IDEAL Annual Meeting and Industry Day 2024

Thursday, June 6th, 2024

Logistics

Date: Thursday, June 6th, Day 1: IDEAL Annual Meeting and Friday, June 7th, Day 2: IDEAL Industry Day 

Location:  Academic and Residential Complex (ARC) 241, University of Illinois Chicago, 940 W. Harrison St.

Parking: For those driving to the workshop, attendees can park in any lot with visitor access. The closest options to the ARC are Lots 1B and the Harrison Street lot. Please refer to the image at the end of the page for marked parking structures.

Parking passes will be provided at the workshop for free parking in designated UIC parking buildings. Please remember to ask for a pass before leaving the workshop.

Registration: click here to register 

Day 1: IDEAL Annual Meeting (June 6)

Synopsis

The IDEAL Annual Meeting day is an in-person-only meeting held every June in Chicago that brings together all researchers, students, and members affiliated with the institute.  We start the day off with a “state of the institute” address by the site directors, followed by distinguished external keynotes and recent research highlights from IDEAL members. There will be plenty of time to meet other members over breakfast, coffee breaks and lunch together. 

Schedule 

 
9:00 am – 9:30 am Arrival, Casual Breakfast 
9:30 am – 10:00 am IDEAL Update (Site Directors)
10:00 am – 11:00 am

Lightning Talks: Thrust 1: Foundations of Machine Learning

(10 mins)  Srebro, TTIC, Interpolation Learning with Neural Networks 

(10 mins)  Blum, TTIC, On the Vulnerability of Fairness Constrained Learning to Adversarial Corruptions 

(10 mins)  Gao, UChicago, Is adaptive robust confidence interval possible? 

(10 mins)  Gutfriand, UIC, Analytical complexity of decision problems under uncertainty 

(10 mins)  Wei, Northwestern, Federated learning 
11:00 am – 11:30 am Coffee Break
11:30 am – 12:15pm

Keynote: Sendhil Mullainathan, University of Chicago

Do Large Language Models Understand and How would we Know if They Did?
12:15 pm – 1:30 pm Lunch: Senior-Junior Lunch
1:30 pm – 2:15 pm

IDEAL Student talks

(3 mins) Yunis, TTIC

(3 mins) Ravichandran, TTIC

(3 mins) Manoj, TTIC

(3 mins) Patel, TTIC

(3 mins) Ovsiankin, TTIC

(3 mins) Dutz, TTIC

(3 mins) Ahmadi, TTIC
2:15 pm – 2:45 pm Lightning Talks: Thrust 2: High-dimensional Data Analysis and Inference

(10 mins)  Zheleva, UIC, Causal Discovery from Network Data 

(10 mins)  Orecchia, U Chicago, Non-linear Diffusions and Spectral Methods with Applications 

(10 mins)  Jia,  Northwestern
2:45 pm – 3:15 pm Coffee Break
3:15 pm – 4:15 pm

Lightning Talks: Thrust 3: Data Science and Society

(10 mins) Cetin, UIC, Wildfire Recognition 

(10 mins)  Diana, TTIC, Adaptive Algorithmic Interventions for Escaping Pessimism Traps in Dynamic Sequential Decisions 

(10 mins)  Linna, Northwestern, Developing Trustworthy and Reliable AI Systems for Law

(10 mins)  Cohen, U Chicago, Enhancing Watermarked Language Models to Identify Users

(10 mins)  Shan, TTIC, Error-Tolerant E-Discovery Protocols 

4:15 pm – 5:00 pm

IDEAL Student Talks

(3 mins) Fournier, UIC

(3 mins) Wu, UIC

(3 mins)  Duan Tu, UIC

(3 mins) Chai, Northwestern

(3 mins) Charlie Guan, Northwestern

(3 mins) Jia, IIT

(3 mins) Yao, U Chicago

 

 

Keynote speaker

Sendhil Mullainathan

Title: Do Large Language Models Understand and How would we Know if They Did?

Bio: Sendhil Mullainathan is the Roman Family University Professor of Computation and Behavioral Science at Chicago Booth, where he is also the inaugural Faculty Director of the Center for Applied Artificial Intelligence. His latest research is on computational medicine—applying machine learning and other data science tools to produce biomedical insights. In past work he has combined insights from behavioral science with empirical methods—experiments, causal inference tools, and machine learning—to study social problems such as discrimination and poverty. He currently teaches a course on Artificial Intelligence.

