- talks - lists.cs.princeton.edu

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

Satadal Sengupta will present his Pre-FPO "Protocol-Aware Telemetry For In-Network Control" on June 11th, 2025 at 12:30pm in Friend 125 and Zoom.
by Gradinfo 04 Jun '25

04 Jun '25

Satadal Sengupta will present his Pre-FPO "Protocol-Aware Telemetry For In-Network Control" on June 11th, 2025 at 12:30pm in Friend 125 and Zoom. Committee: Jennifer Rexford (Advisor) Ravi Netravali (Examiner) Maria Apostolaki (Examiner, Princeton ECE) David Walker (Reader) Hyojoon Kim (Reader, University of Virginia CS) Zoom: https://princeton.zoom.us/j/2332835952?pwd=tYVB2ta8EI4aC3eX4biZCbu4VyYK0J.1 Abstract: The average Internet user consumes a staggering 4 GB of data per day in 2025, compared to just 1 MB per month in 1993 when the web went public. Yet, despite major improvements, the underlying infrastructure still does not guarantee performance or security, and is afflicted by issues such as packet loss, excessive latency, and suboptimal paths. To address these deficiencies, sophisticated network protocols have been developed: for example, TCP aims to provide reliable delivery of on-demand traffic, and WebRTC addresses timely delivery of interactive traffic—together accounting for over 80% of all Internet traffic. These protocols rely on the Border Gateway Protocol for policy-driven routing. Despite the effectiveness of these protocols, significant performance and security problems still persist. Among these problems, short‑lived or stealthy issues are particularly damaging because they often go undetected by traditional coarse‑grained (flow‑level or packet sampling-based) monitoring, yet cause disproportionate impacts on network performance and security. For example, nearly a quarter of business users of video-conferencing applications report lags during peak hours due to transient network congestion, and billions of dollars are lost yearly to stealthy hijacks of cryptocurrency services. Tackling these problems requires protocol‑aware methods to monitor and control the network in real time at a fine granularity (i.e., packet level). This is a challenging task at Internet scale. State-of-the-art networking hardware, such as programmable switches, is characterized by Tbps-scale speed and programmability, offering the only practical opportunity for fine‑grained, real‑time network visibility and control. However, their high speed comes at the cost of limited programmability in the form of stringent memory and computational constraints. This makes it challenging to analyze traffic governed by complex protocols since the hardware cannot retain complete protocol state under such tight resource constraints. In this dissertation, we address these challenges by developing efficient, hardware‑amenable techniques for real-time network monitoring and control. First, we measure a critical packet‑level metric—TCP round‑trip time (RTT)—to enable prompt detection and mitigation of stealthy long‑distance routing attacks. Second, we analyze popular, representative video‑conferencing applications, such as Zoom (proprietary) and MediaSoup (open‑source, WebRTC‑based), and extract detailed packet‑level performance metrics. Our insights drive the design of a scalable, hardware‑software co‑designed architecture for video‑conferencing, significantly improving performance and reducing operational costs compared to conventional software solutions. In summary, this work advances scalable, protocol‑aware techniques for network monitoring and control to enhance the performance and security of the Internet.

1 0

Zhou Lu will present his FPO "Online Learning: Optimization, Control, and Learning Theory" on Monday, June 16, 2025 at 1:00 PM in Friend 005.
by gradinfo＠cs.princeton.edu 03 Jun '25

03 Jun '25

Zhou Lu will present his FPO "Online Learning: Optimization, Control, and Learning Theory" on Monday, June 16, 2025 at 1:00 PM in Friend 005. The members of Zhou's committee are as follows: Examiners: Elad Hazan (Adviser), Chi Jin, Ryan Adams Readers: Matt Weinberg, David Woodruff (CMU) A copy of his thesis is available upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of thethesis. Everyone is invited to attend his talk. Abstract follows below: Online learning is a foundational paradigm in learning theory, addressing the challenge of making predictions from sequential data. Initially introduced as a mathematical model of learnability, online learning has since evolved into a versatile framework with widespread applications in optimization, control, economics, and beyond, spurring algorithmic innovations and empirical advances. This dissertation investigates various theoretical aspects of online learning, encompassing both its fundamental limits and its applications to optimization and control. The first part of this dissertation focuses on online convex optimization. We develop more efficient algorithms for adaptive regret minimization, improving both query and projection efficiency. We also address the more challenging setting of online non-convex optimization via a novel reduction approach. The second part of the dissertation is devoted to online nonstochastic control, a generalization of classical optimal control that relaxes assumptions on the cost structure and disturbance models. We propose a new framework for controlling marginally stable linear dynamical systems, a notoriously difficult class of systems. We also design an optimal algorithm for bandit nonstochastic control with general loss functions, and develop a meta-algorithm capable of aggregating multiple base control algorithms. The final part of the dissertation turns to core problems in learning theory. We establish a necessary and sufficient condition for inductive reasoning, a foundational problem in epidemiology, by drawing a new connection to online learning theory. Along the way, we also present new results in multimodal learning theory, building provable statistical and computational advantages. Together, these contributions advance the theoretical understanding of online learning and extend its applicability to a broader range of domains, offering new insights and tools for both researchers and practitioners.

