Learning by Turning: Neural Architecture Aware Optimisation

TL;DR

This paper introduces Nero, a neural optimizer that adapts to architecture, requiring minimal tuning, and demonstrates reliable training where traditional methods often fail, with a focus on geometric insights linking architecture and optimization.

Contribution

The paper proposes Nero, a novel optimizer that integrates neural architecture awareness, enabling more reliable training without extensive tuning and reducing memory footprint.

Findings

Nero trains reliably without momentum or weight decay.

Nero outperforms Adam and SGD in certain scenarios.

Nero has a smaller memory footprint than Adam or LAMB.

Abstract

Descent methods for deep networks are notoriously capricious: they require careful tuning of step size, momentum and weight decay, and which method will work best on a new benchmark is a priori unclear. To address this problem, this paper conducts a combined study of neural architecture and optimisation, leading to a new optimiser called Nero: the neuronal rotator. Nero trains reliably without momentum or weight decay, works in situations where Adam and SGD fail, and requires little to no learning rate tuning. Also, Nero's memory footprint is ~ square root that of Adam or LAMB. Nero combines two ideas: (1) projected gradient descent over the space of balanced networks; (2) neuron-specific updates, where the step size sets the angle through which each neuron's hyperplane turns. The paper concludes by discussing how this geometric connection between architecture and optimisation may impact theories of generalisation in deep learning.