useful structure eme rges. Instead of pure exploration for learning tasks\, it is possible to leverage high-level semantic knowledge such as recipes\, instructions\, or labels\, and adapt to new tasks by grounding that prior knowledge in the environment. This talk presents Knowledge Graph-Augmented Reinforce ment Learning (KG-RL)\, a method that augments RL policies with structur ed graph knowledge and injects it into a variety of policy networks acro ss a variety of environments.\n\nThe method takes the form of a plug-in adapter. A task knowledge graph is merged each step with a scene graph\, encoded by a Graph Convolutional Recurrent Network\, and pooled through a small recommender into a fixed-width feature vector concatenated with the policy backbone's observation features. The backbone itself is unto uched\, so the adapter slots into learned policies as varied as CNN-MLP \, SoftMoE-LSTM\, GTrXL\, and PoliFormer without modification\; bringing it to a new environment requires only enumerating a handful of entities and relation templates. The knowledge graphs are instantiated against si mulator-exposed tables in symbolic environments\, or mined by a one-time pre-pass of the same perception pipeline that builds per-step scene gra phs in open-world scenarios.\n\nWe evaluate the adapter on four environm ents (Overcooked-AI\, MiniGrid\, Craftax\, and AI-Habitat ObjectNav) and four backbones. It delivers two improvements independent of backbone ch oice: faster training to the same final policy\, reaching the same asymp totic reward in up to 3x fewer environment steps\, and a higher final po licy under a fixed budget\, where on Craftax and AI-Habitat the method o vertakes the available published baselines. Both gains scale with task c omplexity\, and the adapter is robust to substantial knowledge-graph cor ruption\, retaining its advantage with half of the graph's nodes removed . These results argue that structured task knowledge belongs as a defaul t\, low-cost input channel in modern reinforcement learning.\n\nThesis C ommittee\nKatia Sycara (Advisor)\nChangliu Liu\nRenos Zabounidis\n\nIn P erson and Zoom: \nhttps://cmu.zoom.us/j/99561509043?pwd=xUCFbxJ0Z61DCEOj srh1Mzt97v3H08.1 Participation. See announcement. UID:https://www.cs.cmu.edu/calendar/205719026 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205719026 END:VEVENT BEGIN:VEVENT SUMMARY:Factory Reset - What’s Next in Robotics\, AI\, and Advanced Manu facturing DTSTART:20260722T213000Z DTEND:20260723T003000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:On site - Long Beach\, CA X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :SCS Careers X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.cs.cmu.edu/scs-careers/career-events DTSTAMP:20260708T090407Z DESCRIPTION:What's Next in Robotics\, AI\, and Advanced Manufacturing\n \nJoin the Wells Fargo Technology Banking Group and UP Labs for a convers ation with operators\, founders\, buyers\, and investors shaping the fut ure of robotics\, physical AI\, and advanced manufacturing - alongside a select group of PhD and entrepreneurship students from local engineerin g schools.\n​\nDiscussion Topics include:\n - ​How companies win their f irst enterprise contracts and how it differs for industrial/physical app lications\n - ​What industrial and corporate buyers look for - and what they avoid\n - ​The toughest part of scaling from prototype to productio n\n - ​Where investors and strategic partners see opportunity in the mar ket today\n​Moderator:\n - ​Jess Carter - GM of Industrial Manufacturing \, UP.Labs\n​Panelists:\n - ​Jordan Black - CEO\, Senra Systems\n - ​Coo per Keller - COO\, Divergent\n - ​Richard Preece - Director of Global Pr ocurement\, CAT\n​Agenda [PDT]\n - ​5:30-6:00 pm — Hors d'oeuvres + drin ks\n - ​6:00-6:10 pm — Opening remarks from Wells Fargo Tech Bank + Inve stment Bank\n - ​6:10-7:00 pm — Panel & brief Q&A\n - ​7:00-8:30 pm — Ne tworking\n\nREGISTER: https://luma.com/ngb66pm1 ⇢ RSVP requested before July 17 | Inquiries\nLocation Provided Upon Registration UID:https://www.cs.cmu.edu/calendar/205162112 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205162112 END:VEVENT BEGIN:VEVENT SUMMARY:Computer Science Ph.D. Thesis Proposal - William He DTSTART:20260723T160000Z DTEND:20260723T173000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Reddy Conf erence Room\, Gates Hillman 4405 X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// csd.cmu.edu/calendar/2026-07-23/doctoral-thesis-proposal-william-he DTSTAMP:20260716T152850Z DESCRIPTION:This thesis proposal is about designing efficient algorithms for quantum statistical problems. I will discuss two previous works of mine on testing/learning quantum states with realistic measurements and other efficiency constraints.