A dynamic neural field model of continuous input integration.

The ability of neural systems to turn transient inputs into persistent changes in activity is thought to be a fundamental requirement for higher cognitive functions. In continuous attractor networks frequently used to model working memory or decision making tasks, the persistent activity settles to a stable pattern with the stereotyped shape of a "bump" independent of integration time or input strength. Here, we investigate a new bump attractor model in which the bump width and amplitude not only reflect qualitative and quantitative characteristics of a preceding input but also the continuous integration of evidence over longer timescales. The model is formalized by two coupled dynamic field equations of Amari-type which combine recurrent interactions mediated by a Mexican-hat connectivity with local feedback mechanisms that balance excitation and inhibition. We analyze the existence, stability and bifurcation structure of single and multi-bump solutions and discuss the relevance of their input dependence to modeling cognitive functions. We then systematically compare the pattern formation process of the two-field model with the classical Amari model. The results reveal that the balanced local feedback mechanisms facilitate the encoding and maintenance of multi-item memories. The existence of stable subthreshold bumps suggests that different to the Amari model, the suppression effect of neighboring bumps in the range of lateral competition may not lead to a complete loss of information. Moreover, bumps with larger amplitude are less vulnerable to noise-induced drifts and distance-dependent interaction effects resulting in more faithful memory representations over time.

Brain-inspired multiple-target tracking using Dynamic Neural Fields.

Despite considerable progress in the field of automatic multi-target tracking, several problems such as data association remained challenging. On the other hand, cognitive studies have reported that humans can robustly track several objects simultaneously. Such circumstances happen regularly in daily life, and humans have evolved to handle the associated problems. Accordingly, using brain-inspired processing principles may contribute to significantly increase the performance of automatic systems able to follow the trajectories of multiple objects. In this paper, we propose a multiple-object tracking algorithm based on dynamic neural field theory which has been proven to provide neuro-plausible processing mechanisms for cognitive functions of the brain. We define several input neural fields responsible for representing previous location and orientation information as well as instantaneous linear and angular speed of the objects in successive video frames. Image processing techniques are applied to extract the critical object features including target location and orientation. Two prediction fields anticipate the objects' locations and orientations in the upcoming frame after receiving excitatory and inhibitory inputs from the input fields in a feed-forward architecture. This information is used in the data association and labeling process. We tested the proposed algorithm on a zebrafish larvae segmentation and tracking dataset and an ant-tracking dataset containing non-rigid objects with spiky movements and frequently occurring occlusions. The results showed a significant improvement in tracking metrics compared to state-of-the-art algorithms.

Discrimination of idiopathic Parkinson's disease and vascular parkinsonism based on gait time series and the levodopa effect.

Idiopathic Parkinson's disease (IPD) and vascular parkinsonism (VaP) present highly overlapping phenotypes, making it challenging to distinguish between these two parkinsonian syndromes. Recent evidence suggests that gait assessment and response to levodopa medication may assist in the objective evaluation of clinical differences. In this paper, we propose a new approach for gait pattern differentiation that uses convolutional neural networks (CNNs) based on gait time series with and without the influence of levodopa medication. Wearable sensors positioned on both feet were used to acquire gait data from 14 VaP patients, 15 IPD patients, and 34 healthy subjects. An individual's gait features are affected by physical characteristics, including age, height, weight, sex, and walking speed or stride length. Therefore, to reduce bias due to intersubject variations, a multiple regression normalization approach was used to obtain gait data. Recursive feature elimination using the linear support vector machine, lasso, and random forest were applied to infer the optimal feature subset that led to the best results. CNNs were implemented by means of various hyperparameters and feature subsets. The best CNN classifiers achieved accuracies of 79.33%±6.46, 82.33%±10.62, and 86.00%±7.12 without (off state), with (on state), and with the simultaneous consideration of the effect of levodopa medication (off/on state), respectively. The response to levodopa medication improved classification performance. Based on gait time series and response to medication, the proposed approach differentiates between IPD and VaP gait patterns and reveals a high accuracy rate, which might prove useful when distinguishing other diseases related to movement disorders.

