PUMA publications for /tag/Learning%20MachineWed Apr 22 18:03:14 CEST 2026New York, NY, USAProceedings of the 31th Annual ACM Symposium on User Interface Software and TechnologyUIST '18InfiniTouch: Finger-Aware Interaction on Fully Touch Sensitive Smartphones2018vis(us) learning machine interaction touchscreen finger-aware vis-sks vis visus:henzens visus:mayersn visus:lehy tiger-generation:639124705458569286 Wed Apr 22 18:03:13 CEST 2026New York, NY, USAProceedings of the 2018 CHI Conference on Human Factors in Computing Systems360:1-360:13CHI '18PalmTouch: Using the Palm As an Additional Input Modality on Commodity Smartphones2018vis-sks vis vis(us) capacitive machine learning smartphone image palm visus:lehy visus:baderpk visus:koschts visus:mayersn visus:henzens tiger-generation:639124705458569286 Wed Apr 22 18:03:13 CEST 2026New York, NY, USAProceedings of the 20th International Conference on Human-Computer Interaction with Mobile Devices and Services AdjunctMobileHCI '18Demonstrating PalmTouch: The Palm as An Additional Input Modality on Commodity Smartphones2018vis(us) palm smartphone image learning machine capacitive vis-sks vis visus:koschts visus:henzens visus:mayersn visus:lehy tiger-generation:639124705458569286 Mon Feb 09 11:45:09 CET 2026Real-Time Forecasting of Electricity Prices in Day-Ahead and Intraday-Auction Markets Using Statistical and Machine Learning Approaches2025Approaches Day-Ahead Electricity Forecasting Intraday-Auction Learning Machine Markets Prices Real-Time Statistical Using and in of Wed Jan 21 10:36:02 CET 2026International Conference on Artificial Intelligence and Soft Computing443--454Appropriate data density models in probabilistic machine learning approaches for data analysis2019data density learning machine probabilistic Tue Oct 07 14:17:51 CEST 2025Stahlbau10FlectoLine: eine adaptive Fassade für nachhaltige Architektur FlectoLine: A responsive Façade for sustainable architecture2025Aktuatorik Bionik Fassade Kunststoff Lernen Mechanismen Photovoltaik Zwilling actuation adaptive biomimetic compliant digital digitaler faserverstärkter façade fibre learning machine maschinelles mechanisms nachgiebige photo plastic pneumatic pneumatische reinforced responsive twin voltaic Angesichts des steigenden Energiebedarfs im Gebäudebetrieb gewinnt die Entwicklung adaptiver Fassadensysteme zunehmend an Bedeutung. Insbesondere solaraktive Gebäudehüllen bieten Potenzial, durch gezielte Steuerung von Sonneneinstrahlung, Belüftung und Tageslichtnutzung den Energieverbrauch zu senken und gleichzeitig den Nutzerkomfort zu erhöhen. Das Forschungsprojekt FlectoLine-Fassade verfolgt das Ziel, intelligente, materialbasierte adaptive Fassadenlösungen ohne klassische Antriebs- und Steuerungssysteme zu realisieren. Im Zentrum stehen dabei flexible Faltelemente, die auf bioinspirierten, nachgiebigen Mechanismen basieren und sich durch die elastische Verformbarkeit von FVK-Platten ohne Gelenke auszeichnen. Durch integrierte pneumatische Aktuatoren, präzise programmierte Materialsysteme und gezielte Differenzierung der Steifigkeiten kann eine kontrollierte Faltbewegung bis 90° erreicht werden. Neben einem Hybridmaterialsystem wurde eine thermoplastische Variante entwickelt, die einfacher zu fertigen und recycelbar ist. Umfangreiche Belastungstests belegen die Langlebigkeit beider Systeme. Ergänzend wurden flexible Photovoltaikzellen integriert, um Energie zu gewinnen. Ein digitaler Zwilling mit Sensorik und Wetterdaten steuert die Fassade in Echtzeit. Der Demonstrator mit 101 individuell verformbaren Elementen auf 83,5 m2 zeigt die technische Umsetzbarkeit und das energetische Potenzial adaptiver Fassadentechnologien der nächsten Generation. In view of the increasing energy demand in building operations, the development of adaptive façade systems is gaining growing importance. In particular, solar-active building envelopes offer potential to reduce energy consumption through targeted control of solar radiation, ventilation, and daylight use, while simultaneously increasing user comfort. The research project FlectoLine Façade aims to realize intelligent, material-based adaptive façade solutions without conventional drive and control systems. At its core are flexible folding elements based on bio-inspired, compliant mechanisms characterized by elastic deformability of fibre-reinforced composite (FRC) panels without