\n - In a paper with Meghal Gupta and Ryan O'Donnell\, I gave the first polynomial-in-n sample complexity algorithm for certifying any pure n-qubit quantum state using only single-qubit m easurements. Our algorithm\, which uses only O(n/ε) copies of the input state for ε soundness in fidelity\, answered an open question of Huang\, Preskill\, and Soleimanifar\, who had shown that a Haar-random state co uld with high probability be certified with a comparable number of copie s and single-qubit measurements.\n - In a paper with Sabee Grewal\, Megh al Gupta\, Aniruddha Sen\, and Mihir Singhal\, I gave the first algorith m that for learning an n-qubit pure state to fidelity at least 1-ε using near-optimal runtime Õ(2^n/ε) (and sample complexity) while also using only Pauli measurements. Previous algorithms with optimal sample complex ity required both time Õ(8^n/ε) and entangled measurements.\nI will also propose some directions for improvements on these two works.\n\n - Impr ove the tolerance parameter of the result from [GHO25] from O(1/n) to Ω( 1).\n - Give an algorithm with similar efficiency guarantees as [GGHSS26 ] that learn 2η-close (in fidelity) hypotheses for states that are η-clo se to pure with runtime Õ(2^n/η).\n - Give a time-optimal algorithm for mixed state tomography by combining the pure state reduction from Peleca nos\, Spilecki\, Tang\, and Wright with the pure state tomography algori thm from [GGHSS26].\nThesis Committee\nRyan O'Donnell (Chair)\nAayush Ja in\nJason Li\nHsin-Yuan Huang (California Institute of Technology / Orat omic)\nAdditional Information: \nhttps://drive.google.com/file/d/1loR_8W RziB1Pg2fbseOP_xRsTvaSqVvm/view?usp=sharing UID:https://www.cs.cmu.edu/calendar/205724570 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205724570 END:VEVENT BEGIN:VEVENT SUMMARY:Master of Science in Robotics Thesis Presentation - William Here d DTSTART:20260723T160000Z DTEND:20260723T171500Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Newell-Sim on 4305 and Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.ri.cmu.edu/event/a-robotic-system-for-tree-nursery-automation/ DTSTAMP:20260716T124009Z DESCRIPTION:The United States Green Industry faces a persistent labor sh ortage that motivates the adoption of agricultural automation. However \, existing systems are not designed for the unstructured\, densely plant ed environment of a tree nursery. \n\nThis thesis presents a robotic sys tem intended to alleviate this shortage while remaining usable by non-te chnical farmers\, built around a map-based representation of the nursery environment. A custom robotic platform\, the mini-Amiga\, and an accomp anying LiDAR-camera-IMU sensor rig were developed to satisfy the maneuve rability and payload requirements of tight nursery inter-row spacing. Po int cloud maps constructed with this platform\, using the GLIM LiDAR-ine rtial SLAM framework augmented with a custom GNSS georeferencing extensi on\, were processed with a new constrained Gaussian Mixture Model algori thm to segment individual trees without requiring trunk visibility or la rge annotated training datasets. \n\nThe resulting per-tree map was furt her augmented with photographic colorization and encoded as a hierarchic ally organized Universal Scene Description (USD) scene\, supporting non- destructive\, multi-mode visualization and per-tree metadata storage int ended for intuitive interaction by non-technical operators\, and was use d to derive a Nav2-compatible occupancy grid and row-traversal paths int ended for autonomous task execution. These results demonstrate that indi vidual nursery trees can be accurately and efficiently segmented from po int cloud data\, and that the resulting map can be represented in a form suited to both non-technical human interaction and autonomous navigatio n\, providing a practical foundation for future work integrating localiz ation and autonomous task execution to fully realize the labor-saving po tential of this system.