When human participants repeatedly encounter a velocity-dependent force field that distorts their movement trajectories, they adapt their motor behavior to recover straight trajectories. Computational models suggest that adaptation to a force field occurs at the action selection level through changes in the mapping between goals and actions. The quantitative prediction from these models indicates that early perturbed trajectories before adaptation and late unperturbed trajectories after adaptation should have opposite curvature, i.e. one being a mirror image of the other. We tested these predictions in a human adaptation experiment and we found that the expected mirror organization was either absent or much weaker than predicted by the models. These results are incompatible with adaptation occurring at the action selection level but compatible with adaptation occurring at the goal selection level, as if adaptation corresponds to aiming toward spatially remapped targets.

Costs (e.g. energetic expenditure) and benefits (e.g. food) are central determinants of behavior. In ecology and economics, they are combined to form a utility function which is maximized to guide choices. This principle is widely used in neuroscience as a normative model of decision and action, but current versions of this model fail to consider how decisions are actually converted into actions (i.e. the formation of trajectories). Here, we describe an approach where decision making and motor control are optimal, iterative processes derived from the maximization of the discounted, weighted difference between expected rewards and foreseeable motor efforts. The model accounts for decision making in cost/benefit situations, and detailed characteristics of control and goal tracking in realistic motor tasks. As a normative construction, the model is relevant to address the neural bases and pathological aspects of decision making and motor control.

Strong experimental evidence indicates that protein kinase and phosphatase (KP) cycles are critical to both the induction and maintenance of activity-dependent modifications in neurons. However, their contribution to information storage remains controversial, despite impressive modeling efforts. For instance, plasticity models based on KP cycles do not account for the maintenance of plastic modifications. Moreover, bistable KP cycle models that display memory fail to capture essential features of information storage: rapid onset, bidirectional control, graded amplitude, and finite lifetimes. Here, we show in a biophysical model that upstream activation of KP cycles, a ubiquitous mechanism, is sufficient to provide information storage with realistic induction and maintenance properties: plastic modifications are rapid, bidirectional, and graded, with finite lifetimes that are compatible with animal and human memory. The maintenance of plastic modifications relies on negligible reaction rates in basal conditions and thus depends on enzyme nonlinearity and activation properties of the activity-dependent KP cycle. Moreover, we show that information coding and memory maintenance are robust to stochastic fluctuations inherent to the molecular nature of activity-dependent KP cycle operation. This model provides a new principle for information storage where plasticity and memory emerge from a single dynamic process whose rate is controlled by neuronal activity. This principle strongly departs from the long-standing view that memory reflects stable steady states in biological systems, and offers a new perspective on memory in animals and humans.

For gradient descent learning to yield connectivity consistent with real biological networks, the simulated neurons would have to include more realistic intrinsic properties such as frequency adaptation. However, gradient descent learning cannot be used straightforwardly with adapting rate-model neurons because the derivative of the activation function depends on the activation history. The objectives of this study were to (1) develop a simple computational approach to reproduce mathematical gradient descent and (2) use this computational approach to provide supervised learning in a network formed of rate-model neurons that exhibit frequency adaptation. The results of mathematical gradient descent were used as a reference in evaluating the performance of the computational approach. For this comparison, standard (nonadapting) rate-model neurons were used for both approaches. The only difference was the gradient calculation: the mathematical approach used the derivative at a point in weight space, while the computational approach used the slope for a step change in weight space. Theoretically, the results of the computational approach should match those of the mathematical approach, as the step size is reduced but floating-point accuracy formed a lower limit to usable step sizes. A systematic search for an optimal step size yielded a computational approach that faithfully reproduced the results of mathematical gradient descent. The computational approach was then used for supervised learning of both connection weights and intrinsic properties of rate-model neurons to convert a tonic input into a phasic-tonic output pattern. Learning produced biologically realistic connectivity that essentially used a monosynaptic connection from the tonic input neuron to an output neuron with strong frequency adaptation as compared to a complex network when using nonadapting neurons. Thus, more biologically realistic connectivity was achieved by implementing rate-model neurons with more realistic intrinsic properties. Our computational approach could be applied to learning of other neuron properties.

The responses of neurons to time-varying injected currents are reproducible on a trial-by-trial basis in vitro, but when a constant current is injected, small variances in interspike intervals across trials add up, eventually leading to a high variance in spike timing. It is unclear whether this difference is due to the nature of the input currents or the intrinsic properties of the neurons. Neuron responses can fail to be reproducible in two ways: dynamical noise can accumulate over time and lead to a desynchronization over trials, or several stable responses can exist, depending on the initial condition. Here we show, through simulations and theoretical considerations, that for a general class of spiking neuron models, which includes, in particular, the leaky integrate-and-fire model as well as nonlinear spiking models, aperiodic currents, contrary to periodic currents, induce reproducible responses, which are stable under noise, change in initial conditions and deterministic perturbations of the input. We provide a theoretical explanation for aperiodic currents that cross the threshold.

The parametric variation in neuronal discharge according to the values of sensory or motor variables strongly influences the collective behavior of neuronal populations. A multitude of studies on the populations of broadly tuned neurons (e.g., cosine tuning) have led to such well-known computational principles as population coding, noise suppression, and line attractors. Much less is known about the properties of populations of monotonically tuned neurons. In this letter, we show that there exists an efficient weakly biased linear estimator for monotonic populations and that neural processing based on linear collective computation and least-square error learning in populations of intensity-coded neurons has specific generalization capacities.

Many neurons of the central nervous system are broadly tuned to some sensory or motor variables. This property allows one to assign to each neuron a preferred attribute (PA). The width of tuning curves and the distribution of PAs in a population of neurons tuned to a given variable define the collective behavior of the population. In this article, we study the relationship of the nature of the tuning curves, the distribution of PAs, and computational properties of linear neuronal populations. We show that noise-resistant distributed linear algebraic processing and learning can be implemented by a population of cosine tuned neurons assuming a nonuniform but regular distribution of PAs. We extend these results analytically to the noncosine tuning and uniform distribution case and show with a numerical simulation that the results remain valid for a nonuniform regular distribution of PAs for broad noncosine tuning curves. These observations provide a theoretical basis for modeling general nonlinear sensorimotor transformations as sets of local linearized representations.