joints. Through integrated pneumatic actuators, precisely programmed material systems, and targeted differentiation of stiffness, a controlled folding movement of up to 90° can be achieved. In addition to a hybrid material system, a thermoplastic variant was developed that is easier to manufacture and recyclable. Extensive load tests demonstrate the durability of both systems. Furthermore, flexible photovoltaic cells were integrated to generate energy. A digital twin with sensors and weather data controls the façade in real time. The demonstrator with 101 individually deformable elements covering 83.5 m2 demonstrates the technical feasibility and the energetic potential of next-generation adaptive façade technologies.Wed Sep 10 16:49:59 CEST 2025Optimal information injection and transfer mechanisms for active matter reservoir computing2025(cond-mat.soft), (cs.LG), (nlin.AO), (physics.comp-ph), Adaptation Computational Computer Condensed FOS: Learning Machine Matter Physical Physics Self-Organizing Soft Systems and information manybodyml myown publist sciences sciences, Wed Sep 10 16:49:59 CEST 2025Robustly optimal dynamics for active matter reservoir computing2025(cond-mat.soft), (cs.LG), (nlin.AO), (physics.comp-ph), Adaptation Computational Computer Condensed FOS: Learning Machine Matter Physical Physics Self-Organizing Soft Systems and information manybodyml myown publist sciences sciences, Fri Jan 31 11:03:06 CET 2025Physiological Reportsjan3Quantification of training‐induced alterations in body composition via automated machine learning analysis of MRI images in the thigh region: A pilot study in young females132025body composition deep imaging learning machine magnetic medicine musculoskeletal resonance sports system AbstractThe maintenance of an appropriate ratio of body fat to muscle mass is essentialfor the preservation of health and performance, as excessive body fat is associatedwith an increased risk of various diseases. Accurate body composition assessmentrequires precise segmentation of structures. In this study we developed a novelautomatic machine learning approach for volumetric segmentation and quantita-tive assessment of MRI volumes and investigated the efficacy of using a machinelearning algorithm to assess muscle, subcutaneous adipose tissue (SAT), and bonevolume of the thigh before and after a strength training. Eighteen healthy, young,female volunteers were randomly allocated to two groups: intervention group(IG) and control group (CG). The IG group followed an 8-week strength endur-ance training plan that was conducted two times per week. Before and after thetraining, the subjects of both groups underwent MRI scanning. The evaluation ofthe image data was performed by a machine learning system which is based ona 3D U-Net- based Convolutional Neural Network. The volumes of muscle, bone,and SAT were each examined using a 2 (GROUP [IG vs. CG]) × 2 (TIME [pre-intervention vs. post-intervention]) analysis of variance (ANOVA) with repeatedmeasures for the factor TIME. The results of the ANOVA demonstrate significantTIME × GROUP interaction effects for the muscle volume (F1,16 = 12.80, p = 0.003,ηP2 = 0.44) with an increase of 2.93% in the IG group and no change in the CG(−0.62%, p = 0.893). There were no significant changes in bone or SAT volumebetween the groups. This study supports the use of artificial intelligence systemsto analyze MRI images as a reliable tool for monitoring training responses onbody composition.Fri Nov 29 10:42:09 CET 2024New York, NY, USAProceedings of the 31th Annual ACM Symposium on User Interface Software and TechnologyUIST '18InfiniTouch: Finger-Aware Interaction on Fully Touch Sensitive Smartphones2018vis(us) interaction learning machine touchscreen finger-aware vis-sks vis visus:henzens visus:mayersn visus:lehy tiger-generation:638684700229105045 Fri Nov 29 10:42:07 CET 2024New York, NY, USAProceedings of the 2018 CHI Conference on Human Factors in Computing Systems360:1-360:13CHI '18PalmTouch: Using the Palm As an Additional Input Modality on Commodity Smartphones2018vis-sks vis vis(us) smartphone image machine learning palm capacitive visus:lehy visus:baderpk visus:koschts visus:mayersn visus:henzens tiger-generation:638684700229105045 Fri Nov 29 10:42:04 CET 2024New York, NY, USAProceedings of the 20th International Conference on Human-Computer Interaction with Mobile Devices and