Monty: A Thousand-Brains System for Sensorimotor AI скачать в хорошем качестве

Monty: A Thousand-Brains System for Sensorimotor AI

https://github*com/thousandbrainsproject/tbp*monty The provided source is an excerpt from a technical preprint evaluating Thousand-Brains Systems, a novel artificial intelligence (AI) architecture inspired by neuroscience, specifically the concept of cortical columns as semi-independent sensorimotor modules. The paper introduces Monty, the first implementation of this system, and details its performance on the task of 3D object recognition and pose estimation using the YCB dataset. The core findings demonstrate that Monty achieves robust inference and generalization through the use of structured reference frames and movement-based data integration, contrasting with the passive, data-intensive learning of current deep learning models. Furthermore, the evaluation highlights Monty's ability to perform rapid and continual learning with remarkable computational efficiency, supported by model-free and model-based policies, as well as a novel voting algorithm for multi-module consensus. Overall, the work posits that this bio-inspired, sensorimotor approach offers significant advantages in developing intelligent systems, particularly in domains requiring few-shot learning and generalization.