PROJECT
active
SWAMBIT — Shared Workspace And Memory for Brain-Inspired Transformers
MISSION
The core goal of this project
DATEXIS investigates brain-inspired modular AI architectures combining specialist transformer modules, shared workspace and memory, sparsity, and global feedback mechanisms.
Overview
SWAMBIT develops a brain-inspired AI architecture based on many small specialized modules that share a common global workspace and memory bank (analogous to thalamus and hippocampus), aiming to reduce compute cost compared to large monolithic language models while enabling flexible task adaptation and specialized knowledge acquisition. The project is a cooperation between BHT, Humboldt-Universität zu Berlin (Prof. Matthew Larkum), and the University of Bern (Prof. Walter Senn).
FUNDING BODY
BMFTR · Bundesministerium für Forschung, Technologie und Raumfahrt
PERIOD
2025-2028
PARTNERS
TEAM · CONTRIBUTORS
