January 2025 - talks - lists.cs.princeton.edu

CS Colloquium speakers: week of February 26
by Emily C. Lawrence 09 May '25

09 May '25

CS Colloquium speakers Speaker: Jialin Ding, Amazon Web Services Date: Monday, February 26 Time: 12:30pm EST Location: CS 105 Host: Wyatt Lloyd Event page: https://www.cs.princeton.edu/events/26578 Register for live-stream online here: https://princeton.zoom.us/webinar/register/WN_10z_l8MYRr21ZOuPqA8Odw Title: Instance-Optimization: Rethinking Database Design for the Next 1000X Abstract: Modern database systems aim to support a large class of different use cases while simultaneously achieving high performance. However, as a result of their generality, databases often achieve adequate performance for the average use case but do not achieve the best performance for any individual use case. In this talk, I will describe my work on designing databases that use machine learning and optimization techniques to automatically achieve performance much closer to the optimal for each individual use case. In particular, I will present my work on instance-optimized database storage layouts, in which the co-design of data structures and optimization policies improves query performance in analytic databases by orders of magnitude. I will highlight how these instance-optimized data layouts address various challenges posed by real-world database workloads and how I implemented and deployed them in production within Amazon Redshift, a widely-used commercial database system. Bio: Jialin Ding is an Applied Scientist at AWS. Before that, he received his PhD in computer science from MIT, advised by Tim Kraska. He works broadly on applying machine learning and optimization techniques to improve data management systems, with a focus on building databases that automatically self-optimize to achieve high performance for any specific application. His work has appeared in top conferences such as SIGMOD, VLDB, and CIDR, and has been recognized by a Meta Research PhD Fellowship. To learn more about Jialin’s work, please visit [ https://jialinding.github.io/ | https://jialinding.github.io/ ] . CSML/CS/PSY Colloquium Speaker: Brenden Lake, New York University Date: Tuesday, February 27 Time: 12:30pm EST Location: CS 105 Host: Tom Griffiths Event page: https://csml.princeton.edu/events/csmlpsycs-seminar Register for live-stream online here: https://princeton.zoom.us/webinar/register/WN_aUfLfVQFSIqV2RDSYu4big Title: Towards more human-like learning in machines: Bridging the data and generalization gaps Abstract: There is an enormous data gap between how AI systems and children learn: The best LLMs now learn language from text with a word count in the trillions, whereas it would take a child roughly 100K years to reach those numbers through speech (Frank, 2023, "Bridging the data gap"). There is also a clear generalization gap: whereas machines struggle with systematic generalization, children can excel. For instance, once a child learns how to "skip," they immediately know how to "skip twice" or "skip around the room with their hands up" due to their compositional skills. In this talk, I'll describe two case studies in addressing these gaps. 1) The data gap: We train deep neural networks from scratch (using DINO, CLIP, etc.), not on large-scale data from the web, but through the eyes and ears of a single child. Using head-mounted video recordings from a child as training data (<200 hours of video slices over 26 months), we show how deep neural networks can perform challenging visual tasks, acquire many word-referent mappings, generalize to novel visual referents, and achieve multi-modal alignment. Our results demonstrate how today's AI models are capable of learning key aspects of children's early knowledge from realistic input. 2) The generalization gap: Can neural networks capture human-like systematic generalization? We address a 35-year-old debate catalyzed by Fodor and Pylyshyn's classic article, which argued that standard neural networks are not viable models of the mind because they lack systematic compositionality -- the algebraic ability to understand and produce novel combinations from known components. We'll show how neural networks can achieve human-like systematic generalization when trained through meta-learning for compositionality (MLC), a new method for optimizing the compositional skills of neural networks through practice. With MLC, neural networks can match human performance and solve several machine learning benchmarks. Given these findings, we'll discuss the paths forward for building machines that learn, generalize, and interact in more human-like ways based on more natural input. Related articles: Vong, W. K., Wang, W., Orhan, A. E., and Lake, B. M (2024). Grounded language acquisition through the eyes and ears of a single child. Science, 383, 504-511. Orhan, A. E., and Lake, B. M. (in press). Learning high-level visual representations from a child’s perspective without strong inductive biases. Nature Machine Intelligence. Lake, B. M. and Baroni, M. (2023). Human-like systematic generalization through a meta-learning neural network. Nature, 623, 115-121. Bio: Brenden M. Lake is an Assistant Professor of Data Science and Psychology at New York University. He received his M.S. and B.S. in Symbolic Systems from Stanford University in 2009, and his Ph.D. in Cognitive Science from MIT in 2014. He was a postdoctoral Data Science Fellow at NYU from 2014-2017. Brenden is a recipient of the Robert J. Glushko Prize for Outstanding Doctoral Dissertation in Cognitive Science and the MIT Technology Review 35 Innovators Under 35. His research was also selected by Scientific American as one of the 10 most important advances of 2016. Brenden's research focuses on computational problems that are easier for people than they are for machines, such as learning new concepts from just a few examples, learning by asking questions, learning by generating new goals, and learning by producing novel combinations of known components. CS Colloquium Speaker: Eric Mitchell, Stanford University Date: Thursday, February 29 Time: 12:30pm EST Location: CS 105 Host: Karthik Narasimhan Event page: https://www.cs.princeton.edu/events/26587 Register for live-stream online here: TBA Talk info TBA

