Workshop on the Integration of Large Language Models and Reasoning, April 19th, 13:00-17:00 CEST
Hosted by Work Packages 4 and 5, this online workshop explores the fusion of large language models and reasoning. We are thrilled to welcome two esteemed speakers: Guy Van den Broeck (UCLA) and Scott Sanner (University of Toronto).
The workshop will be held online on Microsoft Teams. An email with instructions on how to join will be sent out before the event.
Call for Presentations
We invite participants to showcase their work through talks or poster sessions. Kindly indicate your preference during registration.
Registration Deadline
Please register by April 1st to secure your spot.
Attendance
Free, but registration is mandatory. Reserve your spot here: https://forms.gle/TcEbmKJVxtBFuCTW8
Agenda
| Hourly Schedule | Program |
|---|---|
| 13.00-13.30 | Opening remarks |
| 13.10-13.25 | Contributed talk: Paul Debjit Title: Refining and Improving the Reasoning Capabilities of LLMs |
| 13.25-13.40 | Contributed talk: Angelos Poulis, Eleni Tsalapati, Manolis Koubarakis Title: Reasoning over Description Logic-based Contexts with Transformers |
| 13.40-13.55 | Contributed talk: Leonardo Lucio Custode, Chiara Camilla Rambaldi Migliore Title: Comparing Evolutionary Methods and LLMs for Program Synthesis |
| 13.55-14.00 | Break |
| 14.00-14.50 | Invited talk: Scott Sanner (University of Toronto) Title: Symbolic AI 3.0 (S3): Rise of the LLMs Abstract: Large Language Models (LLMs) such as ChatGPT, GPT-4, and Gemini have emerged as a revolutionary technology for natural language and visual reasoning and numerous related AI applications. I’ll discuss some of my group’s own work on abstract reasoning and interactive conversational systems leveraging LLMs and the game-changing realizations that I have taken away from these investigations. This talk will then discuss some general implications of the LLM era and my conjectures as to how it will shift (and has already shifted) research foci in the near future and enable levels of user-facing AI deployment that were unthinkable just two years ago. |
| 14.50-15.05 | Break |
| 15.05-15.20 | Contributed talk: Alberto Bugarín-Diz Title: Enriching interactive explanations with fuzzy temporal constraint networks |
| 15.20-15.35 | Rishi Harzra Title: SayCanPay: Heuristic Planning with LLMs using Learnable Domain Knowledge |
| 15.35-15.50 | Tias Guns / Dimos Tsouros / Kostis Michailidis Title: Chatbots and LLMs for Constraint Programming: Opportunities and Challenges |
| 15.50-16.00 | Break |
| 16.00-16.50 | Invited talk: Guy Van den Broeck (UCLA) Title: Symbolic Reasoning for Large Language Models Abstract: Many expect that AI will solve society’s problems by simply being more intelligent than we are. Implicit in this bullish perspective is the assumption that AI will naturally learn to reason from data: that it can form trains of thought that “make sense”, similar to how a human expert might reason about a case, or more formally, how a mathematician might prove a theorem. This talk will investigate the question whether this behavior can be learned from data, and how we can design the next generation of AI techniques that can achieve such capabilities. It will focus on neurosymbolic reasoning for large language models, both at training and generation time, using probabilistic circuits as the architecture that bridges learning and reasoning |
| 16.50-17.00 | Closing remarks |
Organizers
- Antonio Di Stasio (University of Oxford)
- Franesco Giannini (University of Siena)
- Giuseppe De Giacomo (University of Oxford)
- Luc De Raedt (KU Leuven)
- Mehdi Ali (Fraunhofer)
- Michela Milano (University of Bologna)
- Robin Manhaeve (KU Leuven)
- Vincent Derkinderen (KU Leuven)
Contact: Robin Manhaeve (robin.manhaeve@kuleuven.be)
Speakers
-
Guy Van den Broeck (UCLA)
-
Scott Sanner (University of Toronto)