- talks - lists.cs.princeton.edu

Jiayi Geng will present her MSE thesis "Understanding the Failure Modes in Large Language Models"on 4/16 at 2pm in Friend 005.
by Gradinfo 14 Apr '25

14 Apr '25

Jiayi Geng will present her MSE thesis "Understanding the Failure Modes in Large Language Models"on 4/16 at 2pm in Friend 005. Date & Time: April 16th, 2:00–2:30 PM Location: Friend Center, Room 005 Thesis Title: Understanding the Failure Modes in Large Language Models Thesis advisors: Danqi Chen, Tom Griffiths Presentation Abstract: Using AI to create autonomous researchers has the potential to accelerate discovery. A prerequisite for this vision to become reality is assessing how well an AI model can identify the underlying structure of a system from its behavior. In this paper, we explore the question of whether a large language model (LLM) can learn from passive observations and actively collect informative data to refine its own hypotheses. To answer this question, we investigate the ability of LLMs to reverse-engineer four types of black-box systems, chosen to represent problems that might appear in different domains of research: list-mapping programs, formal languages, mathematical equations and board games. We use the Bayesian agents as a normative reference to study the gap between LLMs and ideal inference in reverse-engineering black-box systems. Through extensive experiments, we show that while LLMs have difficulty reverse-engineering these systems from observations alone, data generated by LLM-driven interventions can be effective. By testing edge cases, the LLM is able to refine its own hypotheses and avoid failure modes such as overcomplication, where the LLM falsely assumes prior knowledge about the black box, and overlooking, where the LLM fails to incorporate observations. These insights provide practical guidance for helping LLMs more effectively reverse-engineer black-box systems, supporting their use in making new discoveries.

1 0

Yaxin Tu will will present her General Exam "Separating Succinct Non-Interactive Arguments from Cryptographic Assumptions" on April 15, 2025 at 12pm in CS 401.
by Nicki Mahler 10 Apr '25

10 Apr '25

Yaxin Tu will will present her General Exam "Separating Succinct Non-Interactive Arguments from Cryptographic Assumptions" on April 15, 2025 at 12pm in CS 401. Committee: Alex Lombardi (adviser), Ran Raz(adviser), and Pravesh Kothari . Link to abstract and reading list: [ https://docs.google.com/document/d/1FmkljJIThGRgeHYmuChrSgzIas9uP2mUL41lulM… | https://docs.google.com/document/d/1FmkljJIThGRgeHYmuChrSgzIas9uP2mUL41lulM… ] Abstract: An argument system (computationally sound proof) for NP is succinct and non-interactive, if it contains a single prover-to-verifier message whose length is polylogarithmic the instance and witness sizes. In the past decade, the question of what cryptographic assumptions are sufficient and/or necessary for constructing SNARG, has received answers in both directions. On the one hand, the work of Gentry-Wichs’13 exhibits a barrier of using standard assumptions to construct SNARG, showing that efficient black-box reductions cannot prove the security of any SNARG construction based on any falsifiable cryptographic assumption. On the other hand, a recent line of work bypasses the Gentry-Wichs barrier by constructing SNARG based on iO and one-way functions, with a super-polynomial time security reduction. We prove an impossibility result that rules out black-box constructions of SNARG from polynomially-secure iO and one-way functions. Our result extends the Gentry-Wichs barrier, showing that there’s no black-box SNARG construction from even non-falsifiable assumptions like iO, with an efficient security reduction. Our result also shows that the super-polynomial time reduction is inherent for black-box constructions of SNARG from iO.

1 0

Simon Park will present his MSE talk “Instruct-SkillMix: A Powerful Pipeline for LLM Instruction Tuning” on Tuesday, April 15, 2025 at 12:00pm in Room 238, 41 William Street (AI Lab).
by CS Grad Department 10 Apr '25

