Linking meta-learning to meta-structure.

We propose that a principled understanding of meta-learning, as aimed for by the authors, benefits from linking the focus on learning with an equally strong focus on structure, which means to address the question: What are the meta-structures that can guide meta-learning?

Decentralized control and local information for robust and adaptive decentralized Deep Reinforcement Learning.

Decentralization is a central characteristic of biological motor control that allows for fast responses relying on local sensory information. In contrast, the current trend of Deep Reinforcement Learning (DRL) based approaches to motor control follows a centralized paradigm using a single, holistic controller that has to untangle the whole input information space. This motivates to ask whether decentralization as seen in biological control architectures might also be beneficial for embodied sensori-motor control systems when using DRL. To answer this question, we provide an analysis and comparison of eight control architectures for adaptive locomotion that were derived for a four-legged agent, but with their degree of decentralization varying systematically between the extremes of fully centralized and fully decentralized. Our comparison shows that learning speed is significantly enhanced in distributed architectures-while still reaching the same high performance level of centralized architectures-due to smaller search spaces and local costs providing more focused information for learning. Second, we find an increased robustness of the learning process in the decentralized cases-it is less demanding to hyperparameter selection and less prone to becoming trapped in poor local minima. Finally, when examining generalization to uneven terrains-not used during training-we find best performance for an intermediate architecture that is decentralized, but integrates only local information from both neighboring legs. Together, these findings demonstrate beneficial effects of distributing control into decentralized units and relying on local information. This appears as a promising approach towards more robust DRL and better generalization towards adaptive behavior.

What Does a Hand-Over Tell?-Individuality of ShortMotion Sequences.

How much information with regard to identity and further individual participantcharacteristics are revealed by relatively short spatio-temporal motion trajectories of a person?We study this question by selecting a set of individual participant characteristics and analysingmotion captured trajectories of an exemplary class of familiar movements, namely handover of anobject to another person. The experiment is performed with different participants under different,predefined conditions. A selection of participant characteristics, such as the Big Five personalitytraits, gender, weight, or sportiness, are assessed and we analyse the impact of the three factor groups"participant identity", "participant characteristics", and "experimental conditions" on the observedhand trajectories. The participants' movements are recorded via optical marker-based hand motioncapture. One participant, the giver, hands over an object to the receiver. The resulting time courses ofthree-dimensional positions of markers are analysed. Multidimensional scaling is used to projecttrajectories to points in a dimension-reduced feature space. Supervised learning is also applied.We find that "participant identity" seems to have the highest correlation with the trajectories, withfactor group "experimental conditions" ranking second. On the other hand, it is not possible to find acorrelation between the "participant characteristics" and the hand trajectory features.

Perceptual grouping by entrainment in coupled Kuramoto oscillator networks.

In this article we present a network composed of coupled Kuramoto oscillators, which is able to solve a broad spectrum of perceptual grouping tasks. Based on attracting and repelling interactions between these oscillators, the network dynamics forms various phase-synchronized clusters of oscillators corresponding to individual groups of similar input features. The degree of similarity between features is determined by a set of underlying receptive fields, which are learned directly from the feature domain. After illustrating the theoretical principles of the network, the approach is evaluated in an image segmentation task. Furthermore, the influence of a varying degree of sparse couplings is evaluated.

An adaptive P300-based online brain-computer interface.

The P300 component of an event related potential is widely used in conjunction with brain-computer interfaces (BCIs) to translate the subjects intent by mere thoughts into commands to control artificial devices. A well known application is the spelling of words while selection of the letters is carried out by focusing attention to the target letter. In this paper, we present a P300-based online BCI which reaches very competitive performance in terms of information transfer rates. In addition, we propose an online method that optimizes information transfer rates and/or accuracies. This is achieved by an algorithm which dynamically limits the number of subtrial presentations, according to the subject's current online performance in real-time. We present results of two studies based on 19 different healthy subjects in total who participated in our experiments (seven subjects in the first and 12 subjects in the second one). In the first, study peak information transfer rates up to 92 bits/min with an accuracy of 100% were achieved by one subject with a mean of 32 bits/min at about 80% accuracy. The second experiment employed a dynamic classifier which enables the user to optimize bitrates and/or accuracies by limiting the number of subtrial presentations according to the current online performance of the subject. At the fastest setting, mean information transfer rates could be improved to 50.61 bits/min (i.e., 13.13 symbols/min). The most accurate results with 87.5% accuracy showed a transfer rate of 29.35 bits/min.

The dynamic wave expansion neural network model for robot motion planning in time-varying environments.

We introduce a new type of neural network--the dynamic wave expansion neural network (DWENN)--for path generation in a dynamic environment for both mobile robots and robotic manipulators. Our model is parameter-free, computationally efficient, and its complexity does not explicitly depend on the dimensionality of the configuration space. We give a review of existing neural networks for trajectory generation in a time-varying domain, which are compared to the presented model. We demonstrate several representative simulative comparisons as well as the results of long-run comparisons in a number of randomly-generated scenes, which reveal that the proposed model yields dominantly shorter paths, especially in highly-dynamic environments.