The dominant paradigm in language modeling—scaling next-t oken prediction with parametric knowledge storage—delivers impressive ca pabilities but also fundamental limitations: brittle factual memory\, inef ficient parameters\, and myopic reasoning. Progress requires a shift towar d external memory and architectures that reason globally before committing to tokens.
In this talk I present two recent directions that suppor t this claim. First\, Limited-Memory Language Models externalize factual k nowledge during pre-training\, yielding models that are more controllable\ , verifiable\, and parameter-efficient. Second\, latent diffusion–augmen ted language models demonstrate how planning in continuous latent space ov ercomes the foresight limitations of next-token prediction\, improving rea soning and coherence.
Bio: Kilian Q. Weinberger is a Profess or of Computer Science at Cornell University. He received his Ph.D. in Mac hine Learning from the University of Pennsylvania and his undergraduate de gree in Mathematics and Computing from the University of Oxford. Prior to Cornell\, he served as an Associate Professor at Washington University in St. Louis and as a research scientist at Yahoo! Research. His research foc uses on Machine Learning and its applications\, specifically learning unde r resource constraints\, metric learning\, AI in Science\, computer vision \, autonomous vehicles\, Gaussian Processes\, and deep learning. Dr. Weinb erger is an ACM and AAAI Fellow (2024) and a Blavatnik National Awards Fin alist (2021).
LOCATION:https://cam-ac-uk.zoom.us/j/86890624365?pwd=oYGWpY7d5r3JOaUCaJXTD 0sRECFxab.1 END:VEVENT END:VCALENDAR