News & EventsDepartment Events
Events
-
Jan27
EVENT DETAILSmore info
lessTitle: Bridging causality and deep learning with causal generative models
Bio: Bryon Aragam is an Associate Professor and Topel Faculty Scholar in the Booth School of Business at the University of Chicago. He studies causality, statistical machine learning, and probabilistic modeling. His current interests involve causal machine learning, deep generative models, latent variable models, and statistical learning theory. In particular, this work focuses on applications of artificial intelligence, including tools such as ChatGPT and DALL-E. He is also involved with developing open-source software and solving problems in interpretability, ethics, and fairness in artificial intelligence. His work is supported by the NSF and the NIH. He completed his PhD in Statistics and a Masters in Applied Mathematics at UCLA, where he was an NSF graduate research fellow.
Abstract: Generative models for vision and language have shown remarkable capacities to emulate creative processes but still lack fundamental skills that have long been recognized as essential for genuinely autonomous intelligence. Difficulties with causal reasoning and concept abstraction highlight critical gaps in current models, despite their nascent capacities for reasoning and planning. Bridging this gap requires a synthesis of deep learning’s expressiveness with the powerful framework of statistical causality.
We will discuss our recent efforts towards building generative models that extract causal knowledge from data while retaining the flexibility and expressivity of deep learning. Unlike traditional causal methods that rely on predefined causal structures, we tackle the more complex problem of learning causal structure directly from data—even when the causal variables themselves are not explicitly observed. This introduces significant challenges, including ill-posedness, nonconvexity, and the exponential complexity of combinatorial search. We will outline statistical aspects of these problems and present progress towards resolving these challenges with differentiable approaches to causal discovery and representation learning.
TIME Tuesday, January 27, 2026 at 11:00 AM - 12:00 PM
LOCATION A230, Technological Institute map it
CONTACT Kendall Minta kendall.minta@gmail.com EMAIL
CALENDAR Department of Industrial Engineering and Management Sciences (IEMS)
-
Feb3
TIME Tuesday, February 3, 2026 at 11:00 AM - 12:00 PM
LOCATION A230, Technological Institute map it
CONTACT Kendall Minta kendall.minta@gmail.com EMAIL
CALENDAR Department of Industrial Engineering and Management Sciences (IEMS)
-
Feb10
TIME Tuesday, February 10, 2026 at 11:00 AM - 12:00 PM
LOCATION A230, Technological Institute map it
CONTACT Kendall Minta kendall.minta@gmail.com EMAIL
CALENDAR Department of Industrial Engineering and Management Sciences (IEMS)
-
Feb17
EVENT DETAILS
lessNemhauser Best Paper Award: Chutong Gao, Redundancy or Flexibility? Interplay between their Values in Designing Fork-Join Systems
Coauthors: Seyed Iravani, Ohad Perry
Citation: We consider fork-join systems with redundancy and flexibility. Fork-join systems empower numerous modern online service platforms, such as AI agent workflows, among other real-world applications. In a 𝑘-requirement fork-join (FJ𝑘, 𝑘 ≥ 2) system, each job “forks” into multiple tasks processed in parallel, among which 𝑘 tasks are required to “join” for the completion of the job. To minimize the mean job sojourn times in FJ𝑘 systems, two strategies—redundancy and flexibility—are often employed. Redundancy deploys 𝑛 (> 𝑘) stations and sends one task from a job to each, but only 𝑘 out of 𝑛 tasks are needed to complete the job (rendering the remaining 𝑛 − 𝑘 tasks redundant). Flexibility, on the other hand, enables dynamic service capacity allocation. To understand the impact on the mean sojourn time of each strategy and their interplay, we propose a unified modeling framework that simultaneously considers: (i) how many redundant stations should be deployed, and (ii) how the service capacities should be allocated among all stations. We primarily focus on exponentially distributed task sizes. When flexibility is allowed, we prove a class of longest-queue policies are global-optimal, regardless of having redundancy or lack thereof—so redundancy has no marginal value in the existence of flexibility. In the absence of flexibility, we prove in FJ2 systems that “a little redundancy goes a long way”, in a twofold sense: To achieve heavy traffic optimality relative to the global-optimal system, it suffices to (i) employ only one extra station, and (ii) allocate an arbitrarily small capacity 𝜖 > 0 to that extra station. This suggests that (even a little) redundancy is as attractive as (full) flexibility in reducing sojourn times. Numerical studies further support the above insights in FJ𝑘 systems with 𝑘 > 2. We further consider a family of Weibull task-size distributions, where exponential becomes a special case. We show that when the variability of task sizes is low, redundancy can reduce the stability region (and thus harm both throughput and the sojourn time), so it should not be implemented. For highly variable task sizes, redundancy alone can expand the stability region, and flexibility on top of that further expands the stability region. But the flexible control policies need to be selected with extra caution—otherwise, implementing flexibility can harm the system’s stability.
__________________________________________________________________________________________________
Nelson Best Paper Award: Tong Xu, Integer programming for learning directed acyclic graphs from nonidentifiable Gaussian models
Coauthors: Armeen Taeb, Simge Küçükyavuz, Ali Shojaie
Citation: We study the problem of learning directed acyclic graphs from continuous observational data, generated according to a linear Gaussian structural equation model. State-of-the-art structure learning methods for this setting have at least one of the following shortcomings: (i) they cannot provide optimality guarantees and can suffer from learning suboptimal models; (ii) they rely on the stringent assumption that the noise is homoscedastic, and hence the underlying model is fully identifiable. We overcome these shortcomings and develop a computationally efficient mixed-integer programming framework for learning medium-sized problems that accounts for arbitrary heteroscedastic noise. We present an early stopping criterion under which we can terminate the branch-and-bound procedure to achieve an asymptotically optimal solution and establish the consistency of this approximate solution. In addition, we show via numerical experiments that our method outperforms state-of-the-art algorithms and is robust to noise heteroscedasticity, whereas the performance of some competing methods deteriorates under strong violations of the identifiability assumption. The software implementation of our method is available as the Python package micodag.
TIME Tuesday, February 17, 2026 at 11:00 AM - 12:00 PM
LOCATION A230, Technological Institute map it
CONTACT Kendall Minta kendall.minta@gmail.com EMAIL
CALENDAR Department of Industrial Engineering and Management Sciences (IEMS)
-
Mar3
TIME Tuesday, March 3, 2026 at 11:00 AM - 12:00 PM
LOCATION A230, Technological Institute map it
CONTACT Kendall Minta kendall.minta@gmail.com EMAIL
CALENDAR Department of Industrial Engineering and Management Sciences (IEMS)
-
Mar10
TIME Tuesday, March 10, 2026 at 11:00 AM - 12:00 PM
LOCATION A230, Technological Institute map it
CONTACT Kendall Minta kendall.minta@gmail.com EMAIL
CALENDAR Department of Industrial Engineering and Management Sciences (IEMS)