We present GNNAutoScale (GAS), a framework for scaling arbitrary message-passing GNNs to large graphs. GAS prunes entire sub-trees of the computation graph by utilizing historical embeddings from prior training iterations, leading to constant GPU memory consumption in respect to input node size without dropping any data. While existing solutions weaken the expressive power of message passing due to sub-sampling of edges or non-trainable propagations, our approach is provably able to maintain the expressive power of the original GNN. We achieve this by providing approximation error bounds of historical embeddings and show how to tighten them in practice. Empirically, we show that the practical realization of our framework, PyGAS, an easy-to-use extension for PyTorch Geometric, is both fast and memory-efficient, learns expressive node representations, closely resembles the performance of their non-scaling counterparts, and reaches state-of-the-art performance on large-scale graphs.

We present Spline-based Convolutional Neural Networks (SplineCNNs), a variant of deep neural networks for irregular structured and geometric input, e.g., graphs or meshes. Our main contribution is a novel convolution operator based on B-splines, that makes the computation time independent from the kernel size due to the local support property of the B-spline basis functions. As a result, we obtain a generalization of the traditional CNN convolution operator by using continuous kernel functions parametrized by a fixed number of trainable weights. In contrast to related approaches that filter in the spectral domain, the proposed method aggregates features purely in the spatial domain. In addition, SplineCNN allows entire end-to-end training of deep architectures, using only the geometric structure as input, instead of handcrafted feature descriptors. For validation, we apply our method on tasks from the fields of image graph classification, shape correspondence and graph node classification, and show that it outperforms or pars state-of-the-art approaches while being significantly faster and having favorable properties like domain-independence. Our source code is available on GitHub1.

Cooperative working of man and machine in close proximity is considered as an enabling feature for the manufacturing industry where cognitively and physically demanding work steps concur. Acceptance and performance are crucial aspects for the success of such working environments. Understanding the effects of robots on human's behavior plays a central role. In this study, we investigate the effects of robotic motion on humans within cooperative working environments. We therefore conducted an experiment, within the theoretical context of approach-avoidance behavior, by analyzing human motion trajectories during the interaction with an industrial robot in a real setup. Our findings suggest that an active robot affects the movement behavior of interacting participants considerably. Interestingly, an active robot seems to be more positively evaluated than a non-moving robot. The results suggest that approach and avoidance behavior may be an implicit measure of the affective quality of human-robot interaction.

The present paper describes a novel approach to performing feature extraction and classification in possibly layered circular structures, as seen in two-dimensional cutting planes of three-dimensional tube-shaped objects. The algorithm can therefore be used to analyze histological specimens of blood vessels as well as intravascular ultrasound (IVUS) datasets. The approach uses a radial signal-based extraction of textural features in combination with methods of machine learning to integrate a priori domain knowledge. The algorithm in principle solves a two-dimensional classification problem that is reduced to parallel viable time series analysis. A multiscale approach hereby determines a feature vector for each analysis using either a Wavelet-transform (WT) or a S-transform (ST). The classification is done by methods of machine learning — here support vector machines. A modified marching squares algorithm extracts the polygonal segments for the two-dimensional classification. The accuracy is above 80% even in datasets with a considerable quantity of artifacts, while the mean accuracy is above 90%. The benefit of the approach therefore mainly lies in its robustness, efficient calculation, and the integration of domain knowledge.