Services AdjunctMobileHCI '18Demonstrating PalmTouch: The Palm as An Additional Input Modality on Commodity Smartphones2018vis(us) palm smartphone image learning machine capacitive vis-sks vis visus:henzens visus:mayersn visus:koschts visus:lehy tiger-generation:638684700229105045 Fri Jul 26 11:59:23 CEST 20242024 European Control Conference (ECC)June2441-2447Learning Compensation of the State-Dependent Transmission Errors in Rack-and-Pinion Drives2024compensation drive error feed isw learning machine tool Rack-and-pinion drives are commonly used in large machine tools to provide linear motion of heavy loads over long travel distances. A key concern in this context is the achievable path accuracy, which is limited by assembly and manufacturing tolerances of the gearing components in conjunction with load-dependent deformation and the inherent backlash of the system. To address this issue, this paper presents a method for robust modeling of the individual and state-dependent transmission errors of a drive utilizing a two-stage machine learning approach. Based on this, the position control is extended to include an error compensation, which suppresses the modeled deviations in the mechanical system including the position-dependent backlash. The achievable increase in path accuracy as well as the robustness of the approach are evaluated and quantified by an experimental validation on a system with industry standard components.Mon Feb 19 14:41:03 CET 2024Proceedings of the 37th Conference on Neural Information Processing Systems (NeurIPS 2023)LVM-Med: Learning Large-Scale Self-Supervised Vision Models for Medical Imaging via Second-order Graph Matching2023graph; learning machine Fri Feb 02 08:07:58 CET 2024Procedia CIRP216–221Tool condition monitoring in drilling processes using anomaly detection approaches based on control internal data1212024analysis anomaly condition detection learning machine machining manufacturing monitoring networks neural process series time tool Tue Aug 09 08:52:30 CEST 2022{CIRP} Annals1345--348Adaptive compensation of the transmission errors in rack-and-pinion drives712022Compensation Feed Learning Machine drive tool Thu Jan 20 14:39:51 CET 2022CAADRIA 2020: Re:Anthropocene - Design in the Age of Humans203-212Post-Anthropocene: The Design after the Human Centered Design Age22020approach architetural biodiversity learning machine morre-than-human myown performance post-Anthropocene systemic to urban Wed Jan 19 17:25:49 CET 2022133-142Spiralling Slope as a Real Life Co-Design Laboratory2018biodiversity codesign ecosystem gigamapping learning life machine myown performance The paper and its presentation is to discuss a family house Spiraling Slope (sophia) that is co-designed, inhabited, tested and developed prototype by the second and third author of this submission, the clients. The eco-systemically performing house, literally twisted as a helix into the sloping terrain, gaining its thermal energy, is also covered by extensive greenery to gain this property on its top. Algae, grown on the glass roof, is to moderate its atrium clime. Through its sloping disposition, the house employs natural ventilation for its airing. Though the first author is conducting research that this performance is operated by nature of material properties (Davidová 2016c), in the time of the house’s design stage, this research was not developed enough to meet the building practice. Therefore, the house’s eco-systemic performance that could not have been reached by biology is achieved through the technology of autonomous environment control (sysloop). Sysloop is a real-time knowledge processing software cowering physical computing, where the clients are the main developer, co-designing with all the other professions involved in the house design and construction, including its architects and systemic designer. Since the house’s design is based on natural performance, both its environmental, social, cultural and practical performance is operated through a computer based system AI that relates to BIG Data, the paper therefore presents one of the first attempts of fusion of abiotic and biotic agency with artificial intelligence in architectural practice. Testing such prototype by life co-living experience brings true insights into its design in time. This approach has been defined by Sevaldson as Time-Based Design at the start of this