Description:

Friday / CS Seminar
January 31st / 12:00 PM
Hybrid / Mudd 3514

Speaker
Daniel Halpern, Harvard University

Talk Title
Why AI Needs Social Choice

Abstract
In many modern AI paradigms, we encounter tasks reminiscent of social choice theory: collecting preferences from individuals and aggregating them into a single joint outcome. However, these tasks differ from traditional frameworks in two key ways: the space of possible outcomes is so enormous that we can only hope to collect sparse inputs from each participant, and the outcomes themselves are often highly complex. This talk explores these challenges through two case studies: Polis, a platform for democratic deliberation (https://arxiv.org/abs/2211.15608), and Reinforcement Learning From Human Feedback (RLHF), a method for fine-tuning LLMs to align with societal preferences (https://arxiv.org/pdf/2405.14758). In both cases, the focus is on evaluating existing methods through an axiomatic lens and designing new methods with provable guarantees.

Biography
Daniel Halpern is a final-year PhD student at Harvard University advised by Ariel Procaccia. He is supported by an NSF Graduate Research Fellowship and a Siebel Scholarship. His research broadly sits at the intersection of algorithms, economics, and artificial intelligence. Specifically, he considers novel settings where groups of people need to make collective decisions, such as summarizing population views on large-scale opinion aggregation websites, using participant data to fine-tune large language models, and selecting panel members for citizens’ assemblies. In each, he develops practical and provably fair solutions to aggregate individual preferences.

---
Zoom: https://northwestern.zoom.us/j/96824870974?pwd=QfxpsRpfWcDlx4TXPswbAd8X4Dqhyb.1
Panopto: https://northwestern.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=2159520b-1d46-4ece-ba17-b268017475c6
Community Connections Topic: Being comfortable with being uncomfortable

Description:

Wednesday / CS Seminar
February 5th / 12:00 PM
Hybrid / Mudd 3514

Speaker
Akari Asai, University of Washington

Talk Title
Beyond Scaling: Frontiers of Retrieval-Augmented Language Models

Abstract
Large Language Models (LMs) have demonstrated remarkable capabilities by scaling up training data and model sizes. However, they continue to face critical challenges, including hallucinations and outdated knowledge, which particularly limit their reliability in expert domains such as scientific research and software development. In this talk, I will urge the necessity of moving beyond the traditional scaling of monolithic LMs and advocate for Augmented LMs—a new AI paradigm that designs, trains, and deploys LMs alongside complementary modules to address these limitations. Focusing on my research on Retrieval-Augmented LMs, one of the most impactful and widely adopted forms of Augmented LMs today, I will begin by presenting our systematic analyses of current LM shortcomings and demonstrate how Retrieval-Augmented LMs offer a more effective and efficient path forward. I will then discuss my work to establish new foundations for further reliability and efficiency by designing and training new LMs and retrieval systems to dynamically adapt to diverse inputs. Finally, I will demonstrate the real-world impact of such Retrieval-Augmented LMs through OpenScholar, our fully open Retrieval-Augmented LM designed to assist scientists in synthesizing scientific literature, now used by more than 25,000 researchers and practitioners worldwide. I will conclude by outlining my vision for the future of Augmented LMs, emphasizing advancements in their abilities to handle heterogeneous and diverse modalities, more efficient and effective integration with diverse components, and advancing evaluations with interdisciplinary collaboration.

Biography
Akari Asai is a Ph.D. candidate in the Paul G. Allen School of Computer Science & Engineering at the University of Washington. Her research addresses the limitations of large language models (LMs) by developing advanced systems, such as Retrieval-Augmented LMs, and applying them to real-world challenges, including scientific research and underrepresented languages. Her contributions have received widespread recognition, including multiple paper awards at top NLP and ML conferences, the EECS Rising Stars 2022, and MIT Technology Review's Innovators Under 35  Japan. She has also been honored with the IBM Global Fellowship and several industry grants. Akari actively engages with the research community as a co-organizer of a tutorial and workshops, including the first tutorial on Retrieval-Augmented LMs at ACL 2023, as well as NAACL 2022 Workshop on Multilingual Information Access and NAACL 2025 Workshop on Knowledge-Augmented NLP.

Research/Interest Areas
Natural Language Processing, Machine Learning, Large Language Models
---
Zoom: https://northwestern.zoom.us/j/92736097526?pwd=TEoEMxEcDOanxEAoaNdB4ZIxXGsgwV.1
Panopto: https://northwestern.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=7a06abb9-4dd5-402e-a033-b274015b1e07
Community Connections Topic: Lab counterculture

Description:

Monday / CS Seminar
February 10th / 12:00 PM
Hybrid / Mudd 3514

Speaker
Sidhanth Mohanty, MIT

Talk Title
A quest for an algorithmic theory for high-dimensional statistical inference

Abstract
"When does a statistical inference problem admit an efficient algorithm?
There is an emergent body of research that studies this question by trying to understand the power and limitations of various algorithmic paradigms in solving statistical inference problems; for example, convex programming, Markov chain Monte Carlo (MCMC) algorithms, and message passing algorithms to name a few.

