Universal Reasoning Model (53.8% pass 1 ARC1 and 16.0% ARC 2)
4 months ago
- #Universal Transformers
- #Reasoning Models
- #Artificial Intelligence
- Universal transformers (UTs) are widely used for complex reasoning tasks like ARC-AGI and Sudoku.
- Performance improvements in UTs come from recurrent inductive bias and strong nonlinear components, not elaborate architectural designs.
- The Universal Reasoning Model (URM) enhances UT with short convolution and truncated backpropagation.
- URM achieves state-of-the-art results: 53.8% pass@1 on ARC-AGI 1 and 16.0% pass@1 on ARC-AGI 2.
- Code for the research is available online.