Rapid Adaptive Task Learning Inspired by Hierarchical (Cortical) Networks
Abstract
This research aims to explore the mechanisms underlying rapid task adaptation in biological brains, specifically focusing on the anterior cingulate cortex (ACC) in mice, and apply these insights to the development of advanced bio-inspired AI systems for robotic agents. Rapid task adaptation is a hallmark of mammalian intelligence, enabling efficient navigation of complex environments. However, the precise mechanisms through which the brain integrates new information while retaining prior knowledge remain elusive. This challenge is also prevalent in artificial intelligence, where current systems struggle with efficiently reusing existing knowledge to adapt to new tasks. The project s primary objectives are twofold. First, it seeks to understand how the ACC encodes and updates task schemata during the learning process. To achieve this, the study will analyze existing neuronal activity data from the ACC of mice as they undergo curricular learning involving tasks of increasing complexity. By utilizing techniques such as calcium imaging to monitor neuronal activity, and applying various statistical and machine learning methods to analyze the data at both single-cell and population levels, the research aims to uncover the neural representations and modular encoding strategies employed by the ACC. The second objective is to develop artificial neural network (ANN) models inspired by these biological mechanisms, enabling robotic agents to generalize and adapt to new tasks more effectively. The project will involve creating an ANN model that mimics the ACC s modular task encoding, initially testing it in a virtual environment to assess its generalization and learning capabilities through a curriculum-based approach. Following successful virtual tests, the ANN model will be integrated into a physical robotic agent to evaluate its real-world performance in adapting to new tasks rapidly.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 06, 2025
- Source ID
- FA95502410325
Entities
People
- Benjamin Grewe
Organizations
- Air Force Office of Scientific Research
- United States Air Force