\n\nThesis Committee\nGeorge Kantor (Advisor)\nMi chael Kaess\nEaston Potokar\n\nIn Person and Zoom: \nhttps://cmu.zoom.us /j/92136011809?pwd=Lp24obI3gWP7PVfRtu1tLQbYHVmyuA.1 Participation. See announcement. UID:https://www.cs.cmu.edu/calendar/205719275 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205719275 END:VEVENT BEGIN:VEVENT SUMMARY:SDI / Parallel Data Laboratory Talk - Kimberly Keeton DTSTART:20260723T160000Z DTEND:20260723T170000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Remote Acc ess - Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science\, ECE X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Series Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// pdl.cmu.edu/talk-series/index.shtml DTSTAMP:20260717T125553Z DESCRIPTION:Virtual memory\, invented in the 1960s\, has been tremendous ly successful and has served us well in the last 60 years. However\, 60 years in computing is ancient history\, and many early assumptions have fundamentally changed or vanished\, and new challenges have arisen. Thus \, in 2025\, it’s worthwhile to reexamine how memory systems have change d and how memory management should adapt to these changes. In this talk \, we examine several trends and their implications on memory systems. Fi rst\, the migration of computation to data centers and the cloud has cha nged the question that memory management must answer\, from how to minim ize cache misses in a server’s fixed-size memory to how to minimize memo ry usage while satisfying application performance targets to facilitate cluster-wide management. This shift requires a rethinking of historical algorithms and evaluation methods. Second\, the increasing cost of DRAM has led to tiered memory systems\, prompting questions about how these s ystems should be architected\, managed and evaluated. Finally\, the incr easing heterogeneity of compute and memory technologies means that memor ies are no longer strict hierarchies\, leading to questions about how th ese resource pools should be managed.\n\n—\n\nDr. Kimberly Keeton: \nhtt ps://techsysinfra.google/research/srg/srg-staff/kim-keeton/ is a Princip al Software Engineer in the SystemsResearch@Google group. Her recent res earch focuses on memory tiering and efficiency as well as systems suppor t for AI applications. Prior to joining Google\, she was a Distinguished Technologist at Hewlett Packard Labs\, where she investigated how to im prove the manageability\, dependability and usability of large-scale sto rage and information systems\, and how these systems can exploit emergin g technologies like persistent memory to improve functionality and perfo rmance. Her work has led to numerous publications and granted patents\, garnered multiple awards and contributed to multiple products. Kim recei ved her PhD and MS in Computer Science from the University of California at Berkeley and her BS in Computer Engineering and Engineering and Publ ic Policy from Carnegie Mellon. She is a Fellow of the ACM and the IEEE \, and has served as program chair for SOSP\, OSDI\, EuroSys\, SIGMETRICS \, FAST and DSN Performance and Dependability Symposium. She acts as an industrial advisor to university research groups at Carnegie Mellon\, th e University of California at Berkeley and the University of Texas at Au stin. In her spare time\, she sings with the Grammy-nominated chorus\, Pacific Edge Voices.\n\nZoom Participation. See announcement (includes registration details). UID:https://www.cs.cmu.edu/calendar/205753572 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205753572 END:VEVENT BEGIN:VEVENT SUMMARY:Innovation Works CAFÉ IW DTSTART:20260724T140000Z DTEND:20260724T160000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Tech Forge \, 113 47th Street\, Pittsburgh X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science\, Innovation Works X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :SCS Careers X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.cs.cmu.edu/scs-careers/career-events DTSTAMP:20260707T104043Z DESCRIPTION:Building a successful company outside traditional tech hubs comes with unique advantages and challenges. Join Innovation Works for C afé IW\, a candid conversation exploring how Pittsburgh and Nashville ar e competing in today's innovation economy.