Led by Researcher Zeng Yi of the Laboratory of Brain Atlas and Brain-Inspired Intelligence, the Brain-Inspired Cognitive AI Team has developed the Brain-Inspired Cognitive Intelligence Engine (BrainCog)—a fully spiking neural network–based framework. BrainCog is fully open-source and open-access, providing foundational support for research into brain-inspired intelligence for general artificial intelligence (AGI), and aiding exploration of the computational principles of natural intelligence and the development of next-generation AI. The related article was published as a cover story in Cell Press’s journal Patterns.

BrainCog is built on multi-scale biological plasticity principles and supports fully spiking neural network (SNN) modeling. It features brain-inspired AI models along with the ability to simulate brain function and structure, offering researchers in brain-inspired AI and computational neuroscience a relatively complete and systematic set of interface components. These include neuron models at different levels of biological detail, a rich set of brain-inspired learning and plasticity rules, various network connectivity patterns, multiple neural coding schemes, diverse functional brain area models, and software–hardware co-design systems.

Currently, BrainCog has achieved substantial progress in perception and learning, knowledge representation and reasoning, decision-making, embodied intelligence, social cognition, and developmental evolution. In hardware–software co-design, the team has also developed and released BrainCog Firefly, an FPGA-based brain-inspired SNN accelerator that improves SNN inference performance on edge devices. This technology shows significant potential for real-world tasks such as autonomous visual localization and navigation for intelligent vehicles, high-speed obstacle avoidance for drones using event cameras, and adaptive exploration and complex task coordination for robots.

Based on the BrainCog framework, the team has also developed BORN (Brain-inspired Open-ended Reasoning Network), an SNN-based AI engine designed to achieve general cognitive capability. BORN integrates cognitive functions including perception and learning, working memory, long-term memory, knowledge representation and reasoning, decision-making, motor control, attention and consciousness, empathy, and social cognition.

BrainCog has initially integrated the team's earlier work on self-modeling experiments (mirror tests, rubber-hand illusion simulations, theory-of-mind modeling, and affective empathy simulations). Centered on the BORN AI engine, the system has further enhanced self-organizing learning capabilities for brain-inspired intelligent agents. In terms of coordination among different learning models, BORN, built on BrainCog as its computational foundation, has initially realized the integration of perception and learning, sequence learning and generation, knowledge representation and reasoning, and motor control. It has also been demonstrated in humanoid robotics and emotion-driven music composition and performance, supporting tasks such as visual emotion recognition, emotion-dependent music creation, and musical performance by humanoid robots.

Looking forward, the team will continue to refine and expand its spiking neural network–based general AI engine, providing powerful foundational support for research in brain-inspired AI and neuroscience.

Article Link:
https://www.cell.com/patterns/fulltext/S2666-3899(23)00144-7

BrainCog Homepage:
http://www.brain-cog.network/

Brain-Inspired Cognitive Intelligence Engine “BrainCog” Computational Components and Applications

Cover of Patterns Journal