Learning joint representations for order and timing of perceptual-motor sequences: A dynamic neural field approach

Abstract

Many of our everyday tasks require the control of the serial order and the timing of component actions. Using the dynamic neural field (DNF) framework, we address the learning of representations that support the performance of precisely time action sequences. In continuation of previous modeling work and robotics implementations, we ask specifically the question how feedback about executed actions might be used by the learning system to fine tune a joint memory representation of the ordinal and the temporal structure which has been initially acquired by observation. The perceptual memory is represented by a self-stabilized, multi-bump activity pattern of neurons encoding instances of a sensory event (e.g., color, position or pitch) which guides sequence learning. The strength of the population representation of each event is a function of elapsed time since sequence onset. We propose and test in simulations a simple learning rule that detects a mismatch between the expected and realized timing of events and adapts the activation strengths in order to compensate for the movement time needed to achieve the desired effect. The simulation results show that the effector-specific memory representation can be robustly recalled. We discuss the impact of the fast, activationbased learning that the DNF framework provides for robotics applications.

Towards temporal cognition for robots: A neurodynamics approach

“…Table I compares the delays before and after the application of the adaptation dynamics. It shows that the delay can be drastically reduced in a single trial (for a discussion of the adaptation rate parameter β see [22]).…”

“…The strength of individual bumps controls the timing of action onset during recall in order to finish the individual actions in synchrony with an external cue. Starting with the perceptual memory pattern which u AM EM receives as excitatory input, the bump amplitudes are adjusted according to a detected temporal mismatch between the time courses of suprathreshold activity in the Cue field u S and the Feedback field u F , representing the external synchronization signal and the sensory feedback about executed actions, respectively [22]. Both fields receive in addition to the sensory input also the pattern of u AM EM as subthreshold excitation.…”

“…Elapsed time since cue onset can be robustly encoded in the consistent increase of population activity during a delay period [13], [16]. We have recently implemented the notion of shared neural resources for STM and time measurement in a dynamic neural field (DNF) model of sequence learning [9], [22]. The information about the order and relative timing of successive stimulus events is stored in an activation gradient of persistent activity over stimulus-specific subpopulations.…”

Rapid Learning of Complex Sequences With Time Constraints: A Dynamic Neural Field Model

“…A prerequisite is that the preshape-threshold distance in u D of f has been adjusted to represent duration. This can be achieved by applying the local version of the adaptation dynamics (6) to the memory representations in u M of f [46].…”

Combining Spatial and Parametric Working Memory in a Dynamic Neural Field Model

“…Since their level of abstraction favors analytical treatment [4], dynamic field models are also utilized for the development of new technical solutions inspired by neural processing principles [6]. The two specific challenges we address in the present study are motivated by applications of a multi-item working memory [16]. The first question is concerned with the impact on the memory representation when multiple sensory events are presented sequentially as opposed to simultaneously.…”