Visual attention: Towards a neural network model of visual short-term memory

Anders Petersen

AbstractIn this thesis a new type of model for Visual Short-Term Memory (VSTM) is presented. The model is inspired by previous work of Usher & Cohen [37], and it links closely with Bundesen’s well-established mathematical theory of visual attention, see [5].

We evaluate the model’s ability to fit experimental data from a classical type of psychological study, which is known as ‘whole and partial report’. Further we compare our results with results obtained using earlier types of models. These models have already successfully assessed the spatial distribution of visual attention; our neural network meets this standard and, on top of that, offers a neural interpretation of how objects are consolidated in VSTM.

In cognitive psychology there are two types of studies have so far been held very isolated from each other. These are studies of Whole Report (WR) and Partial Report (PR) and studies using Rapid Serial Visual Presentation (RSVP).

Bundesen has been very successful in modelling WR & PR data; however no one has so far provided a satisfactory account for phenomena related to RSVP.

It is an interesting finding that our model contains an inherent dynamical behaviour, which potentially could lead to an account for temporally dependent phenomena observed with RSVP.

We hope that in the future, the model will be able to yield a computational account of temporally dependent phenomena like the attentional blink effect, lag-1 sparing, and perhaps even the newly reported cross-over effects for very short inter-stimulus lags ([34]).
TypeMaster's thesis [Academic thesis]
PublisherInformatics and Mathematical Modelling, Technical University of Denmark, DTU
AddressRichard Petersens Plads, Building 321, DK-2800 Kgs. Lyngby
NoteSupervised Lars Kai Hansen, and co-supervisor Henrik Aanæs
Electronic version(s)[pdf]
BibTeX data [bibtex]
IMM Group(s)Intelligent Signal Processing, Image Analysis & Computer Graphics

Back  ::  IMM Publications