Logo des Repositoriums

Auflistung nach Schlagwort "machine learning"

1 - 10 von 66
  • Konferenzbeitrag
    Achiever or explorer? gamifying the creation process of training data for machine learning
    (Mensch und Computer 2020 - Tagungsband, 2020) Alaghbari, Sarah; Mitschick, Annett; Blichmann, Gregor; Voigt, Martin; Dachselt, Raimund
    The development of artificial intelligence, e. g., for Computer Vision, through supervised learning requires the input of large amounts of annotated or labeled data objects as training data. The creation of high-quality training data is usually done manually which can be repetitive and tiring. Gamification, the use of game elements in a non-game context, is one method to make tedious tasks more interesting. This paper proposes a multi-step process for gamifying the manual creation of training data for machine learning purposes. We choose a user-adapted approach based on the results of a preceding user study with the target group (employees of an AI software development company) which helped us to identify annotation use cases and the users' player characteristics. The resulting concept includes levels of increasing difficulty, tutorials, progress indicators and a narrative built around a robot character which at the same time is a user assistant. The implemented prototype is an extension of the company’s existing annotation tool and serves as a basis for further observations.
  • Konferenzbeitrag
    Adaptive real-time crop row detection through enhancing a traditional computer vision approach
    (44. GIL - Jahrestagung, Biodiversität fördern durch digitale Landwirtschaft, 2024) Hussaini, Mortesa; Voigt, Max; Stein, Anthony
    Crop row detection is important to enable precise management of fields and optimize the use of resources such as fertilizers and water. Autonomous machines need an effective but also robust real-time row detection system to be able to adapt to different field conditions. In this paper, we present an enhanced crop row detection approach which integrates traditional computer vision methods with further techniques such as k-means clustering or probabilistic Hough transformation. The resulting hybrid method allows for efficient and robust detection of straight and curved crop rows in image and video material. We validate our approach empirically on the crop row benchmark dataset (CRBD) and compare it with other state-of-the-art approaches. Furthermore, we demonstrate that our approach is designed to be adaptive and thus becomes straightforwardly transferable to other experimental setups. To corroborate that, we report on results when our approach is validated on representative corner cases which have been collected in the scope of a research project. Observations and current limitations of our approach are discussed along with possible solutions to overcome them in future work.
  • Textdokument
    Aggregate-based Training Phase for ML-based Cardinality Estimation
    (BTW 2021, 2021) Woltmann, Lucas; Hartmann, Claudio; Habich, Dirk; Lehner, Wolfgang
    Cardinality estimation is a fundamental task in database query processing and optimization. As shown in recent papers, machine learning (ML)-based approaches may deliver more accurate cardinality estimations than traditional approaches. However, a lot of training queries have to be executed during the model training phase to learn a data-dependent ML model making it very time-consuming. Many of those training or example queries use the same base data, have the same query structure, and only differ in their selective predicates. To speed up the model training phase, our core idea is to determine a predicate-independent pre-aggregation of the base data and to execute the example queries over this pre-aggregated data. Based on this idea, we present a specific aggregate-based training phase for ML-based cardinality estimation approaches in this paper. As we are going to show with different workloads in our evaluation, we are able to achieve an average speedup of 63 with our aggregate-based training phase and thus outperform indexes.
  • Konferenzbeitrag
    An Anthropomorphic Approach to establish an Additional Layer of Trustworthiness of an AI Pilot
    (Software Engineering 2022 Workshops, 2022) Regli, Christoph; Annighoefer, Björn
    AI algorithms promise solutions for situations where conventional, rule-based algorithms reach their limits. They perform in complex problems yet unknown at design time, and highly efficient functions can be implemented without having to develop a precise algorithm for the problem at hand. Well-tried applications show the AI’s ability to learn from new data, extrapolate on unseen data, and adapt to a changing environment — a situation encountered in fl ight operations. In aviation, however, certifi cation regulations impede the implementation of non-deterministic or probabilistic algorithms that adapt their behaviour with increasing experience. Regulatory initiatives aim at defining new development standards in a bottom-up approach, where the suitability and the integrity of the training data shall be addressed during the development process, increasing trustworthiness in eff ect. Methods to establish explainability and traceability of decisions made by AI algorithms are still under development, intending to reach the required level of trustworthiness. This paper outlines an approach to an independent, anthropomorphic software assurance for AI/ML systems as an additional layer of trustworthiness, encompassing top-down black-box testing while relying on a well-established regulatory framework.
