Computer Science Colloquium

08.29.24: How to design, document, and implement programming languages
Speaker: Sukyoung Ryu
Host: Nate Foster
Abstract: Since 2015, the JavaScript language has rapidly evolved with a yearly release cadence and open development process. However, it results in the gap between the language specification written in English and tools, such as parsers, interpreters, and static analyzers, which makes language designers and tool developers suffer from manually filling the gap. JISET and its extensions lessen the burden by automatically extracting a mechanized specification from the language specification in prose.

09.05.24: Low-depth quantum circuit constructions and 2D architectures, with applications
Speaker: Mark Wilde
Host: Eshan Chattopadhyay
Abstract: Due to the low coherence times of present-day quantum computers, it seems necessary to reduce quantum circuit depth to be as small as possible in order to increase the reliability of quantum computers. In this talk, I will present some basic ideas for doing so, which date back to Shor's late-1990s constructions for fault-tolerant quantum computing. After that, I will illustrate these quantum circuit constructions for one main application: multivariate trace estimation. This application concerns estimating the trace of a product of quantum states, known to be a BQP-complete problem. Finally, I will sketch a 2D architecture design for these circuit constructions, with the aim of motivating more in-depth studies of this problem from a quantum computer architecture perspective. 
References: https://arxiv.org/abs/2404.07151 , https://arxiv.org/abs/2206.15405

09.12.24: My Six Decade Experience in Visual Imaging
Speaker: Donald P. Greenberg
Host: Steve Marschner
Abstract: Knowledge of the fundamentals of computer science and the availability of the new applications and programming tools is now and will continue to be as important as the reading and writing requirements for entering Freshman. This is already evident with the fact that CS/CIS already teach at least one course to 76% of Cornell’s undergraduate students.
I have now had the fantastic opportunity to teach at Cornell for more than six decades in four colleges and five departments and this now probably my final year. During this more than half-century I have been blessed with superb mentors and excellent students from many unrelated disciplines, but also confronted the hurdles of interdisciplinary barriers which are about to change in the very near future.
Two of my students have won the most prestigious computer graphics awards, eighteen of my former students have won Hollywood’s technical Oscars or Emmy’s, and I have collaborated with or taught at the ETH, Stanford, Hewlett-Packard, Autodesk, Nvidia, Intel to name a few. More recently I have been working with computer scientists, VR and AR experts,  roboticists,  perception psychologists, neuroscientists, experts in the Medical industry, structural and mechanical engineers, as well as artist and photographers and graphic designers. As the boundaries between disciplines become “fuzzier and more blurred”, academia in general, and computer science in particular must adapt! . We owe this to our students! In this talk I will share my experiences and thoughts about education in the field of computer science.

09.19.24: Machine learning for discovery: deciphering RNA splicing logic
Speaker: Oded Regev
Host: Noah Stephens-Davidowitz
Abstract: Recent advances in machine learning, such as deep learning, have led to powerful tools for modeling complex data with high predictive accuracy. However, the resulting models are typically black box, limiting their usefulness in scientific discovery. I will describe an "interpretable-by-design'' machine learning model capturing a fundamental cellular process known as RNA splicing. Our model provides a systematic understanding of RNA splicing logic, recapitulating and extending existing domain knowledge.

10.03.24: Multimodal Spatial Intelligence for Interacting in a Dynamic World
Speaker: Deva Ramanan
Host: Abe Davis
Abstract: Artificial intelligence and machine learning are enjoying a period of tremendous progress, driven in large part by scale, compute, and learnable neural representations. However, such innovations have yet to translate to the physical world, as technologies such as self-driving vehicles are still restricted to limited deployments. In this talk, I will argue that autonomy requires spatial three-dimensional understanding integrated with intuitive physical models of a changing world. To do so, I will discuss a variety of models that revisit classic "analysis by synthesis" approaches to scene understanding, taking advantage of recent advances in differentiable rendering and simulation.  But to enable data-driven autonomy for safety-critical applications, I will also argue that the community needs new perspectives on data curation and annotation. Toward this end, I will discuss approaches that leverage multimodal vision-language models to better characterize datasets and models.

