Efficient Representation of Natural Image Patches

TL;DR

This paper introduces a biologically inspired model for efficient image patch representation, demonstrating that optimizing for information transmission alone is insufficient for accurate probability modeling, and showing potential efficiency gains over deep learning.

Contribution

The paper presents a new abstract model based on minimal assumptions that achieves efficient image patch representation and offers insights into biological visual systems and deep learning efficiency.

Findings

Efficient representation achieved through biologically plausible loss functions.

Optimizing for information transmission alone does not ensure probability modeling accuracy.

Model shows significant efficiency advantages over some deep learning approaches.

Abstract

Utilizing an abstract information processing model based on minimal yet realistic assumptions inspired by biological systems, we study how to achieve the early visual system's two ultimate objectives: efficient information transmission and accurate sensor probability distribution modeling. We prove that optimizing for information transmission does not guarantee optimal probability distribution modeling in general. We illustrate, using a two-pixel (2D) system and image patches, that an efficient representation can be realized through a nonlinear population code driven by two types of biologically plausible loss functions that depend solely on output. After unsupervised learning, our abstract information processing model bears remarkable resemblances to biological systems, despite not mimicking many features of real neurons, such as spiking activity. A preliminary comparison with a contemporary deep learning model suggests that our model offers a significant efficiency advantage. Our model provides novel insights into the computational theory of early visual systems as well as a potential new approach to enhance the efficiency of deep learning models.