\n\nWhile much of the startup conversation focuses on Silicon Valley\, New York\, and Boston\, emergin g innovation hubs across the country are proving that high-growth compan ies can be built anywhere. This session brings together founders\, inves tors\, and ecosystem builders to examine how regional startup communitie s create advantages and where founders must work harder to compete on a national level.\n\nYou'll hear honest insights on accessing capital outs ide traditional venture hubs\, recruiting top talent in regional markets \, building national visibility\, and turning geography into a competiti ve advantage. Whether you're launching your first company or scaling an existing venture\, this interactive conversation will help you understan d how to build where you are.\n\nLearn More: \nhttps://innovationworks-4 9632253.hs-sites.com/2026-iw-event-cafe-iw-nashville-july-24 | REGISTER: \nhttps://innovationworks-49632253.hs-sites.com/2026-iw-event-cafe-iw-n ashville-july-24 ⇢ no fee for registered attendees\n\nCafe IW is designe d for founders\, investors\, and ecosystem builders interested in unders tanding how regional startup communities create competitive advantages. Whether you're launching your first company or scaling an existing ventu re\, you'll gain insights into building beyond traditional tech hubs. UID:https://www.cs.cmu.edu/calendar/205380688 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205380688 END:VEVENT BEGIN:VEVENT SUMMARY:CMU LearnLab Summer School DTSTART:20260727T130000Z DTEND:20260727T210000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Gates HIll man X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website and Registration X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// hcii.cmu.edu/news/2026-learnlab-summer-school DTSTAMP:20260701T154102Z DESCRIPTION:This intensive\, one-week program gives participants hands-o n experience building advanced educational technologies and tutoring sys tems to solve real-world learning challenges. Unlike many other summert ime programs\, this event is not for high school students. This professi onal initiative is designed for researchers\, educators\, graduate stude nts and technology developers who want to bridge the gap between the sci ence of learning and practical classroom technology. \n\nLearn More: \nh ttps://hcii.cmu.edu/news/2026-learnlab-summer-school | Questions UID:https://www.cs.cmu.edu/calendar/205165042 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205165042 END:VEVENT BEGIN:VEVENT SUMMARY:Software Engineering Ph.D. Thesis Defense - Luís Filipe Fernande s Gomes DTSTART:20260727T140000Z DTEND:20260727T153000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:TCS Hall 3 60 and Zoom X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science DTSTAMP:20260716T154722Z DESCRIPTION:Software development is inherently multimodal: developers ex ternalize intent with sketches\, diagrams\, and interactive walkthroughs alongside code. Modern AI assistants remain text-centric\, forcing visu al reasoning into linear prompts and limiting alignment with human menta l models. My thesis argues that AI code assistants should treat informal visual artifacts as first-class\, co-evolving representations\, and it advances this claim across three processes: code generation\, code expla nation\, and code evolution. First\, I study visual code assistants that generate code from informal sketches\, showing how multimodal inputs en able more natural expression of intent and introducing LLM-as-a-Judge ev aluation tailored to open-ended programming tasks. Second\, I present Vi sDocSketcher and JupyterDraw\, systems to generate informal visual expla nations from code\, and AutoSketchEval\, a reference-free round-trip met ric that evaluates sketch quality by reconstructing code from the genera ted diagram. In a within-subjects user study with 26 developers\, Jupyte rDraw's visual overviews raised the detection of silent flaws in noteboo ks from 29% to 67% of review sessions. Third\, I introduce Visual Loop\, a co-evolution framework that maintains fine-grained correspondences be tween code and sketches\, enabling localized updates in both directions. Together\, these contributions establish a foundation for multimodal hu man-AI collaboration in software engineering\, demonstrating how visual artifacts can improve mental model alignment\, documentation quality\, a nd iterative development workflows. Remaining planned evaluations assess the practical impact of sustained visual-textual interaction loops.