Abstract
Efficient search (small search slope) is often attributed to pop-out effects that result from low-level visual features. In contrast, we have recently shown that with adequate reward training, non-human primates efficiently search for a high-value target among low-value distractors (target-present trials) regardless of basic low-level features (Ghazizadeh et al. 2016). However, the mechanism of value driven efficient search is not known. In this study, we try to address this issue in the context of a simultaneous decision-making problem. In particular, we utilize a multi-alternative drift-diffusion (MADD) model with various parameters to model decision noise, attention and decision threshold to fit search times in both value-driven target present and target absent trials. To this end, behavioral data of four macaque monkeys trained with a large number of (>300) random fractals which were arbitrarily associated with small or large rewards (i.e., "bad" or "good" objects, respectively) for varying training durations (1-day, 5-days and 30 days) were analyzed. We applied dynamic programming to fit several parametrization schemes of MADD to the data and assessed its performance using cross validation. Preliminary results indicate that longer reward training increases visual processing area differentially for good vs bad objects without significant changes in decision noise or decision threshold. Consistent with this, longer reward training increased the percentage of long-range saccades toward the good objects. Also consistently, the reduction of search slope is only observed for target-present but not target absent trials (search asymmetry). These effects expose a rich, dynamic interaction between reward history and decision making during visual search that is not necessarily explained by classical low-level guiding features. These results suggest that long-term value training may have modified the spatial extent neurons' receptive field in the ventral stream with larger effects for more valuable objects, a speculation that remains to be tested in the future.