10 Apr '25

Simon Park will present his MSE talk “Instruct-SkillMix: A Powerful Pipeline for LLM Instruction Tuning” on Tuesday, April 15, 2025 at 12:00pm in Room 238, 41 William Street (AI Lab). The members of his committee are as follows: Sanjeev Arora (Adviser) and Danqi Chen (reader) All ae welcome to attend. Please see abstract below. We introduce INSTRUCT-SKILLMIX, an automated approach for creating diverse, high quality SFT data for instruction-following. The pipeline involves two stages, each leveraging an existing powerful LLM: (1) Skill extraction: uses the LLM to extract core “skills” for instruction-following by directly prompting the model. This is inspired by “LLM metacognition” of (Didolkar et al., 2024); (2) Data generation: uses the powerful LLM to generate (instruction, response) data that exhibit a randomly chosen pair of these skills. Here, the use of random skill combinations promotes diversity and difficulty. The estimated cost of creating the dataset is under $600. Vanilla SFT (i.e., no PPO, DPO, or RL methods) on data generated from INSTRUCT-SKILLMIX leads to strong gains on instruction following benchmarks such as AlpacaEval 2.0, MT-Bench, and WildBench. With just 4K examples, LLaMA-3-8B-Base achieves 42.76% length-controlled win rate on AlpacaEval 2.0, a level similar to frontier models like Claude 3 Opus and LLaMA-3.1-405B-Instruct. Ablation studies also suggest plausible reasons for why creating open instruction-tuning datasets via naive crowd-sourcing has proved difficult. In our dataset, adding 20% low quality answers (“shirkers”) causes a noticeable degradation in performance. The INSTRUCT-SKILLMIX pipeline seems flexible and adaptable to other settings.

1 0

Shuyao Zhou will present her MSE talk “Code It, See It, Touch It: Vision-Powered AR Programming Tool for Children” on Tuesday, April 15, 2025 at 11:00am in CS 402.
by CS Grad Department 10 Apr '25

10 Apr '25

Shuyao Zhou will present her MSE talk “Code It, See It, Touch It: Vision-Powered AR Programming Tool for Children” on Tuesday, April 15, 2025 at 11:00am in CS 402. The members of her committee are as follows: Andrés Monroy-Hernández (Adviser) and Parastoo Abtahi (reader) All ae welcome to attend. Please see abstract below. We present an augmented reality (AR) authoring tool that enables children to create interactive experiences by linking visual code blocks to physical objects recognized by vision-based AI models. Building on the accessibility of block-based programming, our system encourages creativity and exploration by extending traditional visual authoring tools into the physical environment. We evaluated the system through workshops with 20 children (ages 7–16) across the United States and Argentina, where participants collaboratively designed AR experiences. Our findings suggest that the system supports playful, explorative learning, while also highlighting challenges in real-time object tracking accuracy and responsiveness.

1 0

Yanjin Chen will present her MSE talk “Predictive modeling reveals novel sequence features of prime editing efficiency and precision” on Thursday, April 10, 2025 at 1:30pm in LSI 253.
by CS Grad Department 07 Apr '25

07 Apr '25

Yanjin Chen will present her MSE talk “ Predictive modeling reveals novel sequence features of prime editing efficiency and precision ” on Thursday, April 10, 2025 at 1:30pm in LSI 253. The members of hercommittee are as follows: Yuri Pritykin (Adviser) and Britt Adamson (reader) All ae welcome to attend. Please see abstract below. Prime editing (PE) is an exciting genome editing technology that enables precise genetic modifications without requiring donor DNA or inducing double-strand breaks. However, its relatively low efficiency limits broader applications. PE efficiency largely depends on the PE guide RNA (pegRNA) design, which consists of a primer binding site (PBS) and a reverse transcriptase template (RTT) that contains the intended edit. Various computational models have been developed to predict pegRNA efficiencies across different target sites and edit types. Most existing models, however, rely on deep learning, making it difficult to interpret the contributions of individual features and their combinatorial effects on editing efficiency. Besides, existing datasets provide limited coverage in terms of the sequence space for each editing length at a single target DNA locus, offering few insights into how the intended edit sequence influences PE efficiency. Recently, in collaboration with Britt Adamson’s lab at Princeton, we acquired new PE screen datasets including 1-25 bps substitutions, insertions, and deletions at two target loci, in the K562 cell line and a variant of this cell line that is mismatch repair (MMR)-deficient. This thesis aims to dissect the influence of sequence attributes of the desired edits and cellular repair pathways on PE outcomes. We introduce generalized linear models with LASSO feature selection to identify important positional sequence features predictive of editing rates and unintended indel (insertion or deletion) rate at each target locus. Separate models were trained on insertion and substitution datasets. Our results demonstrate that linear models achieve performance comparable to that of deep neural networks, suggesting that sequence features alone are sufficient determinants of editing efficiency and precision. Beyond predictive accuracy, linear models offer greater interpretability. Notably, we observe a new factor in both our dataset and previously published datasets: for insertion edits, incorporating cytosine (C) at the first editing position increases editing efficiency, providing a practical guideline for optimizing pegRNA design. Additionally, our model recaptures the C-C mismatch preference in parental cell lines versus in MMR-deficient cell lines. In summary, this work offers novel insights into the sequence composition and cellular repair mechanisms in PE editing efficiency, which can inform researchers to design PE experiments to achieve higher efficiency.

