U(1) Symmetry-breaking Observed in Generic CNN Bottleneck Layers

We report on a novel model linking deep convolutional neural networks (CNN) to biological vision and fundamental particle physics. Information propagation in a CNN is modeled via an analogy to an optical system, where information is concentrated near a bottleneck where the 2D spatial resolution collapses about a focal point . A 3D space is defined by coordinates in the image plane and CNN layer , where a principal ray runs in the direction of information propagation through both the optical axis and the image center pixel located at , about which the sharpest possible spatial focus is limited to a circle of confusion in the image plane. Our novel insight is to model the principal optical ray as geometrically equivalent to the medial vector in the positive orthant of a -channel activation space, e.g. along the greyscale (or luminance) vector in colour space. Information is thus concentrated into an energy potential , which, particularly for bottleneck layers of generic CNNs, is highly concentrated and symmetric about the spatial origin and exhibits the well-known "Sombrero" potential of the boson particle. This symmetry is broken in classification, where bottleneck layers of generic pre-trained CNN models exhibit a consistent class-specific bias towards an angle defined simultaneously in the image plane and in activation feature space. Initial observations validate our hypothesis from generic pre-trained CNN activation maps and a bare-bones memory-based classification scheme, with no training or tuning. Training from scratch using combined one-hot loss improves classification for all tasks tested including ImageNet.
View on arXiv