millennium (Sevaldson 2004; Sevaldson 2005). However, at that time the crucial leading design team member was not at the same time the subject of testing. This brings the Shön’s discussion on ‘reflective practitioner’ (Schön 1983) few steps further. It is not the case when her/his designing and lecturing is enriched by tacit knowledge of i.e. building practice experience, but furthermore, the co-designer’s experience is gained through living within the system s/he is co-designing and co-prototyping in real life and for real life. Therefore, through such case studies, as the approach fuses the life performance with its design and eco-systemic design and living processes, the first author defends to ratify a new design field, Systemic Approach to Architectural Performance that fights for the shift from Anthropocene.Wed Jan 19 17:21:39 CET 2022132-141sysloop2018architecture codesign intelligent learning machine myown responsive systems Unlike preceding “autonomous house systems” sysloop is cross-layered and highly scalable concept of “allopoietic system”, a system that is autonomous though dependent on the exchange across its environment (Dekkers 2015). This is performed through three types of co-design: • co-designing of trans-disciplinary co-authors; • co-designing of environment from which it is learning, users included; • co-designing of artificial intelligence and big data. At the scale of local environment, sysloop is focused mainly on interrelations of individual life space qualities, providing contextual autonomous behaviour across many aspects such as climate, light, sound, smell, safety, access control, etc. Due to such scope, the trans-disciplinary team of experts developing sysloop technology is evolving in time in reference to related fields. We specify key aspects of an alternative information system with ability to make decisions based on automated interpretation of meanings, instead of (conventional) symbol processing. We verify such information system in practice of environment automation, introducing technological support of overlapping values such as information hygiene, lifelong learning, aesthetics, overall comfort, etc. Such environments are integrated at “buildings” units scale in phenomenological terms and at “industrial” units scale focused on adaptive automation and reliability engineering, both processing micro-sensorial data and performing qualified decision making in real-time. These together with other big data available are integrated to support the “cities’” scale layer. This layer is to serve for informed city planning and emergency situations solutions, including automated, personalised assistance to individual citizens, etc. This multi-scaled system is feedback looping across its layers of scales and types of co-design and thus evolving by data and most importantly, its ever-changing relations. It gives to the term “smart buildings” its meaning across the scales towards sustainable development, performance and ecosystems. The authors, among all the team, built the first prototypical family and office building for real-world testing and further development. This “real life co-design laboratory” is elaborated at separate paper for this conference.Sun Aug 08 00:02:03 CEST 2021MICCAI 2021cite arxiv:2106.11048Comment: Accepted at MICCAI 2021CataNet: Predicting remaining cataract surgery duration2021cataract duration learning machine prediction surgery Cataract surgery is a sight saving surgery that is performed over 10 million times each year around the world. With such a large demand, the ability to organize surgical wards and operating rooms efficiently is critical to delivery this therapy in routine clinical care. In this context, estimating the remaining surgical duration (RSD) during procedures is one way to help streamline patient throughput and workflows. To this end, we propose CataNet, a method for cataract surgeries that predicts in real time the RSD jointly with two influential elements: the surgeon's experience, and the current phase of the surgery. We compare CataNet to state-of-the-art RSD estimation methods, showing that it outperforms them even when phase and experience are not considered. We investigate this improvement and show that a significant contributor is the way we integrate the elapsed time into CataNet's feature extractor.Learning MachineCommunity for tag(s) Learning Machine