C-voting: Confidence-Based Test-Time Voting without Explicit Energy Functions

TL;DR

This paper introduces C-voting, a confidence-based test-time scaling method for recurrent neural models that improves reasoning task performance without needing explicit energy functions.

Contribution

It proposes C-voting, a novel test-time scaling strategy applicable to recurrent models, and demonstrates its effectiveness with a new attention-based model on reasoning benchmarks.

Findings

C-voting achieves 4.9% higher accuracy on Sudoku-hard than energy-based voting.

Combined with ItrSA++, C-voting outperforms HRM on Sudoku-extreme and Maze tasks.

C-voting improves reasoning performance without requiring explicit energy functions.

Abstract

Neural network models with latent recurrent processing, where identical layers are recursively applied to the latent state, have gained attention as promising models for performing reasoning tasks. A strength of such models is that they enable test-time scaling, where the models can enhance their performance in the test phase without additional training. Models such as the Hierarchical Reasoning Model (HRM) and Artificial Kuramoto Oscillatory Neurons (AKOrN) can facilitate deeper reasoning by increasing the number of recurrent steps, thereby enabling the completion of challenging tasks, including Sudoku, Maze solving, and AGI benchmarks. In this work, we introduce confidence-based voting (C-voting), a test-time scaling strategy designed for recurrent models with multiple latent candidate trajectories. Initializing the latent state with multiple candidates using random variables, C-voting selects the one maximizing the average of top-1 probabilities of the predictions, reflecting the model's confidence. Additionally, it yields 4.9% higher accuracy on Sudoku-hard than the energy-based voting strategy, which is specific to models with explicit energy functions. An essential advantage of C-voting is its applicability: it can be applied to recurrent models without requiring an explicit energy function. Finally, we introduce a simple attention-based recurrent model with randomized initial values named ItrSA++, and demonstrate that when combined with C-voting, it outperforms HRM on Sudoku-extreme (95.2% vs. 55.0%) and Maze (78.6% vs. 74.5%) tasks.