About
The Neuromorphic Cognitive Localization and Mapping project (CLAM), funded by Volkswagenstiftung, envisions a new class of neuromorphic systems inspired by the hippocampal-entorhinal grid-cell system (a vital component of the brain`s navigation system), to offer better generalization than current AI approaches, thanks to the creation of cognitive maps and a neuro-symbolic approach. This approach leverages a sparse, brain-inspired “hyperdimensional” encoding that reflects the geometry of the world and respects the topology of the hardware.
I'm always happy to welcome students looking for an internship or a final project (minimum 3 months). Here is an overview of current projects, but feel free to contact me with individual project ideas.
Research Topics
We combine neuromorphic computing, computational neuroscience and neuroAI. The Cognitive Localization and Mapping project focuses on navigation and memory formation inspired by the entorhinal-hippocampal formation.
CLAM project (Cognitive Localization and Mapping)
How do brains build cognitive maps, internal models that encode both spatial locations and abstract relationships to guide behavior? This is a fundamental question at the intersection of neuroscience and artificial intelligence. Unlike current AI systems, which require massive amounts of training data, biological brains excel at abstracting structure and disentangling relevant factors from limited experience. This ability to abstract and build maps is crucial for intelligent behavior, as it allows straightforward integration of new knowledge (continual learning) and use of knowledge independent of a specific task (transfer learning). Leveraging the geometry of the map representation allows generalization and reasoning by analogies.
A key brain region supporting map building is the entorhinal-hippocampal complex—the brain’s navigation system. Even though it is one of the best understood brain regions, its computational principles have not yet translated into scalable models for AI, presumably due to the lack of the right layer of abstraction.
The CLAM project (Cognitive Localization and Mapping) aims to address this gap by combining insights from the brain’s navigation system with hyperdimensional computing (HDC)—a theoretical framework that enables robust and efficient cognitive operations in hardware. It facilitates hardware-algorithm co-design and leverages the defining features of neuromorphic hardware: massive parallelism, in-memory compute, and event-based processing, while being robust to noise and fabrication variability.
By grounding AI architectures in biologically inspired principles (NeuroAI), the CLAM project offers a twofold contribution: a practical testbed for theories of cognitive map formation in neuroscience, and a path toward sustainable AI technologies that don't rely solely on ever-increasing computational resources.
Project Team
External Project Partners
Events
No results found.
Loading
Publications
Last Modified: 21.05.2025