Image
About
The success of generative models to scale has come with hallucinations. They force the community to think: "When are model generations accurate and truthful?" In this workshop, we will consider novel perspectives and formulations that can explain and mitigate hallucinations in generative models, making them more robust and leading to overall accurate synthesis. During the workshop, we will focus both on discrete settings, e.g., text or graph-structured data, as well as continuous settings across a broad family of models. The workshop is intended to be exploratory; we welcome novel theoretical understanding and trade-offs with generalization and creativity.
Chairs/Organizers
Invited Participants
Leman Akoglu (CMU), Anastasios N. Angelopoulos (Arena), Elias Bareinboim (Columbia University), Florian Buettner (Goethe University Frankfurt), Moses Charikar (Stanford University), Grigorios Chrysos (University of Wisconsin-Madison), Ilias Diakonikolas (University of Wisconsin-Madison), Adji Bousso Dieng (Princeton University), Yue Dong (UC Riverside), Soheil Feizi (University of Maryland, College Park), Hamed Hassani (University of Pennsylvania), Mohit Iyyer (University of Maryland, College Park), Adam Kalai (RESI), Murat Kocaoglu (Johns Hopkins University), Gitta Kutyniok (LMU Munich), Sharon Li (University of Wisconsin-Madison), Litian Liu (Qualcomm AI research), Eric Nalisnick (Johns Hopkins University), Changdae Oh (University of Wisconsin-Madison), Debmalya Panigrahi (Duke University), Pradeep Ravikumar (Carnegie Mellon University), Jascha Sohl-Dickstein (Anthropic), Mahdi Soltanolkotabi (University of Southern California), Dawn Song (UC Berkeley), Jeremias Sulam (Johns Hopkins University), Grigoris Velegkas (Google Research), Santosh Vempala (Georgia Institute of Technology), James Zou (Stanford University)