1、An Integrated Model of Decision Making and Visual Attention,Philip L. Smith University of Melbourne,Collaborators: Roger Ratcliff, Bradley Wolfgang,Attention and Decision Making,Psychophysical “front end” provides input to decision mechanisms Visual search (saccade-to-target) task is attentional tas
2、k Areas implicated in decision making (LIP, FEF, SC) also implicated in attentional control (e.g., LIP as a “salience map”) Visual signal detection: close coupling of attention and decision mechanisms,Attentional Cuing Effects in Visual Signal Detection,Posner paradigm, 180 ms cue-target interval Or
3、thogonal discrimination (proxy for detection) Do attentional cues enhance detectability of luminance targets? Historically controversial,Attentional Cuing Effects in Visual Signal Detection,Depends on: Dependent variable: RT or accuracy How you limit detectability: with or without backward masks,Smi
4、th, Ratcliff & Wolfgang (2004),Detection sensitivity increased by cues only with masked stimuli (mask-dependent cuing) RT decreased by cues for both masked and unmasked stimuli Interaction between attention and decisions mechanisms Smith (2000), Smith & Wolfgang (2004), Smith, Wolfgang & Sinclair (2
5、004), Smith & Wolfgang (2005), Gould, Smith & Wolfgang (in prep.),A Model of Decision Making and Visual Attention,Link visual encoding, masking, spatial attention, visual short term memory and decision making,A Model of Decision Making and Visual Attention,Link visual encoding, masking, spatial atte
6、ntion, visual short term memory and decision making,Visual Encoding and Masking,Stimuli encoded by low-pass filters Masks limit visual persistence of stimuli Unmasked: slow iconic decay Masked: Rapid suppression by mask (interruption masking) Smith & Wolfgang (2004, 2005),Attention and Visual Short
7、Term Memory,VSTM Shunting Equation,Trace strength modeled by shunting equation (Grossberg, Hodgkin-Huxley) Preserve STM activity after stimulus offset Opponent-channel coding prevents saturation (bounded between -b and +b) Recodes luminances as contrasts,Attentional Dynamics,I. Gain Model. Affects r
8、ate of uptake into VSTM:,II. Orienting Model. Affects time of entry into VSTM:,Attentional Dynamics,I. Gain Model. Affects rate of uptake into VSTM:,II. Orienting Model. Affects time of entry into VSTM:,Decision Model,I. (Wiener) Diffusion Model (Ratcliff, 1978),VSTM trace strength determines (nonst
9、ationary) drift Orientation determines sign of drift Contrast determines size of drift Within-trial decision noise determines diffusion coefficient Between-trial encoding noise determines drift variability,II. Dual Diffusion (Smith, 2000; Ratcliff & Smith 2004),Information for competing responses ac
10、cumulated in separate totals Parallel Ornstein-Uhlenbeck diffusion processes (accumulation with decay) Symmetrical stimulus representation (equal and opposite drifts),Attentional Dynamics (Gain Model),Gain interacts with masking to determine VSTM trace strength via shunting equation,Gain Model + Dif
11、fusion,Quantile probability plot: RT quantiles .1,.3,.5,.7,.9 vs. probability Quantile averaged data Correct and error RT Drift amplitude is Naka-Rushton function of contrast (c):,Gain Model + Diffusion,220 data degrees of freedom 14 parameters: 3 Naka-Rushton drift parameters 3 encoding filter para
12、meters 2 attentional gains 2 drift variability parameters 2 decision criteria 2 post-decision parameters,Model Summary,Dual diffusion has same parameters as single diffusion plus additional OU decay parameter,Conclusions,Need model linking visual encoding, masking, VSTM, attention, decision making S
13、tochastic dynamic framework with sequential sampling decision models Predicts shapes of entire RT distributions for correct responses and errors, choice probabilities Possible neural substrate? Behavioral diffusion from Poisson shot noise Accumulated information modeled as integrated OU diffusion; closely approximates Wiener diffusion,
