Research Theme

Neuro-Symbolic & Reasoning Systems

We fuse neural learning with symbolic structure to build models that reason, explain, and collaborate with experts.

Back to all themes View publications
AIML Lab logo

Neuro-symbolic AI, explained

Neuro-symbolic systems blend neural networks with symbolic knowledge bases and logical rules. They can learn from raw data, but they also know how to reason, prove, and explain—just like domain experts consulting manuals, laws, or ontologies.

Our team investigates questions like:

  • Can a neural network obey logical constraints while staying flexible enough to learn?
  • How do we repair or edit knowledge in a model without retraining it from scratch?
  • What kinds of explanations help practitioners trust and debug AI-assisted decisions?

Quick facts

  • Foundations: Logic programming, probabilistic reasoning, gradient-based learning.
  • Use cases: Safety-critical planning, knowledge-rich perception, interactive tutoring, scientific discovery.
  • Tooling: Differentiable provers, symbolic transpilers, concept-bottleneck architectures, interactive editors.

Differentiable logic

We translate logic programs into neural computations so systems can prove properties, verify actions, and adapt from examples simultaneously.

Knowledge-augmented agents

Agents query knowledge graphs, ontologies, and expert rules on the fly, ensuring decisions respect declarative constraints and domain best practices.

Explainable neuro reasoning

Interactive explanation channels reveal which facts, concepts, or proofs influenced a conclusion—making collaboration with domain experts tangible.

Representative Papers

NeST: The Neuro-Symbolic Transpiler

IJAR 2025 — Pfanschilling, Shindo, Dhami & Kersting

Introduces the Sum Product Loop Language and a transpiler that compiles neuro-symbolic programs into tractable inference routines.

Browse the full publication library