- talks - lists.cs.princeton.edu

Mona Wang will present her FPO "Protecting the Network Traffic of One Billion People: Transport Security in the Global Mobile Ecosystem" on August 18, 2025 at 1pm in 306 Sherrerd Hall and Zoom.
by Gradinfo 18 Aug '25

18 Aug '25

Mona Wang will present her FPO "Protecting the Network Traffic of One Billion People: Transport Security in the Global Mobile Ecosystem" on August 18, 2025 at 1pm in 306 Sherrerd Hall and Zoom. Time: August 18 2025, 1PM Eastern time. Location: 306 Sherrerd Hall Zoom link: https://princeton.zoom.us/j/94818544623 Title: Protecting the Network Traffic of One Billion People: Transport Security in the Global Mobile Ecosystem Examiners: Jonathan Mayer (adviser), Prateek Mittal (adviser), and Arvind Narayanan. Readers: Andres Monroy-Hernandez and Roya Ensafi All are welcome to attend Abstract: TLS is the de-facto standard for encrypting network communications. Today, upwards of 80% of pages loaded on Firefox, Chrome, and Safari are encrypted with TLS. This might be the story for web, but what about mobile? Existing measurements of mobile network en- cryption fall short: they often focus on the Google Play ecosystem, which necessarily excludes mobile users in China, who comprise a massive portion of the global Internet. This thesis demonstrates that HTTPS is, in fact, not everywhere, and that a massive por- tion of mobile network communications remains poorly encrypted and accessible to systems of mass surveillance. These issues are particularly concentrated in mobile applications devel- oped in China, which have been overlooked by the global security community despite their massive popularity and influence. Three studies provide different perspectives that demonstrate both the (1) massive pop- ularity of proprietary network encryption protocols in top mobile applications, and (2) the insecurity of such homegrown protocols. First, I present our reverse-engineering of WeChat’ s proprietary transport encryption protocol and subsequent privacy analysis of the WeChat Mini Program ecosystem. Then, I review the network encryption used by popular Chinese keyboards to encrypt user keystrokes. Finally, I present a large-scale study of encryption pro- tocols used by thousands of popular mobile applications. We discovered severe vulnerabilities enabling network attackers to decrypt sensitive data in the vast majority of the proprietary encryption protocols we analyzed. Through the vulnera- bilities fixed en masse as a result of this work, this research has directly improved the network security of over one billion people.

1 0

Javed Aman will present his FPO "Towards Scalable Inference of Protein-Protein Interfaces: Reasoning with structural, functional, and sequence data" on Monday, August 25, 2025 at 3:00 PM in Jadwin 111.
by CS Grad Department 15 Aug '25

15 Aug '25

Javed Aman will present his FPO " Towards Scalable Inference of Protein-Protein Interfaces: Reasoning with structural, functional, and sequence data " on Monday, August 25, 2025 at 3:00 PM in Jadwin 111. The members of Javed ’s committee are as follows: Examiners: Mona Singh (Adviser), Ellen Zhong , Stanislav Shvartsman , Martin Wühr Readers: Olga Troyanskaya , Yuri Pritykin Everyone is invited to attend his talk. Abstract follows below: Protein–protein interactions (PPIs) are central to virtually all biological processes. Understanding how these interactions are mediated, including identifying which residues form the interface, is critical for studying disease mechanisms and designing targeted therapeutics. However, identifying and characterizing PPI interface residues at the proteome scale remains a major challenge due to limitations in experimental coverage and throughput. This dissertation presents three computational approaches for inferring or annotating protein–protein interfaces across different biological and data scales. The first, protein-protein interacDome (PPI-DOME), uses structural data from X-ray crystallography and cryoelectron microscopy to build domain–domain interaction profiles. These profiles identify which positions within domains mediate interactions and enable efficient screening of millions of candidate interactions in seconds, providing a scalable alternative to structure-based computational modeling methods. The second project, Kinase activity and inference dashboard (KINAID), focuses on identifying transient interactions in phosphorylation cascades. It combines orthology mapping, experimentally-derived position weight matrices capturing the specificities of human kinases, and experimentally observed phosphosites to annotate kinase–substrate relationships across over 2,000 kinases from ten model organisms. KINAID enables rapid analysis of high-throughput phosphoproteomic data. The third project, Comparable embeddings from sequence networks (CESN), addresses the need for high-quality, sequence-level representations, or embeddings, suitable for downstream machine learning models. CESN leverages amino acid embeddings generated by protein language models and learns representations that preserve amino acid-level detail across the entire sequence, overcoming the limitations of standard pooling strategies such as mean or CLS token pooling. The CESN approach provides a foundation for machine learning models that predict protein properties or interactions between pairs of proteins. Together, the methods introduced in this thesis contribute to scalable, interface-centric reasoning by leveraging structural information, functional information, and representation learning. They o er complementary strategies for advancing our understanding of how protein interfaces govern molecular interaction, function, and regulation across diverse biological systems.