Outside of research, he co-founded a non-profit to apply behavioral science (ideas42), a center to promote the use of randomized control trials in development (the Abdul Latif Jameel Poverty Action Lab), has worked in government in various roles, and currently serves on the board of the MacArthur Foundation board. He is also a regular contributor to the New York Times.

YouTube:

 

 

Day 2: IDEAL Industry Day (June 7)

 

Synopsis

The IDEAL Industry Day provides an opportunity for IDEAL researchers to explore relationships with members from industry who are tackling problems in data science. It will feature keynotes and lightning talks by a group of industry researchers and data scientists from a diverse array of disciplines along with an interactive poster session led by IDEAL researchers. In this way, the Industry Day event will enable IDEAL faculty, postdocs, and students to learn about industry priorities and perspectives. At the same time, it will expose corporate partners to foundational research in data science that spans computer science, electrical engineering, mathematics, and statistics.

Schedule 

8:30 am – 9:00 am Breakfast (Provided) 
9:00 am – 9:10 am  Welcoming Remarks
9:10 am – 10:00 am 

Keynote: Ian Foster (Argonne National Laboratory) 

  Embodied agents as scientific assistants
10:00 am – 10:15 am Coffee Break 
10:15 am – 10:35 am

Holli Knight, UL Research Institutes 

  NERIS: A Modern Data Platform for US Fire & Emergency Response Services
10:35 am – 10:55 am 

Nhan Tran, Fermi Labs 

  Accelerating Scientific Discoveries with Real-Time Intelligent Sensing
10:55 am – 11:15am

Aadirupa Saha, Apple 

  Principled Methods for Leveraging Human Feedback towards AI Alignment 
11:15 am – 11:35 am 

Dianqi Li, Citadel Securities

  Demystify financial textual data with LLMs
11:35 am – 11:55 am

Harrison Pielke-Lombardo, Peruse Technologies Inc

  AI in the Transportation Sector: Challenges and Successes
12:00 pm – 1:10 pm Lunch 
1:10pm – 2:00pm 

Keynote: Karthikeyan Natesan Ramamurthy (IBM) 

  Trust and Governance in Foundation Models
2:00 pm – 2:15 pm Coffee Break 
2:15 pm – 2:35 pm

Eve Jennings, City of Chicago

   Data Services Strategy at the City of Chicago
2:35 pm – 2:55 pm

Bo Chen, ComEd

  Next-Generation Technologies for Equitable and Clean Energy
2:55 pm – 3:15 pm
Maggie Wolff, American Express Global Business Travel 
  Measuring the User Experience and the Impact of Effort on Business Outcomes
3:15 pm – 3:35pm

Ziqing Hu, Inai North America

   The Industrial Applications of Generative AI
3:35 pm – 3:55 pm

Khashayar Filom, Discover Financial Services

On marginal feature attributions of tree-based models
4:00 pm – 5:30 pm  Interactive Poster Session/Reception 
Abstracts
 
Speaker: Ian Foster, Argonne
Title: Embodied agents as scientific assistants
Abstract: An embodied agent is a computational entity that can interact with the world through a physical body or representation and adapt its actions based on learnings from these interactions. I discuss the potential for such agents to serve as next-generation scientific assistants, illustrating my presentation with work at Argonne, where we are, among other things, applying supercomputers to train large language models that we employ to drive experiments in robotic laboratories. 
 
Speaker: Karthikeyan Natesan Ramamurthy, IBM
Title: Trust and Governance in Foundation Models
Abstract: The rapid pace of development in foundation models has caught the imagination of the public. The reception for these models range from unbridled enthusiasm to deep mistrust. But what does it take to really trust them? How can we effectively govern them? We will take an industry perspective in this talk and provide examples of the promises and perils of these models. We will continue our discussion into understanding the ingredients of trust and governance, and how they can be incorporated into the AI lifecycle. We will also touch upon the open movement that has been gathering steam and discuss the pathways through which it can end up empowering users.
 
