Topics
Foundation Models, Large Language Models, Deep LearningBiography
Sven Giesselbach is the leader of the Natural Language Understanding (NLU) team at the Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS). His team develops solutions in the areas of medical, legal and general document understanding which in their core build upon (large) pre-trained language models. Sven Giesselbach is also part of the Lamarr Institute and the OpenGPT-X project in which he investigates various aspects of Foundation Models. Based on his project experience of more than 25 natural language understanding projects he studies the effect of Foundation Models on the execution of Natural Language Understanding projects and the novel challenges and requirements which arise with them. He has published several papers on Natural Language Processing and Understanding, which focus on the creation of application-ready NLU systems and the integration of expert knowledge in various stages of the solution design. Most recently he co-authored a book on “Foundation Models for Natural Language Processing – Pre-trained Language Models Integrating Media” which will be published by Springer Nature.
Gerhard Paaß, Sven Giesselbach, Foundation Models for Natural Language Processing – Pre-trained Language Models Integrating Media, , Springer, May, 2023
Talk
Topics
Large Language Models, Deep Learning, Natural Language ProcessingBiography
Vivek Natarajan is a Research Scientist at Google leading research at the intersection of large language models (LLMs) and biomedicine. In particular, Vivek is the lead researcher behind Med-PaLM and Med-PaLM 2, which were the first AI systems to obtain passing and expert level scores on US Medical License exam questions respectively. Med-PaLM was recently published in Nature and has been featured in The Scientific American, Wall Street Journal, The Economist, STAT News, CNBC, Forbes, New Scientist among others. More recently, Vivek also led the development of Med-PaLM M, the first demonstration of a generalist biomedical AI system.
Over the years, Vivek’s research has been published in well-regarded journals and conferences like Nature, Nature Medicine, Nature Biomedical Engineering, JMLR, CVPR, ICCV and NeurIPS. It also forms the basis for several regulated medical device products under clinical trials at Google, including the NHS AI award winning breast cancer detection system Mammo Reader and the skin condition classification system DermAssist.
Prior to Google, Vivek worked on multimodal assistant systems at Facebook AI Research and published award winning research, was granted multiple patents and deployed AI models to products at scale with hundreds of millions of users.
Talk
Topics
Global Optimization, Mathematical Modeling, Energy Systems, Financial applications, and Data SciencesBiography
Panos Pardalos was born in Drosato (Mezilo) Argitheas in 1954 and graduated from Athens University (Department of Mathematics). He received his PhD (Computer and Information Sciences) from the University of Minnesota. He is a Distinguished Emeritus Professor in the Department of Industrial and Systems Engineering at the University of Florida, and an affiliated faculty of Biomedical Engineering and Computer Science & Information & Engineering departments.
Panos Pardalos is a world-renowned leader in Global Optimization, Mathematical Modeling, Energy Systems, Financial applications, and Data Sciences. He is a Fellow of AAAS, AAIA, AIMBE, EUROPT, and INFORMS and was awarded the 2013 Constantin Caratheodory Prize of the International Society of Global Optimization. In addition, Panos Pardalos has been awarded the 2013 EURO Gold Medal prize bestowed by the Association for European Operational Research Societies. This medal is the preeminent European award given to Operations Research (OR) professionals for “scientific contributions that stand the test of time.”
Panos Pardalos has been awarded a prestigious Humboldt Research Award (2018-2019). The Humboldt Research Award is granted in recognition of a researcher’s entire achievements to date – fundamental discoveries, new theories, insights that have had significant impact on their discipline.
Panos Pardalos is also a Member of several Academies of Sciences, and he holds several honorary PhD degrees and affiliations. He is the Founding Editor of Optimization Letters, Energy Systems, and Co-Founder of the International Journal of Global Optimization, Computational Management Science, and Springer Nature Operations Research Forum. He has published over 600 journal papers, and edited/authored over 200 books. He is one of the most cited authors and has graduated 71 PhD students so far. Details can be found in www.ise.ufl.edu/pardalos
Panos Pardalos has lectured and given invited keynote addresses worldwide in countries including Austria, Australia, Azerbaijan, Belgium, Brazil, Canada, Chile, China, Czech Republic, Denmark, Egypt, England, France, Finland, Germany, Greece, Holland, Hong Kong, Hungary, Iceland, Ireland, Italy, Japan, Lithuania, Mexico, Mongolia, Montenegro, New Zealand, Norway, Peru, Portugal, Russia, South Korea, Singapore, Serbia, South Africa, Spain, Sweden, Switzerland, Taiwan, Turkey, Ukraine, United Arab Emirates, and the USA.
