Action choices are influenced by recent past and predicted future action states. Here, we demonstrate that recent hand-choice history affects both current hand choices and response times to initiate actions. Participants reach to contact visible targets using one hand. Hand choice is biased in favour of which hand was used recently, in particular, when the biomechanical costs of responding with either hand are similar, and repeated choices lead to reduced response times. These effects are also found to positively correlate. Participants who show strong effects of recent history on hand choice also tend to show strong effects of recent history on response times. The data are consistent with a computational efficiency interpretation whereby repeated action choices confer computational gains in the efficiency of underpinning processes. We discuss our results within the framework of this model, and with respect to balancing predicted gains and losses, and speculate about the possible underlying mechanisms in neural terms.

Neuropsychological training methods of visual rehabilitation for homonymous vision loss caused by postchiasmatic damage fall into two fundamental paradigms: “compensation” and “restoration”. Existing methods can be classified into three groups: Visual Scanning Training (VST), Audio-Visual Scanning Training (AViST) and Vision Restoration Training (VRT). VST and AViST aim at compensating vision loss by training eye scanning movements, whereas VRT aims at improving lost vision by activating residual visual functions by training light detection and discrimination of visual stimuli. This review discusses the rationale underlying these paradigms and summarizes the available evidence with respect to treatment efficacy. The issues raised in our review should help guide clinical care and stimulate new ideas for future research uncovering the underlying neural correlates of the different treatment paradigms. We propose that both local “within-system” interactions (i.e., relying on plasticity within peri-lesional spared tissue) and changes in more global “between-system” networks (i.e., recruiting alternative visual pathways) contribute to both vision restoration and compensatory rehabilitation, which ultimately have implications for the rehabilitation of cognitive functions.

These results suggest that the primary visual deficit in hemianopic patients might be accompanied by an ipsilesional attentional bias which might be reduced by multisensory stimulation.

The model-free algorithms of "reinforcement learning" (RL) have gained clout across disciplines, but so too have model-based alternatives. The present study emphasizes other dimensions of this model space in consideration of associative or discriminative generalization across states and actions. This "generalized reinforcement learning" (GRL) model, a frugal extension of RL, parsimoniously retains the single rewardprediction error (RPE), but the scope of learning goes beyond the experienced state and action. Instead, the generalized RPE is efficiently relayed for bidirectional counterfactual updating of value estimates for other representations. Aided by structural information but as an implicit rather than explicit cognitive map, GRL provided the most precise account of human behavior and individual differences in a reversallearning task with hierarchical structure that encouraged inverse generalization across both states and actions. Reflecting inference that could be true, false (i.e., overgeneralization), or absent (i.e., undergeneralization), state generalization distinguished those who learned well more so than action generalization. With highresolution high-field fMRI targeting the dopaminergic midbrain, the GRL model's RPE signals (alongside value and decision signals) were localized within not only the striatum but also the substantia nigra and the ventral tegmental area, including specific effects of generalization that also extend to the hippocampus. Factoring in generalization as a multidimensional process in value-based learning, these findings shed light on complexities that, while challenging classic RL, can still be resolved within the bounds of its core computations.