The study suggests a paradigm shift for vision science and artificial intelligence

Reviewed by Emily Henderson, B.Sc.November 24, 2020

The receptive field (RF) of a neuron is the term applied to the space in which the presence of a stimulus alters the response of the neuron itself.

The responses of visual neurons, as well as the phenomena of visual perception in general, are highly nonlinear functions of visual input (in mathematics, nonlinear systems represent phenomena whose behavior cannot be expressed as the sum of the behaviors of its descriptors ).

On the contrary, the models of vision used in science are based on the notion of a linear receptive field; in artificial intelligence and machine learning, since artificial neural networks are based on classical models of vision, they also use linear receptive fields.

Vision modeling based on a linear receptive field poses several inherent problems: it changes with each input, presupposes a set of basic functions for the visual system, and conflicts with recent studies on dendritic calculations “,

Marcelo Bertalmío, First author of the study, Pompeu Fabra University – Barcelona

The study was recently published in the journal of the Nature group, Scientific reports. The article proposes to model the receptive field in a non-linear way, introducing the concept of intrinsically non-linear receptive field or INRF

The paper proposes to model the receptive field in a non-linear way, introducing the inherently non-linear receptive field or INRF. A study conducted by Marcelo Bertalmío, Alex Gómez-Villa, Adrián Martín, Javier Vázquez-Corral and David Kane, researchers from the UPF’s Department of Information and Communication Technologies (DTIC), together with Jesús Malo, researcher at the University of Valencia.

An approach with broad implications

The INRF, in addition to being more physiologically plausible and embodying the principle of efficient representation, has a key property with far-reaching implications: for several phenomena of the science of vision in which a linear RF must vary with the input to predict the responses While linear RF varies for each stimulus, the INRF can remain constant under different stimuli.

Bertalmío adds: “We have also shown that artificial neural networks with INRF modules instead of linear filters have significantly improved performance and better emulate basic human perception.” This research highlights the inherently non-linear nature of receptive fields in vision and suggests a paradigm shift for both vision science and artificial intelligence.