\n\n Thesis Committee\nJonathan Aldrich (Co-Chair\, Carnegie Mellon Universit y)\nRui Abreu (Co-Chair\, University of Porto / Meta)\nVincent Hellendoo rn (Co-Chair\, Carnegie Mellon University / Google)\nBodgan Vasilescu\nD avid Lo (Singapore Management University) \nJoão Saraiva (University of Minho)\n\nAdditional Information: \nhttps://drive.google.com/drive/u/0/f olders/1eoYnhzJzR7W5JhHpqhNRJrLpR0pZyEU9\n\nIn Person and Zoom: \nhttps: //cmu.zoom.us/j/92598246315?pwd=PKDAshVeqbAEuwSBLhbjkZYSwbAPb3.1&jst=2 P articipation. See announcement. UID:https://www.cs.cmu.edu/calendar/205725642 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205725642 END:VEVENT BEGIN:VEVENT SUMMARY:Master of Science in Computational Biology Thesis Presentation - Carmen Sagnier Sanmartin DTSTART:20260729T153000Z DTEND:20260729T170000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Blelloch-S kees Conference Room\, Gates Hillman 8115 X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science DTSTAMP:20260720T161845Z DESCRIPTION:Aptamers offer a chemically synthesized\, low-cost alternati ve to antibodies for protein detection\, but predicting how single-nucle otide variation affects binding affinity remains poorly understood. This thesis develops methods for characterizing how sequence variation affec ts aptamer binding kinetics of interferon-gamma (IFN-γ) aptamers\, using the ARTIST platform\, which converts aptamer binding into real-time flu orescence. First\, it establishes a real-time kinetic characterization a pproach\, screening a systematically designed library of single nucleoti de polymorphisms (SNPs) and extracting per-variant binding parameters by fitting an ordinary differential equations (ODE) model to fluorescence data. Rather than finding a single positional rule\, this analysis shows that a mutation’s effect on binding affinity depends on the specific st ructural context of its parent sequence\, with certain mutation types pr oducing more consistent effects than others. Second\, this thesis evalua tes whether machine learning can replace this per-variant ODE fitting pr ocess\, using a physics-informed neural network (PINN) trained to infer kinetic parameters directly from raw experimental data. Across several a rchitectures\, this effort consistently failed to recover physically mea ningful parameters\, a failure that traces to the same root cause found independently during ODE fitting. The fluorescence signal alone does not carry enough information to constrain all of the current model’s underl ying kinetic parameters. Together\, these contributions indicate that th e barrier to further characterizing this system lies in a structural lim itation of the current ARTIST ODE system itself\, pointing toward future readout strategies that could overcome this constraint.\n\nThesis Commi ttee \nJose Lugo-Martinez (Chair)\nChristian Cuba-Samaniego\nJoshua Kang as UID:https://www.cs.cmu.edu/calendar/205850671 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205850671 END:VEVENT BEGIN:VEVENT SUMMARY:Research Experience for Undergraduates in Software Engineering ( REUSE) and Security and Privacy Undergraduate Research (SPUR) Scholars: Project Presentations DTSTART:20260729T183000Z DTEND:20260729T200000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:TCS Hall X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science DTSTAMP:20260720T134207Z DESCRIPTION:The Research Experience for Undergraduates in Software Engin eering (REUSE: \nhttps://www.cmu.edu/scs/s3d/reuse/) and Security and Pr ivacy Undergraduate Research (SPUR: \nhttps://www.cmu.edu/scs/s3d/reuse/ spur/index.html) Scholars programs are ending. Twenty-five students have spent the summer engaged on great research projects in areas spanning a ll of computer science. Come learn about their work.