1 0

Yunyun Wang will present her FPO "Generative Universal Models for Speech" on Friday, May 30, 2025 at 2:00 PM in CS 302.
by gradinfo＠cs.princeton.edu 28 May '25

28 May '25

Yunyun Wang will present her FPO "Generative Universal Models for Speech" on Friday, May 30, 2025 at 2:00 PM in CS 302. Location: CS 302 The members of Yunyun's committee are as follows: Examiners: Adam Finkelstein (Adviser), Szymon Rusinkiewicz, Felix Heide Readers: Danqi Chen, Zeyu Jin (Adobe Research) A copy of her thesis is available upon request. Please email gradinfo(a)cs.princeton.edu <mailto:gradinfo@cs.princeton.edu> if you would like a copy of the thesis. Everyone is invited to attend her talk. Abstract follows below: This thesis presents a comprehensive framework for controllable speech synthesis through self-supervised generative modeling. We propose Generative Universal Models for Speech (GUMS), a system that decomposes speech into disentangled representations-speaker embeddings, acoustic embeddings, and content representations, and reconstructs it using a synthesis model. Our approach enables detailed control over speaker voice, environmental acoustics, speech content, and speaking rate. We introduce three key representation models. First, GR0 learns global speaker embeddings by disentangling them from time-varying local content without requiring speaker labels. Second, we develop content representation models AIC and GUMS Codec that capture speech content in continuous and quantized forms, respectively. The AIC model enforces speaker and pitch invariance through the alteration invariant content loss. GUMS Codec builds on the speech codec model DAC, incorporating residual vector quantization along with speaker and pitch conditioning. The result is a highly compact, discrete, and language-independent representation that is well-suited for manipulation, control, and efficient transmission. We then integrate these representations into a high-fidelity speech synthesis model, DiTVC, based on a Diffusion Transformer architecture. DiTVC enables direct prompting using target speaker audio instead of relying on fixed embeddings, allowing for more expressive voice conversion and robust prosody control. By combining these models, we achieve controllable, high-quality speech synthesis using unlabeled, in-the-wild data. The unified framework advances both representation learning and generation, offering an interpretable, and editable approach to speech synthesis.

1 1

Tim Kosfeld will present his General Exam "Advances in Supervised Interpretation and Preprocessing of ATAC-seq" on Tuesday, May 27, 2025 at 12:00 PM in Icahn 101.
by CS Grad Department 22 May '25

22 May '25

Tim Kosfeld will present his General Exam "Advances in Supervised Interpretation and Preprocessing of ATAC-seq" on Tuesday, May 27, 2025 at 12:00 PM in Icahn 101. Committee Members: Yuri Pritykin (advisor), Mona Singh, Ben Raphael Abstract: Single-cell assay for transposase-accessible chromatin with sequencing (scATAC-seq) has been widely used to profile chromatin accessibility and explore the epigenomic properties in heterogeneous cell populations. Furthermore, transcription factors (TFs) bind to accessible sites, and scATAC-seq analysis could associate TF binding specificities, typically available as sequence motifs, with variance in chromatin accessibility. Here we present a new supervised TF motif regression method, MotifReg, for computational analysis and interpretation of scATAC-seq data. MotifReg can efficiently analyze large-scale scATAC-seq datasets and learn interpretable associations between TF motifs and accessible peaks. The resultant low-dimensional cell representations in the TF activity space facilitate downstream cell type-specific analyses of scATAC-seq data. The supervised formulation in MotifReg enables ranking TF motifs by significance of their association with chromatin accessibility, implicating cell type-specific TFs. Furthermore, by comparing MotifReg performance across data analysis settings, we develop guidelines and best practices for scATAC-seq peak calling, data preprocessing and normalization. We leverage these best practices to show that aggregation of biologically similar ATAC data allows for the discovery of chromosomal regions enriched for chromatin interaction, known motif binding, or ATAC co-accessibility. We provide an automated methodology to quickly and efficiently scrape, compile, and preprocess this data from online sources and illustrate how construction of these high-quality atlases can complement existing analyses. Reading List: https://docs.google.com/document/d/1gnqGszJkHUIQisI3UL_mw-w_pCR2hD0dmL5N5OA… Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0

