PSYCH 221 Final Project: A Probabilistic Model of the Visual System
Siddhartha Kasivajhula
A Probabilistic Model of the Visual System
Abstract
A probabilistic model of the visual system is proposed. The model exhibits the
following 3-phase high-level behavior: First, a prior distribution is generated over
predicted features in the scene. Then, the eyes supply visual information
(evidence). And finally, the distribution is modified based on the evidence
received, and this posterior distribution again serves as the prior for the
subsequent time-step. The process of observation is modeled by a Hidden
Markov Model; and a 2-level noisy-or Bayes Net classifier is used to model the
process of scene interpretation. Finally, the results of the Bruner and Potter
Experiment are reinterpreted using this probabilistic model, and an explanation
of the results is suggested.
The Paper
A Probabilistic Model of the Visual System (pdf)
A Probabilistic Model of the Visual System (PPT)
MATLAB Code
GenerateInitialScene.m -- Generates features in the initial scene randomly.
GenerateScene.m -- Generates a set of feature vectors representing the scene.
HMM_scene.m -- Hidden Markov Model for the scene. Given an input scene at time t, this will evolve the scene and return the scene at time t+1.
MakeObservation.m -- Simulates an observation by sampling a set of feature vectors from the scene, with noise added.
NoisyOrBN_interpret.m -- interprets the scene using a Noisy-Or Bayes Net and returns the object with the highest weight.
ObserveScene.m -- Tests the limited implementation of the model by generating a scene and then observing it.
RepresentObserverKnowledge.m -- Define observer's prior knowledge of objects.
TestModel.m -- Tests the entire MATLAB implementation: Steps you through consecutive timesteps in the model.
Other filesREADME.txt -- Read this for more information about the MATLAB implementation.
Kasivajhula.tar.gz -- Contains all of the files shown above.