TY - CONF AB - Manufacturing companies face the challenge of reaching required quality standards. Using optical sensors and deep learning might help. However, training deep learning algorithms require large amounts of visual training data. Using domain randomization to generate synthetic image data can alleviate this bottleneck. This paper presents the application of synthetic image training data for optical quality inspections using visual sensor technology. The results show synthetically generated training data are appropriate for visual quality inspections. AU - Gräßler, Iris AU - Hieb, Michael ID - 52816 KW - synthetic training data KW - machine vision quality gates KW - deep learning KW - automated inspection and quality control KW - production control T2 - Lectures TI - Creating Synthetic Training Datasets for Inspection in Machine Vision Quality Gates in Manufacturing ER - TY - JOUR AB - Ideational impact refers to the uptake of a paper's ideas and concepts by subsequent research. It is defined in stark contrast to total citation impact, a measure predominantly used in research evaluation that assumes that all citations are equal. Understanding ideational impact is critical for evaluating research impact and understanding how scientific disciplines build a cumulative tradition. Research has only recently developed automated citation classification techniques to distinguish between different types of citations and generally does not emphasize the conceptual content of the citations and its ideational impact. To address this problem, we develop Deep Content-enriched Ideational Impact Classification (Deep-CENIC) as the first automated approach for ideational impact classification to support researchers' literature search practices. We evaluate Deep-CENIC on 1,256 papers citing 24 information systems review articles from the IT business value domain. We show that Deep-CENIC significantly outperforms state-of-the-art benchmark models. We contribute to information systems research by operationalizing the concept of ideational impact, designing a recommender system for academic papers based on deep learning techniques, and empirically exploring the ideational impact of the IT business value domain. AU - Prester, Julian AU - Wagner, Gerit AU - Schryen, Guido AU - Hassan, Nik Rushdi ID - 20212 IS - January JF - Decision Support Systems KW - Ideational impact KW - citation classification KW - academic recommender systems KW - natural language processing KW - deep learning KW - cumulative tradition TI - Classifying the Ideational Impact of Information Systems Review Articles: A Content-Enriched Deep Learning Approach VL - 140 ER - TY - CONF AB - Over the last years, several approaches for the data-driven estimation of expected possession value (EPV) in basketball and association football (soccer) have been proposed. In this paper, we develop and evaluate PIVOT: the first such framework for team handball. Accounting for the fast-paced, dynamic nature and relative data scarcity of hand- ball, we propose a parsimonious end-to-end deep learning architecture that relies solely on tracking data. This efficient approach is capable of predicting the probability that a team will score within the near future given the fine-grained spatio-temporal distribution of all players and the ball over the last seconds of the game. Our experiments indicate that PIVOT is able to produce accurate and calibrated probability estimates, even when trained on a relatively small dataset. We also showcase two interactive applications of PIVOT for valuing actual and counterfactual player decisions and actions in real-time. AU - Müller, Oliver AU - Caron, Matthew AU - Döring, Michael AU - Heuwinkel, Tim AU - Baumeister, Jochen ID - 24547 KW - expected possession value KW - handball KW - tracking data KW - time series classification KW - deep learning T2 - 8th Workshop on Machine Learning and Data Mining for Sports Analytics (ECML PKDD 2021) TI - PIVOT: A Parsimonious End-to-End Learning Framework for Valuing Player Actions in Handball using Tracking Data ER - TY - CONF AB - Modern services comprise interconnected components, e.g., microservices in a service mesh, that can scale and run on multiple nodes across the network on demand. To process incoming traffic, service components have to be instantiated and traffic assigned to these instances, taking capacities and changing demands into account. This challenge is usually solved with custom approaches designed by experts. While this typically works well for the considered scenario, the models often rely on unrealistic assumptions or on knowledge that is not available in practice (e.g., a priori knowledge). We propose a novel deep reinforcement learning approach that learns how to best coordinate services and is geared towards realistic assumptions. It interacts with the network and relies on available, possibly delayed monitoring information. Rather than defining a complex model or an algorithm how to achieve an objective, our model-free approach adapts to various objectives and traffic patterns. An agent is trained offline without expert knowledge and then applied online with minimal overhead. Compared to a state-of-the-art heuristic, it significantly improves flow throughput and overall network utility on real-world network topologies and traffic traces. It also learns to optimize different objectives, generalizes to scenarios with unseen, stochastic traffic patterns, and scales to large real-world networks. AU - Schneider, Stefan Balthasar AU - Manzoor, Adnan AU - Qarawlus, Haydar AU - Schellenberg, Rafael AU - Karl, Holger AU - Khalili, Ramin AU - Hecker, Artur ID - 19609 KW - self-driving networks KW - self-learning KW - network coordination KW - service coordination KW - reinforcement learning KW - deep learning KW - nfv T2 - IEEE International Conference on Network and Service Management (CNSM) TI - Self-Driving Network and Service Coordination Using Deep Reinforcement Learning ER - TY - CONF AB - This paper deals with aspect phrase extraction and classification in sentiment analysis. We summarize current approaches and datasets from the domain of aspect-based sentiment analysis. This domain detects sentiments expressed for individual aspects in unstructured text data. So far, mainly commercial user reviews for products or services such as restaurants were investigated. We here present our dataset consisting of German physician reviews, a sensitive and linguistically complex field. Furthermore, we describe the annotation process of a dataset for supervised learning with neural networks. Moreover, we introduce our model for extracting and classifying aspect phrases in one step, which obtains an F1-score of 80%. By applying it to a more complex domain, our approach and results outperform previous approaches. AU - Kersting, Joschka AU - Geierhos, Michaela ID - 15580 KW - Deep Learning KW - Natural Language Processing KW - Aspect-based Sentiment Analysis T2 - Proceedings of the 12th International Conference on Agents and Artificial Intelligence (ICAART 2020) -- Special Session on Natural Language Processing in Artificial Intelligence (NLPinAI 2020) TI - Aspect Phrase Extraction in Sentiment Analysis with Deep Learning ER - TY - CONF AB - In this work we focus on the well-known Euclidean Traveling Salesperson Problem (TSP) and two highly competitive inexact heuristic TSP solvers, EAX and LKH, in the context of per-instance algorithm selection (AS). We evolve instances with nodes where the solvers show strongly different performance profiles. These instances serve as a basis for an exploratory study on the identification of well-discriminating problem characteristics (features). Our results in a nutshell: we show that even though (1) promising features exist, (2) these are in line with previous results from the literature, and (3) models trained with these features are more accurate than models adopting sophisticated feature selection methods, the advantage is not close to the virtual best solver in terms of penalized average runtime and so is the performance gain over the single best solver. However, we show that a feature-free deep neural network based approach solely based on visual representation of the instances already matches classical AS model results and thus shows huge potential for future studies. AU - Seiler, Moritz AU - Pohl, Janina AU - Bossek, Jakob AU - Kerschke, Pascal AU - Trautmann, Heike ID - 48897 KW - Automated algorithm selection KW - Deep learning KW - Feature-based approaches KW - Traveling Salesperson Problem SN - 978-3-030-58111-4 T2 - Parallel Problem Solving from {Nature} (PPSN XVI) TI - Deep Learning as a Competitive Feature-Free Approach for Automated Algorithm Selection on the Traveling Salesperson Problem ER -