\n\nPlease join in as your schedule permits. \n\n - Khalid Alamri — From Creation to Detec tion: A Systematic Evaluation of Gen-AI Image Generation and Detection\n - Shaden Almodhy — A Visual IDE for Agentic Cybersecurity Systems\n - R afayel Amirkhanyan — BorrowSanitizer: The Next Generation of Rust (Un)Sa fety\n - Delia Brown — Analyzing Public Interest in In-Person Survey Exp eriences and their Impact on Data Quality\n - Melinda Chang — Program an alysis of AI-enabled Telegram bots at scale\n - Ethan Chen — Synthesizin g Visual Specifications\n - Elizabeth Cooney — Human Browser Agent Robus tness\n - Jennifer Forsyth — Evaluation of the Take9 Cybersecurity Aware ness Campaign\n - Ryan Hung — Optimizing Runtime Aliasing Violation Dete ction Through Reference-Counted Garbage Collection in BorrowSanitizer\n - Dustin Juliano — Type Safe Live Updates In Meerkat\n - Allison Kee — U senet: Uncharted Waters\n - Sally Lee — Automating Geometric Proof Feedb ack\n - Sean Lim and Eva Li — A Taxonomy of Design Patterns in Deployed Privacy-Preserving Systems\n - Tammy Pham — AI Chatbot Design to Support Neurodivergent Students learning Program\n - Matthew Rakauskas — Pygrat e2: A Semantics-aware Python 2 to 3 Migration Tool\n - Anika Sharma — To wards non-invasive blood flow measurement\n - Karan Pratap Singh — Unit Testing Framework for ZoKrates\n - Emma Sudo — Automatically Eliminating Instability in SMT-based Program Verification\n - Hazel Torek — Informa tion Borrowing\n - Quyen Tran — Automating Lightweight Proof Generation and Evaluation for Interactive High-School Geometry Learning\n - Nathan Vaz — Reviving Pixnapping Attacks on Android\n - Anh Vu — Adaptive hint generation for better scaffolding in CTF-based cybersecurity education\n - James Waters — Abstraction of LLM and MLLM Jailbreaking\n - Michael W u and Gordon Jin — Beyond Traditional Documentation: Using Invariants to Improve Software Documentation\n - Leah Zhang — Projector Phone\nFacult y Host: Joshua Sunshine UID:https://www.cs.cmu.edu/calendar/205846441 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205846441 END:VEVENT BEGIN:VEVENT SUMMARY:Summer Term - Last Day of Classes DTSTART:20260730T120000Z DTEND:20260730T210000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science DTSTAMP:20260716T120935Z UID:https://www.cs.cmu.edu/calendar/205719078 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205719078 END:VEVENT BEGIN:VEVENT SUMMARY:Human-Computer Interaction Institute - Summer Undergraduate Rese arch Program (REU) Poster Session DTSTART:20260730T200000Z DTEND:20260730T210000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:3rd Floor Perlis Atrium\, Newell-Simon Hall X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// hcii.cmu.edu/summer-research-program DTSTAMP:20260630T165301Z DESCRIPTION:Join us at this end-of-summer event to celebrate the past 10 weeks with our HCII Summer Undergraduate Research Program: \nhttps://hc ii.cmu.edu/summer-research-program (REU) students. The students have spe nt the summer with us working across a range of research projects. Stop by during this poster session to talk to the students. Learn more abou t their research and bring your questions!\n\nSCS Community welcomed. UID:https://www.cs.cmu.edu/calendar/205165019 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205165019 END:VEVENT BEGIN:VEVENT SUMMARY:Microsoft Bay Area Summer Networking Event - SCS Students DTSTART;VALUE=DATE:20260731 DTEND;VALUE=DATE:20260801 X-MICROSOFT-CDO-ALLDAYEVENT:TRUE LOCATION:Date to be locked down this summer X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :SCS Careers X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// www.cs.cmu.edu/scs-careers/career-events DTSTAMP:20260617T132748Z DESCRIPTION:SCS Students →If you will be interning in the Bay Area this July or August and are working in Data\, Training (pre/mid/post)\, Multi modal\, Systems/Infrastructure\, or Responsible AI/Safety\, you are invi ted to express interest in an invite‑only summer networking event with p eers in similar technical focus areas.