Yuqi Zhang will present her General Exam "Linking Genomic Alterations to Cancer Pathway Expression Changes via Canonical Correlation Analysis" on Wednesday, May 21, 2025 at 2:30 PM in Icahn Lab 280.
by CS Grad Department 16 May '25

16 May '25

Yuqi Zhang will present her General Exam "Linking Genomic Alterations to Cancer Pathway Expression Changes via Canonical Correlation Analysis" on Wednesday, May 21, 2025 at 2:30 PM in Icahn Lab 280. Committee Members: Mona Singh (advisor), Olga Troyanskaya, Yuri Pritykin Abstract: Cancer is a disease where numerous genomic alterations are acquired and cause widespread disruption to cellular function. Understanding how these alterations impact biological pathways is essential for uncovering the underlying mechanisms of complex diseases. We propose a framework based on Canonical Correlation Analysis (CCA) to uncover relationships between genes frequently altered in cancer and dysregulated pathway expression. Specifically, we jointly consider the multi-omic alterations affecting a known cancer-relevant gene—including its copy number alteration (CNA), mutation, and methylation status—and model their joint relationship with the expression of genes within specific biological pathways. A key contribution of our framework is a novel sampling-based hypothesis test for CCA that mitigates random signals often observed in high-dimensional datasets. Additionally, we adapt existing visualization methods to uncover relationships between specific types of genomic alterations and gene expression. Our model inherently accounts for correlation between molecular alterations and gene expression, improving sensitivity in detecting biologically meaningful associations. We apply our method across multiple driver genes and pathways, revealing significant associations that traditional gene-level approaches miss. Our framework also reveals the directionality of genomic alterations' regulatory effects on pathway expression while accounting for correlation and joint effects in high-dimensional data. By capturing the joint influence of genomic alterations on pathway activity while reducing statistical noise, our method offers a powerful approach for discovering key regulatory mechanisms in cancer. Reading List: https://docs.google.com/document/d/1edX-ARuzaajncovwykvHNW8GUXdYFqiw83vb_AY… Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0

Zhiyue Zhang will present her General Exam "Best practices of multiome data analysis and application to anti-tumor immunology in obesity" on Monday, May 19, 2025 at 2:30 PM in Icahn 200.
by CS Grad Department 14 May '25

14 May '25

Zhiyue Zhang will present her General Exam "Best practices of multiome data analysis and application to anti-tumor immunology in obesity" on Monday, May 19, 2025 at 2:30 PM in Icahn 200. Committee Members: Yuri Pritykin (advisor), Lydia Lynch (advisor), Mona Singh, Olga Troyanskaya Abstract: Technological advances in single-cell sequencing have expanded the detection of cell phenotype beyond the transcriptome, allowing for more refined characterization of cell types and discovery of novel therapeutic targets. Notably, paired scRNA-seq and scATAC-seq, called multiome data, facilitates inference of how variation in chromatin accessibility regulates gene expression and cell identity. However, unlike scRNA analysis, procedures of scATAC analysis lack consensus, mainly due to the high sparsity and the unfixed feature set of the original data. In this talk, we first present evaluating differential representations of scATAC by integrating with scRNA and validating against the pairing information of multiome data. We show that several decisions in current scATAC processing tools result in loss of information and recommend best practices of featurization and downstream analysis. Next, we demonstrate the power of multiome through an analysis of new multiome data from tumor-infiltrating CD8+ T cells in obesity. We find an enrichment of progenitor exhausted CD8+ T cells in obese tumors, and generate a new hypothesis for "the obesity paradox", namely why certain obese patients have improved response to immune checkpoint inhibitors despite displaying impaired anti-tumor immunity. Reading List: https://docs.google.com/document/d/1Pl6x0mOhhlj8NIXbllube2f11S-JnxglPgM0kOT… Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0