10.10.24: Testing with Large Language Models, Symbolic Execution, and Fuzzing
Speaker: Koushik Sen
Host: Owolabi Legunsen
Abstract: Automation has significantly impacted software testing and analysis in the last two decades.  Automated testing techniques, such as symbolic execution, concolic testing, and feedback-directed fuzzing, have found numerous critical faults, security vulnerabilities, and performance bottlenecks in mature and well-tested software systems. The key strength of automated techniques is their ability to quickly search state spaces by performing repetitive and expensive computational tasks at a rate far beyond the human attention span and computation speed. In this talk, I will briefly overview our past and recent research contributions in automated test generation using large-language models, symbolic execution, program analysis, constraint solving, and fuzzing.  We have combined these techniques to find and rescue $11M from DeFI Smart Contracts.

10.17.24: Conveying Tasks to Computers: How Machine Learning Can Help
Speaker: Michael Littman
Host: Wen Sun
Abstract: It is immensely empowering to delegate information processing work to machines and have them carry out difficult tasks on our behalf. But programming computers is hard. The traditional approach to this problem is to try to fix people: They should work harder to learn to code. In this talk, I argue that a promising alternative is to meet people partway. Specifically, powerful new approaches to machine learning provide ways to infer intent from disparate signals and could help make it easier for everyone to get computational help with their vexing problems.

10.24.24: An Overview of High Performance Computing and Responsibly Reckless Algorithms
Speaker: Jack Dongarra
Host: Giulia Guidi
Abstract: In this talk we examine how high performance computing has changed over the last 10-year and look toward the future in terms of trends. These changes have had and will continue to have a major impact on our software.  Some of the software and algorithm challenges have already been encountered, such as management of communication and memory hierarchies through a combination of compile--time and run-time techniques, but the increased scale of computation, depth of memory hierarchies, range of latencies, and increased run--time environment variability will make these problems much harder.  
Mixed precision numerical methods turn out to be paramount for increasing the throughput of traditional and artificial intelligence (AI) workloads beyond riding the wave of the hardware alone. Reducing precision comes at the price of trading away some accuracy for performance (reckless behavior) but in noncritical segments of the workflow (responsible behavior) so that the accuracy requirements of the application can still be satisfied.

10.31.24: Inference for an Algorithmic Fairness-Accuracy Frontier
Speaker: Francesca Molinari
Host: Eva Tardos
Abstract: Decision-making processes increasingly rely on the use of algorithms. Yet, algorithms' predictive ability frequently exhibits systematic variation across subgroups of the population. While both fairness and accuracy are desirable properties of an algorithm, they often come at the cost of one another, with policymakers needing to assess this trade-off based on finite data.   We provide a consistent estimator for a theoretical fairness-accuracy frontier put forward in the recent Economics literature, derive its asymptotic distribution, and propose inference methods to test hypotheses that have received much attention in the fairness literature, such as (i) whether fully excluding group identity from use in training the algorithm is optimal and (ii) whether there are less discriminatory alternatives to an existing algorithm. We also provide an estimator for the distance between a given algorithm and the fairest point on the frontier and characterize its asymptotic distribution.  

11.07.24: When You've Got EqSat, Everything Is a Compiler
Speaker: Zachary Tatlock
Host: Nate Foster
Abstract: We are surrounded by compilers in disguise. Whether optimizing circuits, planning database queries, designing machine-knit garments, automating proofs, or generalizing user demonstrations, these processes all rely on transforming program-like models.
In this talk, I will share my group's experiences across several domains where the compiler mindset has led to new techniques, tools, and cross-area connections. Over more than a decade, my students have followed this overarching research theme, eventually leading to the development of "egglog", a general equality saturation (EqSat) framework for building program verifiers, synthesizers, and optimizers, now applied across many fields.

11.12.24: Incorporating Behavioral Science into Computational Science
Speaker: Sendhil Mullainathan
Abstract: We are surrounded by compilers in disguise. Whether optimizing circuits, planning database queries, designing machine-knit garments, automating proofs, or generalizing user demonstrations, these processes all rely on transforming program-like models.
In this talk, I will share my group's experiences across several domains where the compiler mindset has led to new techniques, tools, and cross-area connections. Over more than a decade, my students have followed this overarching research theme, eventually leading to the development of "egglog", a general equality saturation (EqSat) framework for building program verifiers, synthesizers, and optimizers, now applied across many fields.