1 0

Molly Pan will present her FPO "Verifiable Traffic Control with Compact Data Structures in the Data Plane" on Friday, Aug 8, 2025 at 10am in Friend 125 and Zoom.
by Gradinfo 30 Jul '25

30 Jul '25

Molly Pan will present her FPO "Verifiable Traffic Control with Compact Data Structures in the Data Plane" on Friday, Aug 8, 2025 at 10am in Friend 125 and Zoom. Advisers: Jennifer Rexford and Andrew Appel Examiners: Jennifer Rexford (Adviser), David Walker, and Maria Apostolaki Readers: Hyojoon Kim (University of Virginia) and Andrew Appel Zoom: [ https://princeton.zoom.us/j/7979634003 | https://princeton.zoom.us/j/7979634003 ] All are welcome to attend. Abstract: As network traffic grows in volume and diversity, operators increasingly require real-time, in-network control over packets. Tasks such as rate-limiting large flows, identifying heavy downloaders, and dropping unsolicited packets demand stateful processing at line rate—capabilities absent in traditional fixed-function switches. The advent of programmable data planes has made such control feasible, allowing custom stateful logic to run directly in the network. However, programming traffic control remains challenging. Data planes impose strict memory constraints, making conventional data structures impractical. Instead, operators rely on approximate data structures like Bloom filters and hash tables, trading accuracy for efficiency. Selecting and configuring these structures requires expertise, as poor choices can lead to excessive errors or wasted resources. To address this, we developed Network Approximate Programming (NAP), a high-level language centered on a versatile approximate dictionary abstraction. This abstraction captures a broad range of compact data structures while allowing programmers to specify the types of errors an application can tolerate. The NAP compiler automatically selects and configures the appropriate structure to optimize hardware utilization and compiles to P4, the native programming language for data planes---significantly simplifying development. Ensuring correctness of approximate data structures in P4 poses additional challenges. While research has advanced new streaming algorithms, little attention has been paid to adapting them systematically to data-plane constraints and refreshing them at runtime. Moreover, these practical concerns may introduce subtle bugs. To address these concerns, we developed general synthesis and verification frameworks. We deployed an approximate sliding-window Bloom filter on Intel Tofino and verified its correctness, demonstrating how to build P4 data structures with correctness guarantees. Finally, P4 itself presents foundational challenges: its low-level design, ambiguous specification, and lack of formal semantics make stateful programming error-prone. To provide a rigorous foundation, we developed a formal semantics for P4 based on its two-phase evaluation model. Our mechanized formalization precisely defines language constructs, including stateful externs on Intel Tofino. By adhering to P4’s intended behavior, our work uncovered previously undocumented ambiguities and contributed improvements to the specification. Together, these efforts form a cohesive framework for enabling robust network control in programmable data planes.

1 0

Kunal Mittal will present his FPO "Parallel Repetition of Multiplayer Games: New Bounds and Applications" on August 1, 2025 at 12pm is CS 302 and Zoom.
by Gradinfo 29 Jul '25

