CVAIMar 17, 2021

The Untapped Potential of Off-the-Shelf Convolutional Neural Networks

arXiv:2103.09891v1
Originality Highly original
AI Analysis

This addresses performance bottlenecks in computer vision by enabling dynamic inference, offering a novel approach to improve accuracy without increasing model size or training complexity.

The paper tackles the problem of static network topology at inference-time limiting performance, showing that by allowing four layers to dynamically change configuration, off-the-shelf models like ResNet-50 achieve over 95% accuracy on ImageNet, exceeding models with 20x more parameters.

Over recent years, a myriad of novel convolutional network architectures have been developed to advance state-of-the-art performance on challenging recognition tasks. As computational resources improve, a great deal of effort has been placed in efficiently scaling up existing designs and generating new architectures with Neural Architecture Search (NAS) algorithms. While network topology has proven to be a critical factor for model performance, we show that significant gains are being left on the table by keeping topology static at inference-time. Due to challenges such as scale variation, we should not expect static models configured to perform well across a training dataset to be optimally configured to handle all test data. In this work, we seek to expose the exciting potential of inference-time-dynamic models. By allowing just four layers to dynamically change configuration at inference-time, we show that existing off-the-shelf models like ResNet-50 are capable of over 95% accuracy on ImageNet. This level of performance currently exceeds that of models with over 20x more parameters and significantly more complex training procedures.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes