Speed is Confidence

TL;DR

This paper demonstrates that confidence-based early stopping in neural ensembles can significantly reduce reasoning steps and computational cost while maintaining high accuracy, inspired by biological neural speed-accuracy trade-offs.

Contribution

It introduces a training method that internalizes confidence-based halting, achieving ensemble-like accuracy with minimal inference cost and resource constraints.

Findings

Winner-take-all halting improves accuracy and efficiency.

Training with a winner-take-all loss matches ensemble performance.

Resource-efficient training on consumer hardware is feasible.

Abstract

Biological neural systems must be fast but are energy-constrained. Evolution's solution: act on the first signal. Winner-take-all circuits and time-to-first-spike coding implicitly treat when a neuron fires as an expression of confidence. We apply this principle to ensembles of Tiny Recursive Models (TRM) [Jolicoeur-Martineau et al., 2025]. On Sudoku-Extreme, halt-first selection achieves 97% accuracy vs. 91% for probability averaging -- while requiring 10x fewer reasoning steps. A single baseline model achieves 85.5% +/- 1.3%. Can we internalize this as a training-only cost? Yes: by maintaining K=4 parallel latent states but backpropping only through the lowest-loss "winner," we achieve 96.9% +/- 0.6% accuracy -- matching ensemble performance at 1x inference cost, with less than half the variance of the baseline. A key diagnostic: 89% of baseline failures are selection problems, revealing a 99% accuracy ceiling. As in nature, this work was also resource constrained: all experiments used a single RTX 5090. A modified SwiGLU [Shazeer, 2020] made Muon [Jordan et al., 2024] and high LR viable, enabling baseline training in 48 minutes and full WTA (K=4) in 6 hours on consumer hardware.