ANCRe: Adaptive Neural Connection Reassignment for Efficient Depth Scaling

TL;DR

This paper introduces ANCRe, a lightweight adaptive framework that reassigns residual connections in neural networks, leading to faster convergence and better depth utilization across various large models.

Contribution

It provides a theoretical analysis of residual connections and proposes ANCRe, a novel method for data-driven residual reallocation with minimal overhead.

Findings

Accelerated convergence in large language models and diffusion models

Improved performance and depth efficiency in deep ResNets

Theoretical proof of residual connection layout impacting convergence rates

Abstract

Scaling network depth has been a central driver behind the success of modern foundation models, yet recent investigations suggest that deep layers are often underutilized. This paper revisits the default mechanism for deepening neural networks, namely residual connections, from an optimization perspective. Rigorous analysis proves that the layout of residual connections can fundamentally shape convergence behavior, and even induces an exponential gap in convergence rates. Prompted by this insight, we introduce adaptive neural connection reassignment (ANCRe), a principled and lightweight framework that parameterizes and learns residual connectivities from the data. ANCRe adaptively reassigns residual connections with negligible computational and memory overhead ($<1\%$), while enabling more effective utilization of network depth. Extensive numerical tests across pre-training of large language models, diffusion models, and deep ResNets demonstrate consistently accelerated convergence, boosted performance, and enhanced depth efficiency over conventional residual connections.