A dynamic model for action understanding and goal-directed imitation.

The understanding of other individuals' actions is a fundamental cognitive skill for all species living in social groups. Recent neurophysiological evidence suggests that an observer may achieve the understanding by mapping visual information onto his own motor repertoire to reproduce the action effect. However, due to differences in embodiment, environmental constraints or motor skills, this mapping very often cannot be direct. In this paper, we present a dynamic network model which represents in its layers the functionality of neurons in different interconnected brain areas known to be involved in action observation/execution tasks. The model aims at substantiating the idea that action understanding is a continuous process which combines sensory evidence, prior task knowledge and a goal-directed matching of action observation and action execution. The model is tested in variations of an imitation task in which an observer with dissimilar embodiment tries to reproduce the perceived or inferred end-state of a grasping-placing sequence. We also propose and test a biologically plausible learning scheme which allows establishing during practice a goal-directed organization of the distributed network. The modeling results are discussed with respect to recent experimental findings in action observation/execution studies.

The dynamic neural field approach to cognitive robotics.

This tutorial presents an architecture for autonomous robots to generate behavior in joint action tasks. To efficiently interact with another agent in solving a mutual task, a robot should be endowed with cognitive skills such as memory, decision making, action understanding and prediction. The proposed architecture is strongly inspired by our current understanding of the processing principles and the neuronal circuitry underlying these functionalities in the primate brain. As a mathematical framework, we use a coupled system of dynamic neural fields, each representing the basic functionality of neuronal populations in different brain areas. It implements goal-directed behavior in joint action as a continuous process that builds on the interpretation of observed movements in terms of the partner's action goal. We validate the architecture in two experimental paradigms: (1) a joint search task; (2) a reproduction of an observed or inferred end state of a grasping-placing sequence. We also review some of the mathematical results about dynamic neural fields that are important for the implementation work.

Goals and means in action observation: a computational approach.

Many of our daily activities are supported by behavioural goals that guide the selection of actions, which allow us to reach these goals effectively. Goals are considered to be important for action observation since they allow the observer to copy the goal of the action without the need to use the exact same means. The importance of being able to use different action means becomes evident when the observer and observed actor have different bodies (robots and humans) or bodily measurements (parents and children), or when the environments of actor and observer differ substantially (when an obstacle is present or absent in either environment). A selective focus on the action goals instead of the action means furthermore circumvents the need to consider the vantage point of the actor, which is consistent with recent findings that people prefer to represent the actions of others from their own individual perspective. In this paper, we use a computational approach to investigate how knowledge about action goals and means are used in action observation. We hypothesise that in action observation human agents are primarily interested in identifying the goals of the observed actor's behaviour. Behavioural cues (e.g. the way an object is grasped) may help to disambiguate the goal of the actor (e.g. whether a cup is grasped for drinking or handing it over). Recent advances in cognitive neuroscience are cited in support of the model's architecture.

Off-line simulation inspires insight: A neurodynamics approach to efficient robot task learning.

There is currently an increasing demand for robots able to acquire the sequential organization of tasks from social learning interactions with ordinary people. Interactive learning-by-demonstration and communication is a promising research topic in current robotics research. However, the efficient acquisition of generalized task representations that allow the robot to adapt to different users and contexts is a major challenge. In this paper, we present a dynamic neural field (DNF) model that is inspired by the hypothesis that the nervous system uses the off-line re-activation of initial memory traces to incrementally incorporate new information into structured knowledge. To achieve this, the model combines fast activation-based learning to robustly represent sequential information from single task demonstrations with slower, weight-based learning during internal simulations to establish longer-term associations between neural populations representing individual subtasks. The efficiency of the learning process is tested in an assembly paradigm in which the humanoid robot ARoS learns to construct a toy vehicle from its parts. User demonstrations with different serial orders together with the correction of initial prediction errors allow the robot to acquire generalized task knowledge about possible serial orders and the longer term dependencies between subgoals in very few social learning interactions. This success is shown in a joint action scenario in which ARoS uses the newly acquired assembly plan to construct the toy together with a human partner.