1 0

Victor Chu will present his MSE talk “Artifact Resilient Holography” on Monday, April 7, 2025 at 2:309m in CS 105.
by CS Grad Department 02 Apr '25

02 Apr '25

Victor Chu will present his MSE talk “Artifact Resilient Holography” on Monday, April 7, 2025 at 2:309m in CS 105. The members of his committee are as follows: Felix Heid (Adviser) and Szymon Rusinkiewicz (reader) All ae welcome to attend. Please see abstract below. Holographic near-eye displays promise unparalleled depth cues, high-resolution imagery, and realistic three-dimensional parallax while possessing a compact form factor, making them promising candidates for emerging augmented and virtual reality systems. However, existing holographic display methods often assume controlled viewing conditions and overlook external factors such as eye floaters and eyelashes—obstructions that can severely degrade perceived image quality. In this work, we investigate holographic displays in the presence of these obstructions and find that these ocular artifacts are not a fundamental barrier but rather a critical design consideration involving the allocation of étendue and selection of phase profiles. Deviating from widely adopted smooth phase holograms in the field, our findings offer new insights into the design of holographic near-eye displays that maintain visual fidelity across a range of practical viewing scenarios.

1 0

Ishita Chaturvedi will present her FPO "In the SHADOW of GhOSTs: Cross-Domain Lessons in Parallelism" on April 9, 2025 at 10am in Friend 007 and Zoom
by Gradinfo 28 Mar '25

28 Mar '25

Ishita Chaturvedi will present her FPO "In the SHADOW of GhOSTs: Cross-Domain Lessons in Parallelism" on April 9, 2025 at 10am in Friend 007 and Zoom. The members of her committee are as follows: Examiners: David August (adviser), Margaret Martonosi, David Wentzlaff Readers: David August, Niraj Jha, Sharad Malik Zoom link: [ https://princeton.zoom.us/my/ishita.c | https://princeton.zoom.us/my/ishita.c ] All are welcome to attend. Title: In the SHADOW of GhOSTs: Cross-Domain Lessons in Parallelism Abstract: Modern processors struggle to balance instruction-level parallelism (ILP) and thread-level parallelism (TLP) efficiently. GPUs rely on massive TLP to hide latency but suffer in workloads with low occupancy or control divergence due to limited ILP. CPUs, optimized for ILP, become bottlenecked in memory-bound workloads where TLP could improve resource utilization. This thesis challenges the rigid separation of ILP and TLP by introducing GhOST and SHADOW, two architectures that extend ILP in GPUs and TLP in CPUs to improve execution efficiency across diverse workloads. GhOST introduces lightweight out-of-order (OoO) execution to GPUs, enabling warp-level instruction reordering without speculative execution or register renaming. Its instruction buffer-based reordering mechanism allows instructions to execute as soon as operands are ready, significantly reducing stalls in low-occupancy scenarios and mitigating control divergence overhead. GhOST achieves a 6.9 \ % geometric mean speedup (up to 36%) with only 0.007% area overhead, demonstrating the feasibility of ILP-aware GPU architectures. SHADOW is the first asymmetric simultaneous multithreading (SMT) core that dynamically balances ILP and TLP by executing OoO and in-order (InO) threads simultaneously on the same core. By leveraging deep ILP in the OoO thread and high TLP in lightweight InO threads, SHADOW maximizes CPU utilization without sacrificing single-thread performance. Unlike conventional SMT, which uniformly shares resources, SHADOW dynamically reallocates execution resources based on workload behavior. It achieves up to 3.16x speedup and 1.38x geomean improvement over an OoO CPU , with just 1% area and power overhead }. demonstrating its ability to accelerate memory-bound workloads efficiently. This thesis demonstrates how cross-domain architectural techniques can improve execution efficiency by integrating ILP into GPUs and TLP into CPUs. GhOST and SHADOW reduce execution inefficiencies, enhancing performance in workloads that challenge conventional architectures.

