Fig. 9: The effect of three constraints in the CCTELM model on the goodness-of-fit of different eye movement metrics. | Communications Biology

Fig. 9: The effect of three constraints in the CCTELM model on the goodness-of-fit of different eye movement metrics.

From: Human visual search follows a suboptimal Bayesian strategy revealed by a spatiotemporal computational model and experiment

Fig. 9: The effect of three constraints in the CCTELM model on the goodness-of-fit of different eye movement metrics.

In all panels, “CCTELM” represents the full CCTELM model; “Inf Mem” represents the CCTELM model with infinite memory capacity; “No SacAmp Penalty” represents the CCTELM model without saccade amplitude penalty function; “No SLP Error” represents the CCTELM model without saccade landing position offset and variance. The last three models were evaluated using the same parameters as the CCTELM model. a–c Bhattacharyya coefficient (Bc) between the subjects’ and models’ overall fixation duration (a), saccade amplitude (b), and fixation location (c) distributions. d–f Average Bc between the subjects’ and models’ fixation duration (d), saccade amplitude (e), and fixation distance to image center (f) distributions of the initial 20 fixations or saccades. The average Bc was obtained by first calculating 20 Bc for each of the initial 20 fixations or saccades after the start of a trial (thus obtaining 20 Bc values at each ordinal position in eye movement sequences), and then averaging the 20 Bc values. The 99% confidence interval (shown in error bar) was calculated by bootstrapping the experimental and simulation data for n = 2000 times and then multiplying the SD of all Bc by 2.576. The violin plot represents the distribution of Bc from all bootstrapping trials.

Back to article page