"Bayesian Developments in Models of Perceptual Decision Making"
The existence of neurons that seem to represent accumulated evidence for a perceptual decision have spurred increased interest in the underlying mechanisms, but on a behavioral level psychologists have long before postulated corresponding models. Acknowledging a sequential nature of the decision process, a basic mechanism has been identified that explains decision making as a (biased) random walk towards a bound. A broad range of models implements this basic mechanism. The most prominent of them is the drift-diffusion model which formalizes the random walk with a simple stochastic process. This model and its variants, however, typically content themselves with implementing the basic accumulation mechanism while omitting to explain potential sources of randomness and biases in the accumulation. An alternative is to use probabilistic (Bayesian) models, because they explicitly link the decision process to sensory information by defining how evidence is extracted from the observed stimulus based on probabilistic, generative models of the considered stimuli. Such probabilistic models, therefore, provide a richer, more complete description of perceptual decision making than the standard drift-diffusion models. In addition, they allow to easily connect perceptual decision making to further topics surrounding the probabilistic brain. In my talk I will outline these ideas. In particular, I will show that classic accumulation models, in form of the drift-diffusion model, can be seen as a special case of probabilistic model and I will show how results from one model may be transferred to the other.