Refine
Year of publication
Document Type
- Article (33089)
- Part of Periodical (12123)
- Book (8392)
- Doctoral Thesis (5943)
- Part of a Book (4047)
- Working Paper (3432)
- Review (2969)
- Preprint (2589)
- Contribution to a Periodical (2484)
- Conference Proceeding (1793)
Language
- German (43963)
- English (32330)
- French (1110)
- Portuguese (845)
- Multiple languages (335)
- Spanish (311)
- Croatian (302)
- Italian (198)
- mis (173)
- Turkish (168)
Keywords
- Deutsch (1087)
- Literatur (894)
- taxonomy (794)
- Deutschland (559)
- Rezension (521)
- new species (470)
- Rezeption (360)
- Frankfurt <Main> / Universität (341)
- Übersetzung (340)
- Geschichte (304)
Institute
- Medizin (8292)
- Präsidium (5389)
- Physik (5195)
- Wirtschaftswissenschaften (2796)
- Extern (2742)
- Gesellschaftswissenschaften (2474)
- Biowissenschaften (2297)
- Frankfurt Institute for Advanced Studies (FIAS) (2036)
- Biochemie und Chemie (2010)
- Informatik (1808)
Graph Neural Networks (GNNs) are a popular class of machine learning models. Inspired by the learning to explain (L2X) paradigm, we propose L2xGnn, a framework for explainable GNNs which provides faithful explanations by design. L2xGnn learns a mechanism for selecting explanatory subgraphs (motifs) which are exclusively used in the GNNs message-passing operations. L2xGnn is able to select, for each input graph, a subgraph with specific properties such as being sparse and connected. Imposing such constraints on the motifs often leads to more interpretable and effective explanations. Experiments on several datasets suggest that L2xGnn achieves the same classification accuracy as baseline methods using the entire input graph while ensuring that only the provided explanations are used to make predictions. Moreover, we show that L2xGnn is able to identify motifs responsible for the graph’s properties it is intended to predict.
The treatment of isolated calcified lesions involving the popliteal artery are demanding and they often require stent placement to achieve acceptable luminal gain. This study evaluates the safety and performance of the orbital atherectomy system (Cardiovascular Systems Inc.) and percutaneous transluminal angioplasty with a drug-coated balloon (PTA-DCB) for the treatment of chronic atherosclerotic lesions within the popliteal artery segment. From November 2018 to November 2019, a series of six patients with Rutherford classification stage III peripheral arterial disease with isolated, focal, calcified occlusions of the popliteal artery were treated with orbital atherectomy followed by PTA-DCB. No embolic protection devices were used. The technical success rate was 100%, the primary patency rate was 100% at 7.0 (±4.2) months of follow-up. The Rutherford classification improved in all patients from stage III to stage II and the mean ankle brachial pressure index after the procedure was 0.97 (±0.08), with a preoperative mean ankle brachial pressure index of 0.69 (±0.21). In one instance, spasm was noted in a distal arterial bed and it was successfully treated with local nitroglycerine. No distal embolisation, perforation or aneurysmal degeneration was observed. During follow-up there were no deaths, major amputations or revascularisation of target lesions. The use of orbital atherectomy system in combination with PTA-DCB was found to be safe and effective in modifying focal, chronic, calcified plaques in the popliteal artery segment in these six cases.
Archaeological research at Al-Khashbah, Sultanate of Oman, conducted by the University of Tübingen, revealed a large Early Bronze Age (3rd millennium BCE) site. During the intensive surface survey and excavations, several ground stone tools were found. Most of them came from the vicinity of monumental stone and mud-brick structures, so-called towers, and are clearly connected to copper-processing waste such as slag, furnace fragments and prills, i.e., droplets of molten copper. Therefore, it is assumed that these ground stone tools were used within the operational procedures of copper-processing. Interestingly, only the monumental towers from the first half of the 3rd millennium BCE, i.e., the Hafit period, feature larger quantities of ground stone tools as well as copper processing waste. Towers from the second half of the 3rd millennium BCE, i.e., the Umm an-Nar period, have none. Within the scope of this paper, the distribution of the different types of ground stone tools in Al-Khashbah as well as their find context will be presented. They are illustrated with drawings generated from 3D models created using digital photography processed with the software Agisoft Photoscan. Comparisons with other 3rd millennium BCE sites in Eastern Arabia show that there as well, copper-processing remains are often associated with ground stone tools. The overall variety of types seems to be rather homogeneous in the region.
Othering ist nicht nur ein gesellschaftliches Phänomen, sondern muss auch in seiner Relevanz für die wissenschaftliche Analyse reflektiert werden. Anhand von Beispielen aus der Forschungspraxis diskutieren die Beiträger*innen, wie das theoretische Konzept des Othering in der qualitativen Forschung fruchtbar gemacht werden kann. Dabei loten sie dessen kritisches und produktives Potenzial sowohl in theoretischer als auch in epistemologischer, methodologischer und forschungspraktischer Hinsicht aus. Sie analysieren Othering in der postmigrantischen Gesellschaft empirisch, machen es auf diese Weise sichtbar und fragen nach den Möglichkeiten und Grenzen von Reflexivität für eine kritische Wissensproduktion.