Talk
Abstract TBA
Biography
Raniero Romagnoli is currently CTO of Almawave, that he joined in 2011, with the responsibility of defining and implementing the company’s technology strategy, with a special focus on R&D labs, helping Almawave create and evolve its products and solutions, that are based on proprietary Natural Language Processing technology to leverage speech and text information and communications in order to govern processes and improve both self and assisted engagement with users. Before joining Almawave Raniero worked for 2 years in RSA and before that Raniero worked for Hewlett Packard, for almost 10 years, in different technology and divisions, covering roles both in Product Management and R&D in intelligent support systems area. Raniero has a broad experience in the artificial intelligence field, starting from his research activities in the late ’90s on Machine Learning and Neural Networks for image processing, than in the security space, and since he joined Almawave in the field of speech and text analysis.
Talk
Topics
Mathematics of Artificial Neural Networks, Biological Neural Networks, Deep LearningBiography
Johannes Schmidt-Hieber was born in Freiburg im Breisgau, Germany, in 1984. He received the master’s degree from the University of Göttingen, Germany, in 2007, and the joint Ph.D. degree from the University of Göttingen and the University of Bern, Switzerland, in 2010.,His Ph.D. degree was followed by two one-year post-doctoral visits at Vrije Universiteit Amsterdam, The Netherlands, and ENSAE, Paris, France. From 2014 to 2018, he was an Assistant Professor at the University of Leiden. Since 2018, he has been a Full Professor at the University of Twente, The Netherlands. His research interests include mathematical statistics, including nonparametric Bayes and statistical theory for deep neural networks. He serves as an Associate Editor for the Annals of Statistics, Bernoulli, and Information and Inference.
The Prof. Schmidt-Hieber’s ERC CoG grant has been selected by the ERC as one of four highlighted projects.
Talk
Recently a lot of progress has been made regarding the theoretical understanding of machine learning methods. One of the very promising directions is the statistical approach, which interprets machine learning as a collection of statistical methods and builds on existing techniques in mathematical statistics to derive theoretical error bounds and to understand phenomena such as overparametrization. The talk surveys this field and describes future challenges.
Topics
fine-tuning LLMs, LLMs, RL with human feedbackBiography
Michal is a principal llama engineer at Meta Paris, tenured researcher at Inria, and the lecturer of the master course Graphs in Machine Learning at l’ENS Paris-Saclay. Michal is primarily interested in designing algorithms that would require as little human supervision as possible. That is why he is working on methods and settings that are able to deal with minimal feedback, such as deep reinforcement learning, bandit algorithms, self-supervised learning, or self play. Michal has recently worked on representation learning, word models and deep (reinforcement) learning algorithms that have some theoretical underpinning. In the past he has also worked on sequential algorithms with structured decisions where exploiting the structure leads to provably faster learning. Michal is now working on large large models (LMMs), in particular providing algorithmic solutions for their scalable fine-tuning and alignment. He received his Ph.D. in 2011 from the University of Pittsburgh under the supervision of Miloš Hauskrecht and was a postdoc of Rémi Munos before getting a permanent position at Inria in 2012 and starting Google DeepMind Paris in 2018.
Talk
Reinforcement learning from human feedback (RLHF) is a go-to solution for aligning large language models (LLMs) with human preferences; it passes through learning a reward model that subsequently optimizes the LLM’s policy. However, an inherent limitation of current reward models is their inability to fully represent the richness of human preferences and their dependency on the sampling distribution. In the first part we turn to an alternative pipeline for the fine-tuning of LLMs using pairwise human feedback. Our approach entails the initial learning of a preference model, which is conditioned on two inputs given a prompt, followed by the pursuit of a policy that consistently generates responses preferred over those generated by any competing policy, thus defining the Nash equilibrium of this preference model. We term this approach Nash learning from human feedback (NLHF) and give a new algorithmic solution, Nash-MD, founded on the principles of mirror descent. NLHF is compelling for preference learning and policy optimization with the potential of advancing the field of aligning LLMs with human preferences. In the second part of the talk we delve into a deeper theoretical understanding of fine-tuning approaches as RLHF with PPO and offline fine-tuning with DPO (direct preference optimization) based on the Bradley-Terry model and come up with a new class of LLM alignment algorithms with better both practical and theoretical properties.
https://arxiv.org/abs/2312.00886
https://arxiv.org/abs/2310.12036