11.14.24: On the Limits of Function Approximation in Large-Scale MDP Planning and Reinforcement Learning
Speaker: Csaba Szepesvári
Host: Wen Sun
Abstract: At the dawn of the computer age in the 1960s, Bellman and his collaborators found it beneficial to use what is now called linear function approximation to address certain multistage stochastic planning problems. Their approach was straightforward: use linear value function approximation to avoid state-space discretization, thereby maintaining polynomial-time computation while also controlling accuracy. However, the question of when and how this approach is feasible has eluded researchers for over 50 years, even as the prospect of using function approximation to overcome the curse of dimensionality has continued to fuel much of the excitement around reinforcement learning. Early results focused on connecting the approximation spaces with the structure of the underlying problem, and some indicated that it might not be enough for the target function (such as the optimal value function) to simply lie within this space. As it turns out, the emerging picture of when function approximation can help in multistage problems is intricate. In this talk, we will explore recent results, primarily from my group, that contribute to this complex understanding. I will conclude with an outlook on current research directions.

11.21.24: First-Person Fairness in Chatbots
Speaker: Adam Tauman Kalai|
Host: Michael Kim
Abstract: Much research on fairness has focused on institutional decision-making tasks, such as resume screening. Meanwhile, hundreds of millions of people use chatbots like ChatGPT for very different purposes, ranging from resume writing and technical support to entertainment. We study “first-person fairness,” which means fairness toward the user who is interacting with a chatbot. This includes providing high-quality responses to all users regardless of their identity or background, and avoiding harmful stereotypes. We propose a scalable, privacy-preserving method for evaluating one aspect of first-person fairness across a large, heterogeneous corpus of real-world chatbot interactions. Specifically, we assess potential bias linked to users’ names—which can serve as proxies for demographic attributes like gender or race—in chatbot systems like ChatGPT that can store and utilize user names. Our method leverages a second language model to privately analyze name-sensitivity in the chatbot’s responses. We verify the validity of these annotations through independent human evaluation. In addition to quantitative bias measurements, our approach also identifies common tasks, such as "career advice" or "writing a story" and gives succinct descriptions of subtle response differences across tasks. Finally, we publish the system prompts necessary for others to conduct similar experiments that faithfully simulate ChatGPT conversations with arbitrary user profiles.

12.05.24: CS Ph.D. Students Colloquium
Ph. D. Student: Katie Luo; Princewill Okoroafor
Abstract: While machine learning has shown great advances in a variety of fields such as computer vision, the current paradigm for self-driving trains the perception systems on specific environments but then deploys them to end-users into a diverse set of vehicle behavior and appearances. This change in environment makes it hard to guarantee high accuracy outside of the development laboratory. My work focuses on exploring additional channels of information to adapt to diverse, real-world scenarios in a data efficient manner. In this talk, I will go over the motivation for why we need to consider the different ways a user can deploy self driving, and explore the challenges associated with it. To address such challenges, I will discuss a few works proposing datasets, as well as solutions for adapting to these diverse cases of deployment. Finally, I will end the talk with future directions and considerations to bring self driving out of constrained settings and integrating with in-the-wild settings.

12.12.24: Ph. D. Students Colloquium
Ph. D. Students: Xixi Deng and Gloire Rubambiza
Abstract: Reconstructing the real world is a crucial yet challenging problem for applications in both entertainment and science. To address this challenge, we study inverse problems in transport theory, including radiative transfer and neutron transport. In these problems, the field (flow of energy) is described by transport equations. Our key challenge lies in determining the parameters of these equations from measured radiation data, such as transmittance and reflectance. This work has applications across multiple disciplines. I will present our research on reconstructing the geometry and optical properties of semi-translucent and thin objects from photographs, and discuss our recent advances in using Monte Carlo simulation gradients to optimize nuclear reactor shielding design.