Graph Neural Networks (GNNs) are a popular class of machine learning models. Inspired by the learning to explain (L2X) paradigm, we propose L2XGNN, a framework for explainable GNNs which provides faithful explanations by design. L2XGNN learns a mechanism for selecting explanatory subgraphs (motifs) which are exclusively used in the GNNs message-passing operations. L2XGNN is able to select, for each input graph, a subgraph with specific properties such as being sparse and connected. Imposing such constraints on the motifs often leads to more interpretable and effective explanations. Experiments on several datasets suggest that L2XGNN achieves the same classification accuracy as baseline methods using the entire input graph while ensuring that only the provided explanations are used to make predictions. Moreover, we show that L2XGNN is able to identify motifs responsible for the graph's properties it is intended to predict.
Graph Neural Networks (GNNs) are a popular class of machine learning models. Inspired by the learning to explain (L2X) paradigm, we propose L2XGNN, a framework for explainable GNNs which provides faithful explanations by design. L2XGNN learns a mechanism for selecting explanatory subgraphs (motifs) which are exclusively used in the GNNs message-passing operations. L2XGNN is able to select, for each input graph, a subgraph with specific properties such as being sparse and connected. Imposing such constraints on the motifs often leads to more interpretable and effective explanations. Experiments on several datasets suggest that L2XGNN achieves the same classification accuracy as baseline methods using the entire input graph while ensuring that only the provided explanations are used to make predictions. Moreover, we show that L2XGNN is able to identify motifs responsible for the graph's properties it is intended to predict.
Graph Neural Networks (GNNs) are a popular class of machine learning models. Inspired by the learning to explain (L2X) paradigm, we propose L2XGNN, a framework for explainable GNNs which provides faithful explanations by design. L2XGNN learns a mechanism for selecting explanatory subgraphs (motifs) which are exclusively used in the GNNs message-passing operations. L2XGNN is able to select, for each input graph, a subgraph with specific properties such as being sparse and connected. Imposing such constraints on the motifs often leads to more interpretable and effective explanations. Experiments on several datasets suggest that L2XGNN achieves the same classification accuracy as baseline methods using the entire input graph while ensuring that only the provided explanations are used to make predictions. Moreover, we show that L2XGNN is able to identify motifs responsible for the graph's properties it is intended to predict.
Graph Neural Networks (GNNs) are a popular class of machine learning models. Inspired by the learning to explain (L2X) paradigm, we propose L2XGNN, a framework for explainable GNNs which provides faithful explanations by design. L2XGNN learns a mechanism for selecting explanatory subgraphs (motifs) which are exclusively used in the GNNs message-passing operations. L2XGNN is able to select, for each input graph, a subgraph with specific properties such as being sparse and connected. Imposing such constraints on the motifs often leads to more interpretable and effective explanations. Experiments on several datasets suggest that L2XGNN achieves the same classification accuracy as baseline methods using the entire input graph while ensuring that only the provided explanations are used to make predictions. Moreover, we show that L2XGNN is able to identify motifs responsible for the graph's properties it is intended to predict.
The de-Sitter spacetime is a maximally symmetric Lorentzian manifold with constant positive scalar curvature that plays a fundamental role in modern cosmology. Here, we investigate bulk-viscosity-assisted quasi de-Sitter inflation, that is the period of accelerated expansion in the early universe during which −𝐻˙≪𝐻2 , with 𝐻(𝑡) being the Hubble expansion rate. We do so in the framework of a causal theory of relativistic hydrodynamics, which takes into account non-equilibrium effects associated with bulk viscosity, which may have been present as the early universe underwent an accelerated expansion. In this framework, the existence of a quasi de-Sitter universe emerges as a natural consequence of the presence of bulk viscosity, without requiring introducing additional scalar fields. As a result, the equation of state, determined by numerically solving the generalized momentum-conservation equation involving bulk viscosity pressure turns out to be time dependent. The transition timescale characterising its departure from an exact de-Sitter phase is intricately related to the magnitude of the bulk viscosity. We examine the properties of the new equation of state, as well as the transition timescale in the presence of bulk viscosity pressure. In addition, we construct a fluid description of inflation and demonstrate that, in the context of the causal formalism, it is equivalent to the scalar field theory of inflation. Our analysis also shows that the slow-roll conditions are realised in the bulk-viscosity-supported model of inflation. Finally, we examine the viability of our model by computing the inflationary observables, namely the spectral index and the tensor-to-scalar ratio of the curvature perturbations, and compare them with a number of different observations, finding good agreement in most cases.