Speaker: Holli Knight, UL Research Institutes
Title: NERIS: A Modern Data Platform for US Fire & Emergency Response Services Abstract: The development of the NERIS (National Emergency Response Information System) platform aims to modernize data collection and analysis for fire and emergency response services in the United States. This talk will detail the reasons for building such a system, the types of data being collected and augmented, and the machine learning applications that are currently in development to assist stakeholders with understanding the nature of these problems in United States.
 
Speaker: Aadirupa Saha, Apple
Title: Principled Methods for Leveraging Human Feedback towards AI Alignment 
Abstract: With the increasing need of incorporating AI systems for the betterment of human life, it is becoming pervasively crucial for modern ML to make these two powerful entities, human and AI, communicate and interpret each other seamlessly—for efficiency, safety, reliability and mitigate any unfavorable system behavior. This consequently requires machines to interpret human feedback and embed inference capability to serve humans in a desired way, more technically known as the Human-AI alignment problem. “Misaligned AI” can lead to disasters ranging from spam misclassification to more serious outcomes like toxic language model (chatbot) behavior, improper face recognition, robot accidents, autonomous vehicle crashes, amongst many. But how to make AI communicate with and learn from humans in a principled way? While human expressions/feedback can be encoded through several mediums, like rating, demonstration, multiple choices, ranking, engagement time, etc., — many studies in psychology, brain, and neuroscience have shown that we are most comfortable, unbiased and fastest in expressing opinions on a relative scale or through preference feedback, e.g. “Do you prefer Item A over B?”, rather than their absolute rating counterparts: “How much do you score items A and B on a scale of [0-10]?”. Thus in this talk, we will build on a mathematical framework for developing efficient optimal and adaptive prediction algorithms with human preferences and analyze their theoretical guarantees. The goal of this work is to develop simultaneously runtime efficient and near-optimal personalized prediction algorithms exploiting actively queried human preferences. Our results are derived from lifting tools from different interdisciplinary literature including Online learning, Bandits, Reinforcement learning, Game theory, Optimization, Federated learning, Differential privacy, Social Choice theory, and Operations research.
 
Speaker: Ziqing Hu, INAI, North America
Title: The Industrial Applications of Generative AI
Abstract: This presentation explores the industrial applications of generative AI, including its fundamental technologies and implementation strategies. It introduces generative AI models and highlights their capabilities compared to traditional machine learning. Practical examples illustrate how these models can be integrated with SQL and Python for data analysis. Additionally, the presentation covers the potential of generative AI to enhance productivity, creativity, and business processes across various sectors, while also mentioning potential challenges related to responsible AI, data quality, and scalability. 
 
Khashayar Filom, Discover Financial Services
Title: On marginal feature attributions of tree-based models
Abstract: Game-theoretic feature attributions allow one to rank the variables of an ML model for any data point. There are three different ingredients in this framework: the model in hand, the game defined based on the model (e.g. conditional or marginal) whose players are the features, and the game value quantifying each player’s importance (e.g. Shapley). Focusing on tree-based models, we axiomatically characterize game values suitable for explaining them in terms of marginal feature attributions. More importantly, we argue that such feature attributions can be computed explicitly when the trees are symmetric (oblivious). This results in a fast and accurate algorithm for explaining ensemble of symmetric decision trees (e.g. CatBoost models). This research project was carried out at Discover’s R&D group jointly with A. Miroshnikov, K. Kotsiopoulos and A. Ravi Kannan.

YouTube:

 

About IDEAL

The Institute for Data, Econometrics, Algorithms, and Learning (IDEAL) is a multi-institution and transdisciplinary institute led by the University of Illinois Chicago in collaboration with Northwestern University; Toyota Technological Institute at Chicago; the University of Chicago; and Illinois Institute of Technology, in partnership with members of the Learning Theory team at Google. The institute involves more than 60 researchers working on key aspects of the foundations of data science across computer science, electrical engineering, mathematics, statistics, and fields such as economics, operations research and law. Research will center around the foundations of machine learning, high-dimensional data analysis and inference, and data science and society. Topics include foundations of deep learning, reinforcement learning, machine learning and logic, network inference, high-dimensional data analysis, trustworthiness and reliability, fairness, and data science with strategic agents.

 

Parking visual for UIC:

 

 

Join Our Newsletter