1 23

PPPL Seminar - Sunita Chandrasekaran, January 30, 2:45 PM
by Emily C. Lawrence 29 Jan '25

29 Jan '25

The Princeton Plasma Physics Laboratory's Computational Science Department is pleased to announce. [ https://sites.google.com/d/13WaEf9N_dRV6D56spW_7SebByoF71VQe/p/1u7bBH2T5Z_D… | The Artificial Scientist - Leveraging In-transit Machine Learning for Plasma Simulations ] Presented by: Sunita Chandrasekaran , Associate Professor , University of Delaware Thursday, January 30, 2025, 2:45 PM Richard Hawryluk Conference Room, B31 8, [ https://pppl.zoom.us/j/91058597465?pwd=bLV9xHE4WFbznMBZFh45MJl4jiZ1pg.1 | Zoom ] Host: Shantenu Jha Abstract: With the rapid advancements in the computer architecture space, the migration of legacy applications to new architectures remains a continuous challenge. To effectively navigate this ever-evolving hardware landscape, software and toolchains must evolve in tandem, staying ahead of the curve in terms of architectural innovation. While this synchronization between hardware and software is inherently complex, it is essential for fully harnessing the potential of advanced hardware platforms. In this context, a marriage between HPC and AI is gaining increasing prominence. By effectively orchestrating the workflow of HPC and AI, we can not only accelerate scientific progress but also achieve significant gains in computational efficiency. One promising strategy to further optimize large-scale workflows is to stream simulation data directly into machine learning (ML) frameworks. This approach bypasses traditional file system bottlenecks, allowing for the transformation of data in transit— asynchronously with both the simulation process and model training. This talk will explore these strategies in detail, demonstrating the synergy between hardware innovation and software adaptation. Using real-world scientific applications as case study, Plasma-in-Cell on GPU, i.e. PIConGPU will showcase how these techniques can be applied at scale to drive both scientific and computational advancements. Sponsorship of an event does not constitute institutional endorsement of external speakers or views presented.

1 0

Yun Cheng will present her General Exam "Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs?" on Wednesday, January 22, 2025 at 10:00 AM in CS 402 and via zoom.
by CS Grad Department 17 Jan '25

17 Jan '25

Yun Cheng will present her General Exam "Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs?" on Wednesday, January 22, 2025 at 10:00 AM in CS 402 and via zoom. Zoom link: https://princeton.zoom.us/j/91346752488 Committee Members: Sanjeev Arora (advisor), Danqi Chen, Benjamin Eysenbach Abstract: While Vision Language Models (VLMs) are impressive in tasks such as visual question answering (VQA) and image captioning, their ability to apply multi-step reasoning to images has lagged, giving rise to perceptions of modality imbalance or brittleness. Towards systematic study of such issues, we introduce a synthetic framework for assessing the ability of VLMs to perform algorithmic visual reasoning (AVR), comprising three tasks: Table Readout, Grid Navigation, and Visual Analogy. Each has two levels of difficulty, SIMPLE and HARD, and even the SIMPLE versions are difficult for frontier VLMs. We seek strategies for training on the SIMPLE version of tasks that improve performance on the corresponding HARD task, i.e., S2H generalization. This synthetic framework, where each task also has a text-only version, allows a quantification of the modality imbalance and how it is impacted by training strategy. Ablations highlight the importance of explicit image-to-text conversion in promoting S2H generalization when using auto-regressive training. We also report results of mechanistic study of this phenomenon, including a measure of gradient alignment that seems to identify training strategies that promote better S2H generalization. Reading List: https://docs.google.com/document/d/1DbFfQZyCJb3HSYLvhy2SZ9voh_IN_6JZXcrFoiy… 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

Amaya Kelegedara will present her General Exam "The Impact of Coreset Selection on Bias and Group Robustness" on Friday, January 17, 2025 at 3:00 PM in CS 105 and via zoom.
by CS Grad Department 14 Jan '25

14 Jan '25

Amaya Kelegedara will present her General Exam "The Impact of Coreset Selection on Bias and Group Robustness" on Friday, January 17, 2025 at 3:00 PM in CS 105 and via zoom. Zoom link: https://princeton.zoom.us/j/95739855237 Committee Members: Olga Russakovsky (advisor), Lydia Liu, Ryan Adams Abstract: Coreset selection methods have shown promise in reducing the training data size while maintaining model performance for data-efficient machine learning. However, many real-world datasets contain spurious features and unintended biases, and how coreset selection algorithms could impact the group robustness of their respective downstream models remains unexplored. I will talk about my research project where we conducted the first extensive analysis of the implications of coreset selection on the downstream group-robustness in the presence of spurious correlations. Our findings point to counterintuitive interactions between data bias and data quality, highlighting the importance of evaluating biases and robustness in coreset selection. Reading List: https://docs.google.com/document/d/1SMJIM-L98tOrmW0xY9P0hXqyymvaVyEB33Z71QA… 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

