Could be an interesting paper. However, the authors appear to be fairly unknown according to Google Scholar (I checked the last two authors). They are affiliated with the Chinese Ubiquant (who?) a hedge fund management firm that apparently also established an AI Lab.
From the abstract:
"Universal transformers (UTs) have been widely used for complex reasoning tasks such as ARC-AGI and Sudoku, yet the specific sources of their performance gains remain underexplored.
In this work, we systematically analyze UTs variants and show that improvements on ARC-AGI primarily arise from the recurrent inductive bias and strong nonlinear components of Transformer, rather than from elaborate architectural designs. Motivated by this finding, we propose the Universal Reasoning Model (URM), which enhances the UT with short convolution and truncated backpropagation. Our approach substantially improves reasoning performance, achieving state-of-the-art 53.8% pass@1 on ARC-AGI 1 and 16.0% pass@1 on ARC-AGI 2. ..."
No comments:
Post a Comment