1 0

PPH Lunch Seminar: Matt Salganik and Mark Verhagen, tomorrow at 12pm
by Emily C. Lawrence 27 Mar '25

27 Mar '25

Please join the Princeton Precision Health (PPH) Initiative for the next talk of our spring semester series on Friday, March 28 , at 12:00 pm , at 252 Nassau Street. Matt Salganik (Princeton University) and Mark Verhagen (Oxford University) National population registries, large language models, and precision health This talk will present our interdisciplinary, international collaboration exploring the genetic, social, and environmental factors shaping human health. Our approach leverages two innovations: 1) secure access to the Netherlands’ complete population registries within a high-performance computing environment, and 2) open-weight large language models (LLMs). We’ll share our journey from award-winning LLM-based predictive modeling with Dutch registry data to our current work fusing these government records with the Netherlands Twin Registry. This fusion aims to create the world’s largest twin study with comprehensive health information, enabling 1) high-throughput heritability estimates across various health conditions and 2) estimates of the social and environmental contributions to disease by studying identical twins who have different health outcomes. Looking ahead, we plan to incorporate genetic and other “omic” data from 150,000 people to further clarify genetic contributions. Our work operates within a robust ethical and legal framework, which helps foster diverse and growing collaborations across disciplines. Supported in part by the New Jersey Alliance for Clinical and Translational Science. Lunch will be provided. Please note that 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 ] Check out our website for our full speaker lineup: [ https://pph.princeton.edu/events | https://pph.princeton.edu/events ]

1 0

March 28 PPH Lunch Seminar: Matt Salganik and Mark Verhagen National population registries, large language models, and precision health
by Emily C. Lawrence 25 Mar '25

25 Mar '25

Please join the Princeton Precision Health (PPH) Initiative for the next talk of our spring semester series on Friday, March 28 , at 12:00 pm , at 252 Nassau Street. Matt Salganik (Princeton University) and Mark Verhagen (Oxford University) National population registries, large language models, and precision health This talk will present our interdisciplinary, international collaboration exploring the genetic, social, and environmental factors shaping human health. Our approach leverages two innovations: 1) secure access to the Netherlands’ complete population registries within a high-performance computing environment, and 2) open-weight large language models (LLMs). We’ll share our journey from award-winning LLM-based predictive modeling with Dutch registry data to our current work fusing these government records with the Netherlands Twin Registry. This fusion aims to create the world’s largest twin study with comprehensive health information, enabling 1) high-throughput heritability estimates across various health conditions and 2) estimates of the social and environmental contributions to disease by studying identical twins who have different health outcomes. Looking ahead, we plan to incorporate genetic and other “omic” data from 150,000 people to further clarify genetic contributions. Our work operates within a robust ethical and legal framework, which helps foster diverse and growing collaborations across disciplines. Supported in part by the New Jersey Alliance for Clinical and Translational Science. Lunch will be provided. Please note that 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 ] Check out our website for our full speaker lineup: [ https://pph.princeton.edu/events | https://pph.princeton.edu/events ]

1 0

AI² Research Talk Series-Join us! March 24th, 12 p.m., "Physical Computing in Elastic Metamaterials", Marc Serra-Garcia, AMOLF
by Emily C. Lawrence 24 Mar '25

24 Mar '25

Please join us! 12 p.m., Monday, March 24 th AI² Research Talk Series Bendheim House, Classroom 103 Lunch available at 11:45 a.m. [ https://forms.gle/bbUbLvZ14CDbHvAW9 | Kindly RSVP ] Marc Serra-Garcia (Manresa, 1987) is a tenure-track group leader at the AMOLF research institute in Amsterdam. After a BSc in Physics at Universitat Autònoma de Barcelona , he did a MSc degree at Caltech, and a PhD at ETH Zurich. His research is currently focused on novel (beyond-CMOS) approaches to information processing, with the goal of using a richer and more diverse set of physical phenomena to implement relevant information-processing tasks with low power. The research has been recognized by an ERC Starting grant, and various other grants from the Dutch Research Council and Swiss National Science Foundation. “Physical Computing in Elastic Metamaterials” There is a significant range of physical phenomena—from nonlinear elasticity, to symmetry, noise, topology, and disorder — that are rarely utilized in traditional computing paradigms. Yet these phenomena can unlock new efficiencies, by directly processing signals in their natural domain, and by bypassing the traditional abstraction stack associated with digital CMOS technology. However, building physical computers is challenging. Information processing tasks generally involve complex input-output relations, thus requiring designs that are highly expressive; and for these designs, the relation between function and structure is nontrivial, complicating the simulation, design, and fabrication of devices. In my talk, I will illustrate our journey towards using metamaterials for physical computing, with two recent examples. First, I will talk about our results in passive speech recognition, where we leverage a phononic metamaterial to implement wake-up-word detection with zero standby power consumption. Second, I will discuss our ongoing work in self-learning materials, that autonomously adapt to improve their performance—driven by their ability to form long-term memories in response to examples and external feedback. [ https://invent.ai.princeton.edu/about/get-involved | AI² Mailing List ] [ https://invent.ai.princeton.edu/about/get-involved | Get Involved ] We are rapidly developing AI for Accelerated Invention activities and programs. Sign up for the mailing list and reach out to us with your questions.

1 0