29 Jul '25

Kunal Mittal will present his FPO "Parallel Repetition of Multiplayer Games: New Bounds and Applications" on August 1, 2025 at 12pm is CS 302 and Zoom. The committee is as follows: Examiners: Ran Raz (adviser), Gillat Kol,Mark Braverman Readers: Pravesh Kothari and Zeev Dvir Zoom Link: [ https://princeton.zoom.us/j/98962124632?pwd=9Or6rUipHulgmrqsgRfqvIn24haJzh.1 | https://princeton.zoom.us/j/98962124632?pwd=9Or6rUipHulgmrqsgRfqvIn24haJzh.1 ] Meeting ID: 989 6212 4632 Passcode: 750823 All are welcome to attend. Abstract: This thesis studies the parallel repetition of multiplayer games. These games were first (implicitly) introduced in quantum mechanics, to study quantum entanglement with regard to local hidden-variable theories, in the context of the Einstein-Podolsky-Rosen (EPR) paradox. Later, multiplayer games were formally defined and extensively studied in complexity theory, in the framework of interactive proof systems and probabilistically checkable proofs (PCPs). Parallel repetition is a natural hardness-amplification technique used to decrease the value of multiplayer games while maintaining their single-round structure. Despite its apparent simplicity, analyzing the effects of parallel repetition on the value of a game remains a challenging problem, particularly for games involving three or more players. This work makes two main contributions: establishing stronger bounds on the value of parallel repetition for several classes of multiplayer games, and developing new connections to computational complexity and extremal combinatorics. In our first contribution, we substantially improve the best-known bounds on the value of parallel repetition for several classes of multiplayer games. In particular, we prove an inverse-polynomial bound for all 3-player games with 1-bit questions, as well as an inverse polynomial bound for a very large class of k-player games called playerwise-connected games. Furthermore, we prove exponential bounds for certain special 3-player games like the anticorrelation game. Prior to our work, nothing better than weak inverse-Ackermann bounds were known, for any of these games. Our second contribution shows that a sufficiently strong parallel repetition theorem for a special case of multiplayer games implies super-linear lower bounds for multi-tape Turing machines with advice, which is a long-standing open problem in complexity theory. Additionally, we demonstrate equivalences between several high-dimensional problems in extremal combinatorics and parallel repetition of multiplayer games over large answer alphabets.

1 0

Zhijun Zhang will present his FPO "Efficient and Robust Communication for Massive Data: Sublinear Algorithms and Coding with Interaction" on Monday, 7/21/2025 at 10:00am in CS 302
by Gradinfo 15 Jul '25

15 Jul '25

Zhijun Zhang will present his FPO "Efficient and Robust Communication for Massive Data: Sublinear Algorithms and Coding with Interaction" on Monday, 7/21/2025 at 10:00am in CS 302. The members of his committee are as follows: Examiners: Gillat Kol (Adviser), Ran Raz, Mark Braverman Readers: Huacheng Yu, Pravesh Kothari All are welcome to attend. Abstract: The world has been witnessing a massive surge in data accumulation, which, combined with the distributed nature of the data, necessitates efficient and robust methods for information transmission. The study of communication addresses diverse aspects of such transmission between different parties, and has become a powerful tool across a wide range of seemingly unrelated research fields. In this dissertation, we first examine sublinear models, where algorithms for handling massive data are designed to minimize the usage of valuable computational resources. We study fundamental graph problems, such as MST, MIS, and Maximal Matching, for which celebrated multi-pass algorithms are known in streaming and distributed sketching settings. Using novel information theoretic techniques, we prove almost matching lower bounds for any number of passes. Additionally, we explore how interaction can improve the robustness of communication. While the optimal noise resilience of one-way binary error correcting codes has long been known to be 1/4, we show how to achieve resilience beyond 1/4 by allowing a single short round of noisy communication from the receiver back to the sender. In the setting where the receiver’s feedback is noiseless, we characterize the resilience as a function of the number of feedback rounds.

1 0

Fan Yi will present his FPO "Cross-Layer Telemetry and Optimization for Real-Time Interactive Applications in 5G Networks" on Tuesday, July 8, 2025 at 3pm in Friend 007.
by Gradinfo 30 Jun '25

