CIMeC Colloquium Series
Announcing the 2023-2024 May Colloquium Speaker
ONLINE Thursday, May 2, 2024 at 4 p.m.
Knowledge Dependencies in Large Language Models
Mor Geva, Ph.D. is Assistant Professor (Senior Lecturer) at the School of Computer Science at Tel Aviv University and a Visiting Researcher at Google Research. Her research focuses on understanding the inner workings of large language models, to increase their transparency and efficiency, control their operation, and improve their reasoning abilities. Mor completed a Ph.D. in Computer Science and a B.Sc. in Bioinformatics at Tel Aviv University, and was a postdoctoral researcher at Google DeepMind and the Allen Institute for AI. She was nominated as one of the MIT Rising Stars in EECS and is a laureate of the Séphora Berrebi Scholarship in Computer Science. She was awarded the Dan David Prize for graduate students in the field of AI and received an Outstanding Paper Award at EACL 2023.
Some of the most pressing issues with large language models (LLMs), such as the generation of factually incorrect text and logically incorrect reasoning, may be attributed to the way models represent and recall knowledge internally. In this talk, we will evaluate the representation and utilization of knowledge dependencies in LLMs from two different perspectives. First, we will consider the task of knowledge editing, showing that (a) using various editing methods to edit a specific fact does not implicitly modify other facts that depend on it, and (b) some facts are often hard to disentangle. Next, we will consider the setting of latent multi-hop reasoning, showing that LLMs only weakly rely on knowledge dependencies when answering complex queries. While these shortcomings could potentially be mitigated by intervening on the LLM computation, they call for better training procedures and possibly new architectures.
The CIMeC Colloquia Series is an annual set of invited talks given by leading researchers in the mind/brain sciences, both from Italy and abroad, aimed principally at our PhD Students. Given the multi-disciplinary backgrounds of the CIMeC students and researchers, the Colloquia are aimed at a general scientific level rather than at a more specialized audience. Colloquia usually take place on the first Thursday of each month except August.
The 2023/24 Colloquium Committee: Elisa Pasquini, Matteo De Matola, Davide Mazzaccara.
Academic Councilor: Dr. Moritz Wurm
2023-2024 academic year Colloquium speakers
Karl J. Friston Scientific Director: Wellcome Trust Centre for Neuroimaging, Institute of Neurology, UCL The Physics of Sentience - November 9
Christoph Zrenner Scientist, Neurologist Center for Addiction and Mental Health Temerty Centre for Therapeutic Brain Intervention Personalised therapeutic brain stimulation with closed-loop EEG-synchronized TMS - December 7
Eric-Jan Wagenmakers Professor, University of Amsterdam Robustness Reports: An Initiative to Prevent Model Myopia by Consilience of Inductions - February 1
Michael S. Graziano Professor, Princeton University A Conceptual Framework for Consciousness - April 4
Dan S. Bassett Professor, J. Peter Skirkanich Professor, University of Pennsylvania Quantifying Economy in Brain Networks - April 11
2022-2023 academic year Colloquium speakers
Yaoda Xu Yale University - Understanding visual object representations in human occipito-temporal and posterior parietal cortices and convolutional neural networks
Lisa Giocomo Stanford University - Learning and adapting the structure of spatial maps
Earl K. Miller MIT - Cognition is an emergent property
Stefanie Hoehl University of Vienna - How do infants learn? Neural oscillations shed light on infant attention & learning
Bratislav Misic McGill University - Tools for multi-scale, multi-modal annotation of brain networks
Nicolas Schuck University of Hamburg - Learning and replay of state representations in the human brain
Marlene Behrmann University of Pittsburgh Medical School - Hemispheric organization in humans: two hemispheres, one mind
Nikolaus Kriegeskorte Columbia University - Neural network models as mechanistic explanations of brain information processing
Giacomo Rizzolatti Parma University - The mirror brain: past, present, and future
Nancy Kanwisher MIT - Functionally Specific Cortical Regions in Humans: What Others and Why These?