Less Is More: Recursive Reasoning with Tiny Networks
4 hours ago
- #Machine Learning
- #Recursive Reasoning
- #Neural Networks
- Introduces Tiny Recursive Model (TRM), a simpler recursive reasoning approach.
- TRM uses a single tiny network with only 2 layers and 7M parameters.
- Outperforms Hierarchical Reasoning Model (HRM) and many Large Language Models (LLMs) on tasks like ARC-AGI.
- Achieves 45% test-accuracy on ARC-AGI-1 and 8% on ARC-AGI-2.
- Demonstrates potential for solving complex problems with minimal computational resources.