An Adaptive Strategy for Sensory Processing

Raj, Rishabh (2021). An Adaptive Strategy for Sensory Processing. PhD thesis The Open University.



Recognizing objects and detecting associations among them is essential for the survival of organisms. The ability to perform these tasks is derived from the representations of objects obtained through processing information along sensory pathways. Our current understanding of sensory processing is based on two sets of foundational theories – The Efficient Coding Hypothesis and hierarchical assembly of object representations. These theories suggest that sensory processing aims to identify independent features of the environment and progressively represent objects in terms of comprehensive combinations of these features. Separately, the two sets of theories have successfully explained the detection of associations and perceptual invariance, respectively; however, reconciling them together in one unified theory has remained challenging. Independent features are deemed essential for detecting association by the Efficient coding hypothesis, but to achieve consistency in representations, multiple comprehensive structures corresponding to the same object must be hierarchically assembled, ignoring independence among such structures.

Here we propose an alternative framework for sensory processing in which the system, instead of finding the truly independent components of the environment, aims to represent objects based on their most informative structures. Using theoretical arguments, we show that following such a strategy allows the system to efficiently represent sensory cues without necessarily acquiring knowledge about statistical properties of all possible inputs. Through mathematical simulations, we find that the framework can describe the known characteristics of early sensory processing stages and permits consistent input representations observed at later stages of processing. We also demonstrate that the framework can be implemented in a biologically plausible neuronal circuit and explain aspects of experience and learning from corrupted inputs. Thus, this framework provides a novel perspective and a unified description of sensory processing in its entirety.

Viewing alternatives

Download history


Public Attention

Altmetrics from Altmetric

Number of Citations

Citations from Dimensions

Item Actions