Neural Encoding and Decoding

Neural encoding models form the first stepping stone of most neural signal processing frameworks. These models relate the observed neural activity to intrinsic and extrinsic stimuli as well as to neural states and sources. From a signal processing viewpoint, neural encoding models have two main applications. First, when combined with models accounting for the dynamics of brain activity, they can be used to infer information regarding the function of the underlying neural systems given the observed activity and extrinsic stimuli. Second, they can be used as a basis for solving the so-called neural decoding problems. Neural decoding can be thought of as a dual to neural encoding, where the parameters of the underlying neuronal models are assumed to be fixed, and the observations are used to estimate extrinsic stimuli, or intrinsic processes such as perception, decision-making, intention, and attention. In this chapter, we present formal definitions and mathematical formulations of neural encoding and decoding problems from a Bayesian perspective and provide two case studies involving electrophysiology and magnetoencephalography (MEG) data.