Talk "Reasoning with Large Language Models"
Abstract
Large Language Models (LLMs) such as DeepSeek R-1, OpenAI o1 and many others have recently demonstrated very robust reasoning capabilities, arousing considerable interest in the scientific community. The success of these models usually depends on several factors, such as inference-time scaling and a fine-tuning phase based on Reinforcement Learning. In this talk, we will analyze the main ingredients of reasoning LLMs and we will introduce a new emerging paradigm based on reasoning in a continuous latent space.
Short bio
Enver Sangineto is Associate Professor at the University of Modena and Reggio Emilia, where he teaches Machine Learning, Image Processing, and AI for Automotive. Among his research interests, he has worked on Tabular Data Generation, Compositional AI, Vision-Language Models, Diffusion Models, Transformer Training Regularization and Self-Supervised Learning. He is the author of over 100 publications in international journals and conferences and associate editor of Machine Learning Journal and Machine Vision and Application. In 2014, he worked at the Center for Research in Computer Vision laboratory at the University of Central Florida (UCF), under the supervision of Mubarak Shah, where he worked on Object Detection.