Are you fascinated by a future where generative AI models fully comply with the specifications and the requirements of the systems that use them? We are excited to offer a postdoc position, which is focused on integrating generative AI (e.g., LLMs) with symbolic reasoning, where the latter can verify the output of the former
is compliant with explicit and formal system constraints. The position is within the context of the EU-funded
SmartEM project, which aims to use AI-assisted methods to create industrial surrogate models from heterogeneous data sources. In SmartEM, we you will have the possibility to directly collaborate with industry partners such as Canon, Philips, ThermoFisher, Siemens and others.
InformationLarge Language Models are known to hallucinate. Techniques such as In-Context Learning, Retrieval-Augmented Generation, Chain-of-Thought Prompting, and Self-Refinement can help reduce hallucinations, but they do not eliminate them entirely. This poses a challenge in applications where trustworthiness and reliability are critical.
In this project, you will explore neuro-symbolic methods that integrate LLMs (or Generative AI more broadly) with Symbolic AI techniques. In this hybrid approach, the Generative AI component proposes candidate solutions to a problem, while the Symbolic AI component validates these solutions against explicit, formalized constraints by symbolic reasoning. An example is a LLM translating natural language specification into a symbolic representation (e.g. knowledge graph (KG) or logic program) and a symbolic solver computing the solution. Another example is the generation of surrogate model architectures for complex industrial systems: the generative component proposes models based on natural language specifications, while the symbolic component ensures compliance with system constraints, encoding domain knowledge and business requirements, provided in symbolic form.
The research will be conducted in collaboration with industrial partners as part of the European project SmartEM, focused on surrogate modeling. The goal of SmartEM is to develop surrogate models, or systems of surrogates, for complex industrial systems or their high-fidelity simulations. The European project will supply concrete, real-world applications to validate and refine the general methods developed during the Postdoc research.
You work will take place within the
Interconnected Resource-aware Intelligent Systems (IRIS) cluster at the Mathematics and Computer Science department. The research topic will be tailored to your research interests. The project offers collaboration with industrial partners, providing a chance to test your models in real-world scenarios. You will also have the chance to closely collaborate with a SmartEM PhD candidate.