30 Jun '25

Fan Yi will present his FPO "Cross-Layer Telemetry and Optimization for Real-Time Interactive Applications in 5G Networks" on Tuesday, July 8, 2025 at 3pm in Friend 007. The members of his committee are as follows: Examiners: Kyle Jamieson (adviser), Ravi Netravali, Maria Apostolaki (ECE) Readers: Wyatt Lloyd, Yaxiong Xie (University at Buffalo, SUNY) All are welcome to attend. Abstract: Real-time applications such as interactive video conferencing now underpin daily communication, yet their performance over 5G networks is often inconsistent and unreliable. While 5G promises unprecedented speed, its internal complexity introduces significant fluctuations in latency and throughput that directly degrade the user's Quality of Experience (QoE). These performance issues persist because the network stack operates in functional silos, obscuring the complex, cross-layer interactions that are the true root cause of application-level degradation. The fundamental challenge lies in the highly dynamic wireless channel and the lack of cross-layer visibility. Dynamic channel conditions and millisecond-level scheduling at the physical and link layers continually reshape available capacity, yet application-layer congestion controllers perceive only averaged, coarse-grained feedback and therefore misinterpret the wireless artifacts as generic network congestion. This flawed perception causes them to overreact by unnecessarily throttling sending rates or increasing buffer delays, which impairs the user experience. This thesis tackles this challenge by building a suite of intelligent wireless systems that first expose millisecond-level cross-layer telemetry, and then exploit it for real-time 5G channel optimization and sensing. This work first introduces Athena, a novel measurement framework that correlates high-resolution telemetry from the 5G physical layer up to the video conferencing application. Building on this framework, Domino is developed as a first-of-its-kind diagnostic tool that automates the root-cause analysis of QoE degradation by modeling and searching a graph of causal relationships, successfully identifying 24 previously unknown event chains that impair 5G video conferencing. Moving from diagnosis to direct network enhancement, WaveFlex is the first smart surface for private CBRS 5G that monitors control channels and steers signals in real time, increasing median throughput by 19.5%. Finally, CAPER repurposes cellular measurements for privacy-preserving proximity sensing, achieving highly accurate proximity detection via a custom deep learning pipeline. Collectively, these contributions deliver a comprehensive paradigm for building intelligent wireless systems. By providing the essential tools for deep measurement (Athena), automated diagnostics (Domino), proactive network optimization (WaveFlex), and novel sensing capabilities (CAPER), this thesis establishes the foundational visibility and control necessary to unlock the full potential of 5G and future wireless networks.

1 0

PPH Spring Seminar- Prof. Zohar Yakhini, Monday, June 23rd at 12pm at 252 Nassau St
by Emily C. Lawrence 20 Jun '25

20 Jun '25

Reminder! Please join the Princeton Precision Health (PPH) Initiative for the next talk of our spring semester series on Monday, June 23rd, at 12:00 pm, at 252 Nassau Street. Prof. Zohar Yakhini, from the Efi Arazi School of Computer Science and Head of Machine Learning and Data Science Program, at Reichman University, will present virtually, “Statistical enrichment, image analysis and spatial biology”. The spatial context of molecular and morphological characteristics of cells and cell populations can be very important in determining pathogenesis or other higher level biological processes. It is therefore desirable to understand molecular profiles in a spatial context. In this talk I will describe the application of computational and statistical techniques - ranging from assessing enrichment in ranked lists to applying deep learning-based image analysis - to various spatial/molecular measurement modalities. Modalities addressed include H&E imaging, IHC and spatial transcriptomics. I will describe methods, and some example results and will discuss the potential impact in terms of life science and medical science. Lunch will be provided. Getting to the seminar space currently requires that you climb a set of stairs. If an accommodation is needed, please contact PPH in advance at: [ mailto:PrincetonPPH@princeton.edu | PrincetonPPH(a)princeton.edu ] Thank you, Princeton Precision Health

1 0

Anunay Kulshrestha will present his FPO "Advancing Public Policy Through Cryptography: Building Private and Accountable Systems" on Friday, June 20, 2025 at 11am in Sherrerd 308 (CITP Conference Room) and Zoom.
by Gradinfo 18 Jun '25

18 Jun '25

Anunay Kulshrestha will present his FPO "Advancing Public Policy Through Cryptography: Building Private and Accountable Systems" on Friday, June 20, 2025 at 11am in Sherrerd 308 (CITP Conference Room) and Zoom. Zoom: https://princeton.zoom.us/j/94232769862 All are welcome to join. Examiners: Jonathan Mayer (Adviser), Prateek Mittal, Aleksandra Korolova Readers: Jennifer Rexford, Seny Kamara (Brown University), and Kobbi Nissim (Georgetown University) Abstract: We illustrate how modern cryptography can offer novel solutions to longstanding policy challenges improving both privacy and accountability—–objectives that are often in tension with each other—by discussing examples from content moderation and foreign intelligence surveillance. We formalize the relevant functionalities and design practical protocols that realize them. We provide security proofs, open-source implementations, and benchmarks to demonstrate feasibility of our protocols. We describe limitations of technical answers to complex socio-technical questions. We conclude by discussing the policy impact of our contributions and promising avenues for future work.

