BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:America/New_York
X-WR-TIMEZONE:America/New_York
BEGIN:VEVENT
UID:935@fds.yale.edu
DTSTART;TZID=America/New_York:20260216T130000
DTEND;TZID=America/New_York:20260216T140000
DTSTAMP:20260205T150816Z
URL:https://fds.yale.edu/events/fds-seminar-lingjing-kong/
SUMMARY:FDS Seminar: Lingjing Kong (CMU)\, "Causal AI for Transferable\, In
 terpretable\, and Controllable Machine Learning"
DESCRIPTION:\nZoom Password: 123\n\n\n\nAbstract: Foundation models are rap
 idly becoming capable assistants for knowledge work\, but their deployment
  in real settings is limited by three gaps: they do not transfer reliably 
 across environments\, their internal reasoning is opaque\, and their behav
 ior is hard to precisely control. In this talk\, I argue that these limita
 tions are not only about model size — they are fundamentally about wheth
 er learning captures and leverages the underlying structure of the data-ge
 nerating process. I use causal thinking as a practical lens to model what 
 is invariant\, what changes\, and what can be intervened on\, and I furthe
 r show how this leads to learning principles that improve trustworthiness.
 \n\n\n\nI will first present methods for learning unifying mechanisms from
  heterogeneous data\, across domains and modalities\, to enable reliable t
 ransfer and controllable generation. Next\, I will show how structured con
 cepts can be recovered even from seemingly unstructured data\, by analyzin
 g and improving self-supervised objectives (such as masking and diffusion)
  through hierarchical latent-variable models. These concept structures can
  then be used to interpret generative models and support targeted\, multi-
 level edits. Finally\, I connect these two threads to generalization beyon
 d the training distribution. I will discuss natural conditions for extrapo
 lation and a compositional generation framework that improves prompt follo
 wing for novel concept combinations. I will conclude with a brief outlook 
 on self-improving world models and AI-assisted scientific discovery.\n\n\n
 \nSpeaker Bio: Lingjing Kong is a Ph.D. candidate in the Computer Science 
 Department at Carnegie Mellon University. His research focuses on Causal A
 I for transferable\, interpretable\, and controllable systems\, with an em
 phasis on understanding and exploiting the structure of real-world data to
  make foundation models actionable and more reliable. He develops identifi
 cation principles and scalable algorithms for learning unified models from
  heterogeneous data\, uncovering hierarchical concept structures in unstru
 ctured data (e.g.\, images and text)\, and generalizing beyond training su
 pport through compositionality and extrapolation. His work has appeared in
  top ML venues\, including ICML\, NeurIPS\, CVPR\, ICLR\, and EMNLP\, and 
 has been prototyped and applied in industry through research internships.\
 n\n\n\n\n
CATEGORIES:FDS Events,Postdoctoral Applicants
END:VEVENT
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:STANDARD
DTSTART:20251102T010000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR