2008 Poster Sessions : Cognitive Memory

Student Name : Brian Percival
Advisor : Benard Widrow
Research Areas: Information Systems
We are developing a model of human memory. The model stores 'memories' of objects in the world obtained through sensory systems (e.g. images, audio clips). It stores these memories in chronologically grouped locations in memory, called memory folders, such that the location of specific memories is lost during storage, but memories that are located in the same memory folder were obtained in the same time period. Other data, that can be interpreted as 'thoughts' , can be stored in these memory folders at the same time. These may provide identities of the observed objects, recalled memories that contain related objects, or other abstract information. In order to retrieve a group of memories from a memory folder, the memory must be presented with sensory data that match at least one of the contents of the memory folder. However, the model does not perform an exhaustive search of all contents of all memory folders for each sensory input. Instead, an autoassociative memory that has been trained on the contents of the memory folders acts as a recognition unit that is used to identify whether the current input represents an input similar to any of the stored memories. This event is similar to the phenomenon of deja vu: the memory has determined that the current input is familiar, but has not retrieved the corresponding memory yet. The memory then performs a search of the contents of memory folders and extracts the contents of the memory folder that contains the best match to the input that triggered deja vu. The entire contents of the memory folder are now available, including the original pattern as well as identifying information and related patterns. The retrieved patterns can then in turn be used to recall further memories, creating a stream-of-consciousness-like phenomenon. We are currently pursuing face detection and recognition and will soon be working on speech and voice recognition.

Brian Percival is a PhD student in Electrical Engineering at Stanford working with Professor Bernard Widrow on this project. Brian is also working on a neural computational modeling project with fellow student Sridhar Devarajan in the Neurosciences Program under Bioengineering Professor Kwabena Boahen. Brian earned a degree in Engineering Sciences from Harvard University, and worked as a software engineer for several years before coming to Stanford.