1 1

Please Join us at 252 Nassau Street, on June 23rd at 12 pm for our PPH Spring Seminar- Prof. Zohar Yakhini will present virtually “Statistical enrichment, image analysis and spatial biology”.
by Emily C. Lawrence 17 Jun '25

17 Jun '25

Please join the Princeton Precision Health (PPH) Initiative for the next talk of our spring semester series on Monday, June 23rd, at 12:00 pm, at 252 Nassau Street. Prof. Zohar Yakhini, from the Efi Arazi School of Computer Science and Head of Machine Learning and Data Science Program, at Reichman University, will present virtually, “Statistical enrichment, image analysis and spatial biology”. The spatial context of molecular and morphological characteristics of cells and cell populations can be very important in determining pathogenesis or other higher level biological processes. It is therefore desirable to understand molecular profiles in a spatial context. In this talk I will describe the application of computational and statistical techniques - ranging from assessing enrichment in ranked lists to applying deep learning-based image analysis - to various spatial/molecular measurement modalities. Modalities addressed include H&E imaging, IHC and spatial transcriptomics. I will describe methods, and some example results and will discuss the potential impact in terms of life science and medical science. Lunch will be provided. Getting to the seminar space currently requires that you climb a set of stairs. If an accommodation is needed, please contact PPH in advance at: [ mailto:PrincetonPPH@princeton.edu | PrincetonPPH(a)princeton.edu ] Thank you, Princeton Precision Health

1 0

Ted Zadouri will present his General Exam "Hardware-Efficient Attention for Inference" on Friday, June 6, 2025 at 1:00 PM via Zoom.
by gradinfo＠cs.princeton.edu 06 Jun '25

06 Jun '25

Ted Zadouri will present his General Exam "Hardware-Efficient Attention for Inference" on Friday, June 6, 2025 at 1:00 PM via Zoom. Zoom link: https://princeton.zoom.us/j/94965926485 <https://www.google.com/url?q=https://princeton.zoom.us/j/94965926485&sa=D&s ource=calendar&ust=1749645969349347&usg=AOvVaw3t8uU_z00Dbt6kPc88ZdtS> Committee Members: Tri Dao (advisor), Kai Li, Chi Jin Abstract: In light of test-time compute, inference efficiency drives progress in AI, demanding a greater emphasis on inference-aware architectures. The sequential nature of decoding limits parallelism. For large batches and long contexts, the key value cache often bottlenecks decoding as it consumes scarce GPU memory, and each decoding step must load the large cache from HBM. Fetching this cache dominates latency relative to the small matrix-vector computation per decoding step, causing prolonged low GPU utilization. This bottleneck hinders a wide range of use cases, including latency-sensitive requests, multi-step reasoning agents, long-context video models, and test-time compute scaling. This work redesigns attention through the lens of arithmetic intensity to shrink the KV cache size, thereby accelerating decoding without sacrificing quality or distributed scalability. We first propose Grouped-Tied Attention (GTA), a simple variant that combines and reuses key and value states, reducing memory transfer while preserving accuracy. We then introduce Grouped Latent Attention (GLA), a parallel-friendly latent attention paired with low-level optimizations for fast decoding at full quality. Experiments show that GTA matches Grouped-Query Attention (GQA) quality while using roughly half the KV cache and that GLA matches Multi-head Latent Attention (MLA) and is easier to shard. Our optimized GLA kernel is up to 2x faster than FlashMLA, e.g., in a speculative decoding setting where the query length exceeds one. Additionally, loading a smaller KV cache per device further reduces latency and boosts throughput in online serving benchmarks. Reading List: https://docs.google.com/document/d/1aNdCzisnXfuicP_7-sqVECLbILa55uwhWha3oXPu GoM/edit?usp=sharing <https://www.google.com/url?q=https://docs.google.com/document/d/1aNdCzisnXf uicP_7-sqVECLbILa55uwhWha3oXPuGoM/edit?usp%3Dsharing&sa=D&source=calendar&us t=1749645969349347&usg=AOvVaw2IiMs_Y0APTE_5i7jhQttW> Everyone is invited to attend the talk, and those facultywishing to remain for the oral exam following are welcome to do so.

1 0