English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Recent work suggests that changing Convolutional Neural Network (CNN)architecture by introducing a bottleneck in the second layer can yield changesin learned function. To understand this relationship fully requires a way ofquantitatively comparing trained networks. The fields of electrophysiology andpsychophysics have developed a wealth of methods for characterising visualsystems which permit such comparisons. Inspired by these methods, we propose anapproach to obtaining spatial and colour tuning curves for convolutionalneurons, which can be used to classify cells in terms of their spatial andcolour opponency. We perform these classifications for a range of CNNs withdifferent depths and bottleneck widths. Our key finding is that networks with abottleneck show a strong functional organisation: almost all cells in thebottleneck layer become both spatially and colour opponent, cells in the layerfollowing the bottleneck become non-opponent. The colour tuning data canfurther be used to form a rich understanding of how colour is encoded by anetwork. As a concrete demonstration, we show that shallower networks without abottleneck learn a complex non-linear colour system, whereas deeper networkswith tight bottlenecks learn a simple channel opponent code in the bottlenecklayer. We further develop a method of obtaining a hue sensitivity curve for atrained CNN which enables high level insights that complement the low levelfindings from the colour tuning data. We go on to train a series of networksunder different conditions to ascertain the robustness of the discussedresults. Ultimately, our methods and findings coalesce with prior art,strengthening our ability to interpret trained CNNs and furthering ourunderstanding of the connection between architecture and learnedrepresentation. Code for all experiments is available athttps://github.com/ecs-vlc/opponency.

How Convolutional Neural Network Architecture Biases Learned Opponency and Colour Tuning