Review of Robotic Technology for Stereotactic Neurosurgery.

The research of stereotactic apparatus to guide surgical devices began in 1908, yet a major part of today's stereotactic neurosurgeries still rely on stereotactic frames developed almost half a century ago. Robots excel at handling spatial information, and are, thus, obvious candidates in the guidance of instrumentation along precisely planned trajectories. In this review, we introduce the concept of stereotaxy and describe a standard stereotactic neurosurgery. Neurosurgeons' expectations and demands regarding the role of robots as assistive tools are also addressed. We list the most successful robotic systems developed specifically for or capable of executing stereotactic neurosurgery. A critical review is presented for each robotic system, emphasizing the differences between them and detailing positive features and drawbacks. An analysis of the listed robotic system features is also undertaken, in the context of robotic application in stereotactic neurosurgery. Finally, we discuss the current perspective, and future directions of a robotic technology in this field. All robotic systems follow a very similar and structured workflow despite the technical differences that set them apart. No system unequivocally stands out as an absolute best. The trend of technological progress is pointing toward the development of miniaturized cost-effective solutions with more intuitive interfaces.

Dynamic field theory of movement preparation.

A theoretical framework for understanding movement preparation is proposed. Movement parameters are represented by activation fields, distributions of activation defined over metric spaces. The fields evolve under the influence of various sources of localized input, representing information about upcoming movements. Localized patterns of activation self-stabilize through cooperative and competitive interactions within the fields. The task environment is represented by a 2nd class of fields, which preshape the movement parameter representation. The model accounts for a sizable body of empirical findings on movement initiation (continuous and graded nature of movement preparation, dependence on the metrics of the task, stimulus uncertainty effect, stimulus-response compatibility effects, Simon effect, precuing paradigm, and others) and suggests new ways of exploring the structure of motor representations.

Neuro-cognitive mechanisms of decision making in joint action: a human-robot interaction study.

In this paper we present a model for action preparation and decision making in cooperative tasks that is inspired by recent experimental findings about the neuro-cognitive mechanisms supporting joint action in humans. It implements the coordination of actions and goals among the partners as a dynamic process that integrates contextual cues, shared task knowledge and predicted outcome of others' motor behavior. The control architecture is formalized by a system of coupled dynamic neural fields representing a distributed network of local but connected neural populations. Different pools of neurons encode task-relevant information about action means, task goals and context in the form of self-sustained activation patterns. These patterns are triggered by input from connected populations and evolve continuously in time under the influence of recurrent interactions. The dynamic model of joint action is evaluated in a task in which a robot and a human jointly construct a toy object. We show that the highly context sensitive mapping from action observation onto appropriate complementary actions allows coping with dynamically changing joint action situations.

Effects of attention on a relative mislocalization with successively presented stimuli.

Previous studies yielded evidence that the precision, with which stimuli are localized in the visual periphery, is improved under conditions of focused attention. The present study examined whether focused attention is able to correct a mislocalization recently observed with successively presented stimuli: when observers are asked to localize the peripheral position of a briefly presented target with respect to a previously presented comparison stimulus, they tended to judge the target as being more towards the fovea than was its actual position. In three experiments the mislocalization was tested under conditions with focused and distributed attention. Results revealed that the mislocalization increased with distributed attention and disappeared when stimuli appeared consistently at predictable positions and thus under conditions of focused attention. However, when a procedure with a trial-by-trial cueing was applied the mislocalization was only reduced, but not wiped out completely. In a recently developed dynamic field model consisting of interacting excitatory and inhibitory neuronal cell populations the results were explained as an attentional modulation of spontaneous (baseline) levels of neural activity.

Integrating verbal and nonverbal communication in a dynamic neural field architecture for human-robot interaction.

