Program

Friday, March 25 9:00 - 9:15

S0: Welcome and Introduction

Chairs: Pascal Hirmer (University of Stuttgart, Germany), Jadwiga Indulska (The University of Queensland, Australia), Gabriele Civitarese (University of Milan, Italy)

Friday, March 25 09:15 - 10:00

S1: Keynote: Machine Learning on the Edge - A Software Engineering Perspective

Chair: Jadwiga Indulska

Philippe Lalanda (Grenoble University, France)

Abstract: Pervasive services are often located in cloud data centers . This implies constant and costly communications between connected objects and data centers that are usually located at a great distance. This architectural solution does not scale well and is challenged by the massive increase in data to be transported and stored in the cloud. Also, recent trends in pervasive computing, like augmented reality and machine learning for instance, increase the number of applications that require fast data processing. Analysing this data in the cloud does not meet the response time requirements of these applications due to the high communication latency between the devices and the cloud infrastructure. This major evolution has marked the end of all-cloud solutions and the inevitable emergence of more decentralized architectures where computations are done near the data sources. This new architectural approach is based on edge computing, which refers to all the enabling technologies allowing computation to be performed at the edge of the network. Another major evolution relates to the use of Artificial Intelligence. Machine Learning based approaches have indeed received a very strong interest for the development of pervasive applications. This can be explained by the difficulty to build models of dynamic phenomena in complex physical environments. The goal of a machine learning system is to train an algorithm to automatically make decisions by identifying patterns that may be hidden within massive data sets whose exact nature is unknown and therefore cannot be programmed explicitly. The growing attention towards machine learning stems from different sources: efficient algorithms, availability of massive amounts of data in pervasive environments, advances in highperformance computing, broad accessibility of these technologies, and impressive successes reported by industry, academia, and research communities, in fields such as vision, natural language processing or decision making. The development of software systems based on learning techniques is however very different than the development of traditional software systems since the behavior of the system is not specified in the code by programmers but is learned by a machine from data. This has profound implications for the development of pervasive applications. First, it demands the creation of new development teams with different and complementary skills, using different languages, processes and tools. Such heterogeneity is difficult to manage from a human point of view but also from a technical point of view. Strong heterogeneity is also introduced in the management of the software components life cycle. Learning-based components have their own cycle, entirely data-driven, that is very different from the one of more traditional software components. While the major activities remain the same, they are treated in a very different way and require, here again, specific skills and tools. All these differences call into question the current Software Engineering techniques and processes but also the support that has to be provided by the edge platforms.

Friday, March 25 10:00 - 11:00

Paper Session 1: Human Activity Recognition

Chair: Jadwiga Indulska

Junaid Younas; Hector Margarito; Paul Lukowicz: FAirWrite - Movement Reconstruction and Recognition using a Low-cost IMU

Riccardo Presotto; Gabriele Civitarese; Claudio Bettini: Preliminary Results on Sensitive Data Leakage in Federated Human Activity Recognition

Friday, March 25 11:30 - 13:00

Paper Session 2: Learning Based Approaches and Context Models

Chair: Gabriele Civitarese

Anastasiia Usmanova; François Portet; Philippe Lalanda; German Vega: Federated Continual Learning through distillation in pervasive computing

Lea Cohausz; Nils Wilken; Heiner Stuckenschmidt: Plan-Similarity Based Heuristics for Goal Recognition

Daniel Del Gaudio; Boshra Ariguib; Arne Bartenbach; Georgios Solakis: A live context model for semantic reasoning in IoT applications

Friday, March 25 14:00 - 15:30

Paper Session 3: Human Activity Recognition (2)

Chair: Pascal Hirmer

Lahiru Wijayasingha; John Stankovic: Generalized Few-Shot Learning For Wearable Sensor-based Human Activity Recognition

Abduallah Mohamed; Fernando Lejarza; Stephanie Cahail; Christian Claudel; Edison Thomaz: HAR-GCNN: Deep Graph CNNs for Human Activity Recognition From Highly Unlabeled Mobile Sensor Data

Soroush Shahi; Rawan Alharbi; Yang Gao; Sougata Sen; Aggelos K Katsaggelos; Josiah Hester; Nabil Alshurafa: Impacts of Image Obfuscation on Fine-grained Activity Recognition in Egocentric Video

Friday, March 27 15:30 - 15:45

S5: Discussion, Feedback and Farewell