\n\nIf you’d like to be considered for an invitation\, please complete this short form: \nBay Area Summer Networking Event (Date TBD): \nhttps://forms.office.com/Pages/ResponsePa ge.aspx?id=v4j5cvGGr0GRqy180BHbR4uEikV5JRNPuKCuQKvk6Y9UQTNSMEVXWFA4WVlQO VBYTlFHMURaMVg0MC4u UID:https://www.cs.cmu.edu/calendar/204735180 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D204735180 END:VEVENT BEGIN:VEVENT SUMMARY:Welcome FUJITSU DTSTART:20260803T123000Z DTEND:20260803T210000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:By Invitat ion Only X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:Sch ool of Computer Science DTSTAMP:20260615T131532Z DESCRIPTION:Meetings and participation by invitation only. UID:https://www.cs.cmu.edu/calendar/204678505 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D204678505 END:VEVENT BEGIN:VEVENT SUMMARY:Hack the Impossible (October 24) DTSTART:20260803T220000Z DTEND:20260803T220000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:APPLICATIO N DEADLINE: 3 August 2026 @ 6:00 pm (PST) DTSTAMP:20260708T161423Z DESCRIPTION:You want ownership\, mentorship\, and problems that don’t ha ve answers yet. You want to build for real customers—fast.\n\nWhat this is\nAn invite-only hackathon for mission-driven outliers. You’ll work on record-breaking challenges at the intersection of autonomy\, CV/ML\, ro botics\, sensor fusion\, distributed systems\, and real-time controls—wi th mentors who’ve shipped at the frontier.\n\nTravel (hotel\, flight\, a nd airport transfers) is sponsored to San Francisco.\n\nWho’s hiring\n\n 1. A stealth AI startup protecting safety and freedom worldwide. Immedia te roles open in Silicon Valley and Münich building:\n\n - Breakthrough technology to safeguard lives with resilient\, low-cost systems that can counter the growing risks from aerial threats\; and\n - an AI-powered a erial system.\n2. A design for manufacturing foundational model startup founded by Sebastian Thrun: \nhttps://en.wikipedia.org/wiki/Sebastian_Th run (Stanford CS professor\; founded Google X\, Waymo\, Google Brain) in San Francisco.\n\nWhy join\n - Ownership & autonomy from day one\;\n - Mentorship from engineers who built Stanley (DARPA Grand Challenge auton omous vehicle\n - winner) and eVTOL/Kitty Hawk lineage\; and\n - Immedia te impact at fast growing early-stage startups with real customers.\nWho should apply\n - BS/MS/PhD builders (or equivalent) in SW/ML/HW/Robotic s\;\n - You thrive in ambiguity\, execute fast\, and want to ship the “i mpossible”\; and\n - You’re open to full-time in-person in 2027. No remo te\, part-time\, or internships.\nHow to get in ⇢ Apply by 3 August 2026 @ 6:00 pm PST (California Time)\n\n - Submit your resume/portfolio/GitH ub here: \nhttps://orion-talent-management.breezy.hr/p/660aa7454ac0-sf-h ackathon-october-2026\; and a\n - Cover Letter re: an instance you shipp ed under brutal constraints (200 words).\nQUESTIONS ?\n\nSpace is limite d. Comfort zones are not invited. UID:https://www.cs.cmu.edu/calendar/205423097 URL;TYPE=URI:https://www.cs.cmu.edu/calendar/?trumbaEmbed=view%3Devent% 26eventid%3D205423097 END:VEVENT BEGIN:VEVENT SUMMARY:VentureBridge Webinar: Why Most Startups Can't Explain What The y Do DTSTART:20260807T170000Z DTEND:20260807T180000Z X-MICROSOFT-CDO-ALLDAYEVENT:FALSE LOCATION:Webinar fo r VentureBridge and CMU Startup Founders X-TRUMBA-CUSTOMFIELD;NAME="Organization(s)";ID=58018;TYPE=SingleLine:SCS \, Swartz Center for Entrepreneurship X-TRUMBA-CUSTOMFIELD;NAME="Event Website Title";ID=52347;TYPE=SingleLine :Event Website and Register X-TRUMBA-CUSTOMFIELD;NAME="Event Website URL";ID=52348;TYPE=Url:https:// luma.com/r1xus0bw DTSTAMP:20260717T113403Z DESCRIPTION:Most startups struggle to explain what they do\, both to cus tomers and to their own teams. This session breaks down the two language s every founder needs: external messaging that converts customers\, and internal language that aligns your team and guides product decisions.\n \n​Adam Chen\, Chief Revenue Officer at FinStrat Management\, will walk t hrough how to build a specific\, actionable Ideal Customer Profile\, cra ft a value proposition that answers "why should I choose you\," and writ e a behavioral statement that defines the exact behavior change your pro duct is trying to create. You'll leave with practical templates for all three\, plus a validation playbook for pressure-testing your messaging w ith real customers before you scale it.\n\nREGISTER: https://luma.com/r1 xus0bw\n\nThis event is organized in partnership between the CMU Swartz Center for Entrepreneurship\, CMU VentureBridge\, CMU Alumni Association \, and FinStrat Management.\n\nFinStrat Management : \nhttps://www.cmu.e du/swartz-​This event is organized in partnership between the CMU Swartz Center for Entrepreneurship\, CMU VentureBridge\, CMU Alumni Associatio n\, and FinStrat Management.​