How do humans coordinate their intentions, goals and motor behaviors when performing joint action tasks? Recent experimental evidence suggests that resonance processes in the observer's motor system are crucially involved in our ability to understand actions of others', to infer their goals and even to comprehend their action-related language. In this paper, we present a control architecture for human-robot collaboration that exploits this close perception-action linkage as a means to achieve more natural and efficient communication grounded in sensorimotor experiences. The architecture is formalized by a coupled system of dynamic neural fields representing a distributed network of neural populations that encode in their activation patterns goals, actions and shared task knowledge. We validate the verbal and nonverbal communication skills of the robot in a joint assembly task in which the human-robot team has to construct toy objects from their components. The experiments focus on the robot's capacity to anticipate the user's needs and to detect and communicate unexpected events that may occur during joint task execution.

Learning to Time: a perspective.

In the last decades, researchers have proposed a large number of theoretical models of timing. These models make different assumptions concerning how animals learn to time events and how such learning is represented in memory. However, few studies have examined these different assumptions either empirically or conceptually. For knowledge to accumulate, variation in theoretical models must be accompanied by selection of models and model ideas. To that end, we review two timing models, Scalar Expectancy Theory (SET), the dominant model in the field, and the Learning-to-Time (LeT) model, one of the few models dealing explicitly with learning. In the first part of this article, we describe how each model works in prototypical concurrent and retrospective timing tasks, identify their structural similarities, and classify their differences concerning temporal learning and memory. In the second part, we review a series of studies that examined these differences and conclude that both the memory structure postulated by SET and the state dynamics postulated by LeT are probably incorrect. In the third part, we propose a hybrid model that may improve on its parents. The hybrid model accounts for the typical findings in fixed-interval schedules, the peak procedure, mixed fixed interval schedules, simple and double temporal bisection, and temporal generalization tasks. In the fourth and last part, we identify seven challenges that any timing model must meet.

Relative mislocalization of successively presented stimuli.

When observers were asked to localize the peripheral position of a briefly presented target with respect to a previously presented comparison stimulus, they tended to judge the target as being more towards the fovea than the comparison stimulus. Three experiments revealed that the mislocalization only emerged when the comparison stimulus and the target were presented successively. Varying the temporal interval between stimuli showed that the mislocalization reversed with longer stimulus-onset asynchronies. Further, the mislocalization was increased with a decrease of the spatial distance between stimuli. These findings suggested that the mislocalization originated from local excitatory and inhibitory mechanisms. Corroborating this idea a neuronal dynamic field model was successfully developed to account for the findings.

Internal models for visual perception.

Although the extrapolation of past perceptual history into the immediate and distant future is a fundamental phenomenon in everyday life, the underlying processing mechanisms are not well understood. A network model consisting of interacting excitatory and inhibitory cell populations coding for stimulus position is used to study the neuronal population response to a continuously moving stimulus. An adaptation mechanism is proposed that offers the possibility to control and modulate motion-induced extrapolation without changing the spatial interaction structure within the network. Using an occluder paradigm, functional advantages of an internally generated model of a moving stimulus are discussed. It is shown that the integration of such a model in processing leads to a faster and more reliable recognition of the input stream and allows for object permanence following occlusion. The modeling results are discussed in relation to recent experimental findings that show motion-induced extrapolation.

Shorter latencies for motion trajectories than for flashes in population responses of cat primary visual cortex.

Psychophysical evidence in humans indicates that localization is different for stationary flashed and coherently moving objects. To address how the primary visual cortex represents object position we used a population approach that pools spiking activity of many neurones in cat area 17. In response to flashed stationary squares (0.4 deg) we obtained localized activity distributions in visual field coordinates, which we referred to as profiles across a 'population receptive field' (PRF). We here show how motion trajectories can be derived from activity across the PRF and how the representation of moving and flashed stimuli differs in position. We found that motion was represented by peaks of population activity that followed the stimulus with a speed-dependent lag. However, time-to-peak latencies were shorter by approximately 16 ms compared to the population responses to stationary flashes. In addition, motion representation showed a directional bias, as latencies were more reduced for peripheral-to-central motion compared to the opposite direction. We suggest that a moving stimulus provides 'preactivation' that allows more rapid processing than for a single flash event.