jiaxiaogang/HE

What it solves

he4o aims to create a General Artificial Intelligence (AGI) system that balances pre-designed code rules with dynamically acquired knowledge. It seeks to move beyond the dichotomy of purely compute-driven competitive learning (which can be too "unstable") and purely rule-based reasoning (which can be too "rigid").

How it works

Based on the "Helix Entropy Reduction Theory," the system operates as a "spiral entropy reduction machine." It uses a Dynamic Oriented Programming (DOP) approach where the system is born with a basic controller and storage structure, but its knowledge evolves post-natally from concrete experiences to abstract concepts.

Key technical components include:

• Cognitive Loop: A spiral-shaped thinking controller managing perception, recognition, learning, tasking, planning, solving, transfer, and behavior.
• Knowledge Representation: Uses sparse coding, features, concepts, and temporal sequences to represent information, focusing on the relationship between the general and the individual.
• Memory Structure: Employs a network for long-term memory, a tree for short-term memory, and a sequence for instantaneous memory.
• Learning Mechanisms: Primarily utilizes transfer learning (structural) supported by reinforcement learning (competitive).

Who it’s for

Developers and researchers interested in embodied intelligence, AGI architectures, and non-traditional neural network approaches that combine symbolic-like rules with dynamic learning.

Highlights

• Embodied Intelligence Demos: Successfully demonstrated a virtual agent (crow) that learns to find food, avoid obstacles, and use simple tools through interaction.
• Visual Recognition: Capable of learning to recognize objects (e.g., mice, cups, cats) from a small number of examples through an iterative process.
• Low Resource Requirement: Designed to run on a single machine terminal (currently tested on iOS devices), prioritizing disk I/O over heavy computation.
• Lifelong Learning: Supports dynamic, lifelong learning and rapid response to environmental changes.