Wei Tang will present his FPO "Enabling Large-scale Quantum Computing via Distributed and Hybrid Architectures" on Monday, January 27, 2025 at 10am in CS 302.
by Gradinfo 14 Jan '25

14 Jan '25

Wei Tang will present his FPO "Enabling Large-scale Quantum Computing via Distributed and Hybrid Architectures" on Monday, January 27, 2025 at 10am in CS 302. The members of his committee are as follows: Examiners: Margaret Martonosi (adviser), Jeff Thompson (Princeton ECE), Kai Li Readers: Fred Chong (Univ of Chicago), Amit Levy All are welcome to attend. Abstract follows below. Quantum Computing (QC) is an emerging paradigm that may o!er significant runtime advantages over classical computing by harnessing quantum physics principles to perform computations. Through years of software and hardware developments, QC has achieved notable progress. The theoretical computational advantages of QC over classical computing only manifest themselves as greater e”ciency when handling large and complex applications. Yet, QC encounters significant challenges in scaling and maintaining the precision necessary to fulfill its potential. In simpler or smaller problems, these theoretical advantages might not present a clear benefit over traditional methods. Such complex applications demand large Quantum Processing Units (QPUs). Moreover, in contrast to the exceptionally low error rates of classical computers, current QPUs are much more susceptible to errors. This vulnerability necessitates QPUs that not only support large-scale tasks but also mitigate error accumulation. The dual requirements of size and accuracy put a heavy toll on QC, hindering its adoption in practical scenarios. Conversely, classical computing is characterized by its high precision and reliability across various applications. Nonetheless, it encounters intrinsic limitations in computational complexity for solving complex, real-world challenges. In fact, the best-known classical algorithms for many important real-world problems remain intractable as the size of the problems increases. Despite decades of semiconductor technology advancements, the miniaturization of chip components to enhance their processing capacity is approaching its physical boundaries. This development poses a challenge to classical computing’s ability to keep pace with increasing computational demands, highlighting the need for innovative paradigms such as QC. This dissertation tackles the challenges of scaling up QC by distributing quantum algorithms across quantum and classical resources. The distributed hybrid approach carries two main benefits. First, it significantly reduces the quantum resource requirements compared to an exclusively quantum setup. Second, it achieves potentially faster runtimes than those attainable on purely classical platforms. This dissertation pioneers the field of Distributed Hybrid QC (DHQC) and develops the first end-to-end DHQC toolchain. By integrating with cutting-edge classical computing methods, this dissertation further enhances the scalability of DHQC systems. Additionally, this dissertation also contributes to the evolution of QC compiler design, optimizing QPU performance towards practical applications.

1 0

Tyler Zhu will present his General Exam "Unifying Specialized Visual Encoders for Video Language Models" on Tuesday, January 14, 2025 at 10:00 AM in CS 402 and via zoom.
by CS Grad Department 09 Jan '25

09 Jan '25

Tyler Zhu will present his General Exam " Unifying Specialized Visual Encoders for Video Language Models " on Tuesday, January 14, 2025 at 10:00 AM in CS 402 and via zoom. Zoom link: https://princeton.zoom.us/my/tylerzhu Committee Members: Olga Russakovsky (advisor), Jia Deng , Danqi Chen Abstract: The recent advent of Large Language Models (LLMs) has ushered sophisticated reasoning capabilities into the realm of video through Video Large Language Models (VideoLLMs). However, VideoLLMs currently rely on a single vision encoder for all of their visual processing, which limits the amount and type of visual information that can be conveyed to the LLM. Our method, MERV, Multi-Encoder Representation of Videos, instead leverages multiple frozen visual encoders to create a unified representation of a video, providing the VideoLLM with a comprehensive set of specialized visual knowledge. Spatio-temporally aligning the features from each encoder allows us to tackle a wider range of open-ended and multiple-choice video understanding questions and outperform prior state-of-the-art works on their data mixes. MERV is up to 3.7% better in accuracy than Video-LLaVA across the standard suite video understanding benchmarks, while also having a better Video-ChatGPT score. We also improve upon SeViLA, the previous best on zero-shot Perception Test accuracy, by 2.2%. MERV introduces minimal extra parameters and trains faster than equivalent single-encoder approaches. Finally, we provide qualitative evidence that our model captures domain knowledge from each encoder simultaneously, such as on the motion classification tasks found in Something-Something v2. Our results offer promising directions for future research in utilizing multiple vision encoders for comprehensive video understanding. Reading List: https://docs.google.com/document/d/1jBXhv-IhFRsFsC6HblTb9ZE4nZkQIfGgm2-NAMO… 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