  • Konferenzbeitrag
    Assisting Service Robots on their Journey to become Autonomous Agents: From Apprentice to Master by Participatory Observation
    (Mensch und Computer 2019 - Tagungsband, 2019) Golchinfar, David; Vaziri, Daryoush; Stevens, Gunnar; Schreiber, Dirk
    Natural and reliable application of service robots (SR) in service domains, for instance health service or elderly care, is currently not possible and full autonomy and automatization of SR is still in far distance. Hence, methodologies are needed that promote human-robot collaboration and allow the robot to learn from its human mentor to become more autonomous and reliable. This demo illustrates an environment for such human-robot collaboration that provides an infrastructure for SR manipulation and teaching. The basic idea is that the robot becomes an apprentice that learns new skills by observing a trained human mentor that performs relevant tasks in the service domain by operating the robot. By observation and collaboration, the SR gradually becomes more autonomous and capable to carry out relevant healthcare tasks.
  • muc: langbeitrag (vorträge)
    Automatic Classification of Mobile Phone s Contacts
    (Mensch & Computer 2013: Interaktive Vielfalt, 2013) Sahami Shirazi, Alireza; Le, Huy Viet; Henze, Niels; Schmidt, Albrecht
    Current smartphones have virtually unlimited space to store contact information. Users typically have dozens or even hundreds of contacts in their address book. The number of contacts can make it difficult to find particular contacts from the linear list provided by current phones. Grouping contacts ease the retrieval of particular contacts and also enables to share content with specific groups. Previous work, however, shows that users are not willing to manually categorize their contacts. In this paper we inves-tigate the automatic classification of contacts in phones contact lists, using the user s communication history. Potential contact groups were determined in an online survey with 82 participants. We collect-ed the call and SMS communication history from 20 additional participants. Using the collected data we trained a machine-learning algorithm that correctly classified 59.2% of the contacts. In a pilot study in which we asked participants to review the results of the classifier we found that 73.6% of the re-viewed contacts were considered correctly classified. We provide directions to further improve the performance and argue that the current results already enable to ease the manual classification of mo-bile phone contacts.
  • Konferenzbeitrag
    Can algorithms help us manage dairy cows?
    (41. GIL-Jahrestagung, Informations- und Kommunikationstechnologie in kritischen Zeiten, 2021) Cockburn, Marianne
    Digitalisation has reached agricultural production and specifically dairy farming, where a wide range of sensing technologies are now available. From farm management systems over body condition scoring systems to those that detect behavioural changes. All these systems have one aim: to offer decision support to the farmer and aid his management decisions. Currently, however, little is known about the return of investment that these systems offer, or even the effectiveness of their functionality. Only little information is available about the underlying algorithms, despite them presenting the essence of performance. Thus, we can only consider the published literature to get an impression of such systems’ outcome. In the current study, we therefore evaluated machine-learning related studies published in the scientific literature between 2015 and 2020. We found that machine-learning algorithms were implemented across all fields of dairy science, but only a minority of them could reliably aid management decisions in practice. In this publication, we aim to give an overview of the achievements of current machine-learning algorithms published in dairy science literature and give an outlook on how they could develop further in the future.
  • Konferenzbeitrag
    Challenges in Data Preservation for AI and ML Systems
    (INFORMATIK 2024, 2024) Tonkin, Emma L.; Tourte, Gregory J. L.
    The management and preservation of machine learning (ML) and artificial intelligence (AI) data is increasingly a concern for research institutions, as well as for institutions and industry organisations making use of this type of data and method. This paper summarises key issues in this area, presenting the case that there are significant benefits to the industry in developing best practices and joint standards in this area, and identifying the benefits of this approach, as well as highlighting risks and a current paucity of best practice in the area.
  • Konferenzbeitrag
    Code Smell Detection using Features from Version History
    (SE 2024 - Companion, 2024) Engeln, Ulrike
    Code smells are indicators of bad quality in software. There exist several detection techniques for smells, which mainly base on static properties of the source code. Those detectors usually show weak performance in detection of context-sensitive smells since static properties hardly capture information about relations in the code. To address this information gap, we propose a strategy to extract information about interdependencies from version history. We use static and the new historical features to identify code smells by a random forest. Experiments show that the introduced historical features improve detection of code smells that focus on interdependencies.
  • Konferenzbeitrag
    Code Smell Detection using Features from Version History
    (Softwaretechnik-Trends Band 43, Heft 2, 2023) Engeln, Ulrike
    Code smells are indicators for bad quality of source code. A well suited approach for the development of a smell detector are machine learning techniques that learn based on features, i.e., measurable properties of the software under investigation, e.g., code metrics. One major objective of our machine learning approach is to decide how to express information from the version history by features. we introduce a method to draw historical features that improve smell detection.