Of these, MCMC algorithms are easy to adapt to new inference problems and have shown strong performance in practice, which makes them promising as a universal algorithm for inference.  However, provable guarantees for MCMC have been scarce, lacking even for simple stylized models of inference.

In this talk, I will survey some recent strides that I have made with my collaborators on achieving provable guarantees for MCMC in inference, and some new tools we introduced for analyzing the behavior of slow-mixing Markov chains."

Biography
"Sidhanth is broadly interested in theoretical computer science and probability theory, and his primary interests are on the algorithms and complexity of statistical inference, and spectral graph theory.

Sidhanth is currently a postdoctoral researcher at MIT, hosted by Sam Hopkins. Previously, he received his PhD in Computer Science at UC Berkeley in 2023 where he was advised by Prasad Raghavendra."

Research/Interest Areas
Theoretical computer science, algorithmic statistics, analysis of Markov chains, spectral graph theory, semidefinite programming, random matrix theory
---

Zoom: https://northwestern.zoom.us/j/98400432043?pwd=t4PaTy1pIWxa9R0QrhxojIjKeJ8pho.1
Panopto: https://northwestern.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=d9a67624-ddf1-4295-bb3a-b27b011768d7
Community Connections Topic: Lab counterculture

Description:

Wednesday / CS Seminar
February 12th / 12:00 PM
Hybrid / Mudd 3514

Speaker
Wenting Zhao

Talk Title
Reasoning in the Wild 

Abstract
In this talk, I will discuss how to build natural language processing (NLP) systems that solve real-world problems requiring complex reasoning. I will address three key challenges. First, because real-world reasoning tasks often differ from the data used in pretraining, I will introduce WildChat, a dataset of reasoning questions collected from users, and demonstrate how training on it enhances language models’ reasoning abilities. Second, because supervision is often limited in practice, I will describe my approach to enabling models to perform multi-hop reasoning without direct supervision. Finally, since many real-world applications demand reasoning beyond natural language, I will introduce a language agent capable of acting on external feedback. I will conclude by outlining a vision for training the next generation of AI reasoning models.

Biography
Wenting Zhao is a Ph.D. candidate in Computer Science at Cornell University, advised by Claire Cardie and Sasha Rush. Her research focuses on the intersection of natural language processing and reasoning, where she develops techniques to effectively reason over real-world scenarios. Her work has been featured in The Washington Post and TechCrunch. She has co-organized several tutorials and workshops, including the VerifAI: AI Verification in the Wild workshop at ICLR 2025 and the Complex Reasoning in Natural Language tutorial at ACL 2023. In 2024, she was recognized as a rising star in Generative AI and was named Intern of the Year at the Allen Institute for AI in 2023.

Research/Interest Areas
natural language processing, AI, reasoning
---
Zoom: https://northwestern.zoom.us/j/93876523769?pwd=T8R2mKPRz0mgkb1Ygjokp3OcRwAUnX.1
Panopto: https://northwestern.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=f01e4368-11b0-45f1-bdb6-b27b0117a5cd
Community Connections Topic: Building a community of care

Description:

Friday / CS Seminar
February 13th / 12:00 PM
Hybrid / Mudd 3514

Speaker
Kaitlyn Zhou, Stanford University

Talk Title
Broadening AI Utility Through Natural Language Interfaces

Abstract
In this talk, I will present the novel dynamics of human interaction with large language models (human-LM interaction), focusing on how these systems shape human decision-making, trust, and reliance. As the world seeks to integrate the innovations of foundation models into everyday work and life, my mission is to design human-centered natural language interfaces to augment human intelligence and democratize access to AI. My work pioneers key advancements in natural language processing and human-computer interaction by: 1) uncovering core algorithmic risks in current human-LM interactions, 2) articulating the factors that complicate human-AI interactions, and 3) proposing new human-LM interactions to serve the needs of a broader population.

Biography
Kaitlyn Zhou is PhD candidate in computer science at Stanford University, advised by Dan Jurafsky. Her contributions have been recognized at top-tier conferences in NLP and HCI. She has received awards such as MIT EECS Rising Star, Stanford Graduate Fellowship, and the College of Engineering Dean’s Medal, and her methods have been featured in high-profile news outlets like the New York Times and Wall Street Journal. Kaitlyn has long advocated for increased access, inclusion, and equity in higher education and was appointed by the Washington State Governor to serve on the University of Washington Board of Regents.

Research/Interest Areas
natural language processing, human-computer interactions
---
Zoom: https://northwestern.zoom.us/j/98362439009?pwd=c89YlFRcBHSe1XH0PEl5rKu0HtuRpA.1
Panopto: https://northwestern.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=6719566d-cf13-45ec-a0f3-b27b0117b56e
Community Connections Topic: Black History Month