{"693194":{"#nid":"693194","#data":{"type":"event","title":"PhD Defense by Bodhisatwa Chatterjee","body":[{"value":"\u003Cp\u003E\u003Cstrong\u003ETitle\u003C\/strong\u003E: Predicting Dynamic Program Behavior with Compiler-Guided Machine Learning\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003EDate\u003C\/strong\u003E: Friday, October 16th, 2026\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003ETime\u003C\/strong\u003E: 1:00 - 3:00 PM ET\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003ELocation\u003C\/strong\u003E: Klaus 2126\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003EVirtual Link\u003C\/strong\u003E: Teams\u0026nbsp;(\u003Ca href=\u0022https:\/\/teams.microsoft.com\/meet\/293867738134584?p=GI8Tcf7EIQ4ih1fkL7\u0022 target=\u0022_blank\u0022 title=\u0022Meeting join\u0022\u003Ehttps:\/\/teams.microsoft.com\/meet\/293867738134584?p=GI8Tcf7EIQ4ih1fkL7\u003C\/a\u003E\u003C\/p\u003E\u003Cp\u003E\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; Meeting ID:\u0026nbsp;293 867 738 134 584\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; Passcode:\u0026nbsp;CN9DD3Zy)\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003EBodhisatwa Chatterjee\u003C\/strong\u003E\u003Cbr\u003EPh.D. Student\u003C\/p\u003E\u003Cp\u003ESchool of Computer Science\u003C\/p\u003E\u003Cp\u003ECollege of Computing\u003C\/p\u003E\u003Cp\u003EGeorgia Institute of Technology\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003ECommittee:\u003C\/strong\u003E\u003C\/p\u003E\u003Cp\u003EDr. Santosh Pande (Advisor) \u2013 School of Computer Science, Georgia Institute of Technology\u003C\/p\u003E\u003Cp\u003EDr. Qirun Zhang \u2013 School of Computer Science, Georgia Institute of Technology\u003C\/p\u003E\u003Cp\u003EDr. Vijay Ganesh \u2013 School of Computer Science, Georgia Institute of Technology\u003C\/p\u003E\u003Cp\u003EDr. Rajeev Gupta \u2013 Department of Computer Science \u0026amp; Engineering, University of California, Riverside\u003C\/p\u003E\u003Cp\u003EDr. Mark Stephenson \u2013 NVIDIA Research, NVIDIA\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EAbstract\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EAccurate knowledge of program execution behavior is critical for both compiler optimizations and system-level resource management. However, modern workloads exhibit complex, input-dependent behavior with irregular control flow and dynamic phase changes, making static reasoning inadequate. While dynamic optimization techniques offer an alternative, their high analysis overhead often negates potential performance gains. In this work, we propose \u003Cem\u003Elightweight predictive techniques that leverage runtime values, program analysis, machine learning models, and LLM reasoning to predict dynamic program behavior at multiple levels of granularity.\u0026nbsp;\u003C\/em\u003EAlthough such predictions do not provide strict guarantees, we demonstrate empirically that they enable a range of \u201csoft\u201d optimizations that yield significant performance improvements.\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EIn our first work, we showcase \u003Cstrong\u003ECompiler-Guided Cache Apportioning System (Com-CAS)\u003C\/strong\u003E\u0026nbsp;for effectively apportioning the shared Last-Level Cache (LLC), through the \u003Cem\u003Euse of runtime and compile-time cooperative framework\u003C\/em\u003E. \u003Cem\u003ECom-CAS\u003C\/em\u003E improves weighted throughput by \u003Cstrong\u003E15% on average,\u003C\/strong\u003E outperforming the state-of-the-art schemes by \u003Cstrong\u003E20%\u003C\/strong\u003E on average. In our second work, we extend this approach to estimate trip count for irregular loops, and present\u003Cstrong\u003E\u0026nbsp;Beacons Framework\u003C\/strong\u003E, an end-to-end system that estimates dynamic loop characteristics and leverages them for proactive workload scheduling. \u003Cem\u003EBeacons\u003C\/em\u003E achieves approximately \u003Cstrong\u003E85--95% prediction accuracy\u003C\/strong\u003E and improves throughput by up to \u003Cstrong\u003E2.62x\u003C\/strong\u003E over existing schedulers.\u003C\/p\u003E\u003Cp\u003EIn our third work, we introduce\u003Cstrong\u003E\u0026nbsp;Phaedrus\u003C\/strong\u003E, a framework for profile-driven software optimization via predictive modeling of dynamic application behavior. Phaedrus combines \u003Cem\u003Eapplication profile generalization\u003C\/em\u003E\u0026nbsp;and \u003Cem\u003Eapplication profile synthesis\u003C\/em\u003E, integrating generative deep learning models and LLMs with static compiler analysis to predict function calls and control-flow behavior. This enables profile-guided optimization without exhaustive profiling across application inputs and demonstrates performance gains over traditional offline profiling approaches. \u003Cem\u003EPhaedrus\u003C\/em\u003E improves performance by \u003Cstrong\u003E6% on average\u003C\/strong\u003E and reduces binary size by \u003Cstrong\u003E13.68%.\u0026nbsp;\u003C\/strong\u003E\u003C\/p\u003E\u003Cp\u003EFinally, we present \u003Cstrong\u003ECassandra\u003C\/strong\u003E, a compiler-LLM co-reasoning framework that selectively predicts fine-grained, input-specific program behavior within computationally important regions identified through coarse-grained analysis. Cassandra first uses LLM reasoning to identify hot function chains, then applies compiler analyses and loop categorization to predict loop trip counts, invocation frequencies, and other intra-function behavior required for profile synthesis. \u003Cem\u003ECassandra PGO\u003C\/em\u003E produces leaner code than Traditional PGO and achieves aggregate runtime improvements of \u003Cstrong\u003E1.32-1.44%\u003C\/strong\u003E over Traditional PGO.\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003EPredicting Dynamic Program Behavior with Compiler-Guided Machine Learning\u003C\/p\u003E","format":"limited_html"}],"field_summary_sentence":[{"value":"Predicting Dynamic Program Behavior with Compiler-Guided Machine Learning"}],"uid":"27707","created_gmt":"2026-10-12 13:29:50","changed_gmt":"2026-10-12 13:30:21","author":"Tatianna Richardson","boilerplate_text":"","field_publication":"","field_article_url":"","field_event_time":{"event_time_start":"2026-10-16T13:00:12-04:00","event_time_end":"2026-10-16T15:00:12-04:00","event_time_end_last":"2026-10-16T15:00:12-04:00","gmt_time_start":"2026-10-16 17:00:12","gmt_time_end":"2026-10-16 19:00:12","gmt_time_end_last":"2026-10-16 19:00:12","rrule":null,"timezone":"America\/New_York"},"location":"Klaus 2126","extras":[],"groups":[{"id":"221981","name":"Graduate Studies"}],"categories":[],"keywords":[{"id":"100811","name":"Phd Defense"}],"core_research_areas":[],"news_room_topics":[],"event_categories":[{"id":"1788","name":"Other\/Miscellaneous"}],"invited_audience":[{"id":"78771","name":"Public"}],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[],"email":[],"slides":[],"orientation":[],"userdata":""}}}