Logo des Repositoriums

Auflistung nach Schlagwort "neural networks"

1 - 10 von 12
  • Textdokument
    Building a GAN for Replicating Epithelial Impedance Spectra for ML-based Pattern Recognition
    (SKILL 2021, 2021) Jurkschat, Lena; Schindler, Benjamin
    Impedance spectroscopy is a common method in the field of biotechnology to measure electrical conductivity of special cell lines (i.e. ephitelial). Based on the measured impedance spectra, machine learning (ML) techniques including random forests and feedforward networks are increasingly used to determine physiological properties of the underlying cell tissue and to detect a wide range of diseases. However, training ML models for this purpose typically requires large amounts of data and real cell tissue measurements are costly to obtain due to their experimental setup. This paper introduces a Generative Adversarial Network (GAN) which meets the high demand for training data by replicating impedance spectra from a given data set. As a proof of concept, we show that GANs are capable of generating spectra that have a similar shape to the original ones and could therefore be used to overcome a lack of training data.
  • Textdokument
    Challenges of Network Traffic Classification Using Deep Learning in Virtual Networks
    (INFORMATIK 2022, 2022) Spiekermann,Daniel; Keller,Jörg
    The increasing number of network-based attacks like denial-of-service and ransomware have become a serious threat in nowadays digital infrastructures. Therefore, the monitoring of network communications and the classification of network packets is a critical process when protecting the environment. Modern techniques like deep learning aim to help the providers when detecting anomalies or attacks by learning details extracted from a network packet or a flow of packets. Most of these models are trained in networks without any kind of virtualisation, especially network virtualisation overlay environments are not investigated in detail. In this paper, we analyse the classification rate of a Convolutional Neural Network (CNN) faced with encapsulated packets. We evaluate this approach with a proof-of-concept based on a binary classification of a self-curated data-set.
  • Textdokument
    Deep Quality-informed Score Normalization for Privacy-friendly Speaker Recognition in unconstrained Environments
    (BIOSIG 2017, 2017) Nautsch,Andreas; Steen,Søren Trads; Busch,Christoph
    In scenarios that are ambitious to protect sensitive data in compliance with privacy regulations, conventional score normalization utilizing large proportions of speaker cohort data is not feasible for existing technology, since the entire cohort data would need to be stored on each mobile device. Hence, in this work we motivate score normalization utilizing deep neural networks. Considering unconstrained environments, a quality-informed scheme is proposed, normalizing scores depending on sample quality estimates in terms of completeness and signal degradation by noise. Utilizing the conventional PLDA score, comparison i-vectors, and corresponding quality vectors, we aim at mimicking cohort based score normalization optimizing the Cmin llr discrimination criterion. Examining the I4U data sets for the 2012 NIST SRE, an 8.7% relative gain is yielded in a pooled 55-condition scenario with a corresponding condition-averaged relative gain of 6.2% in terms of Cmin llr . Robustness analyses towards sensitivity regarding unseen conditions are conducted, i.e. when conditions comprising lower quality samples are not available during training.
  • Workshopbeitrag
    Detecting Hands from Piano MIDI Data
    (Mensch und Computer 2019 - Workshopband, 2019) Hadjakos, Aristotelis; Waloschek, Simon; Leemhuis, Alexander
    When a pianist is playing on a MIDI keyboard, the computer does not know with which hand a key was pressed. With the help of a Recurrent Neural Network (RNN), we assign played MIDI notes to one of the two hands. We compare our new approach with an existing heuristic algorithm and show that RNNs perform better. The solution is real-time capable and can be used via OSC from any programming environment. A non real-time capable variant provides slightly higher accuracy. Our solution can be used in music notation software to assign the left or right hand to the upper or lower staff automatically. Another application is live playing, where different synthesizer sounds can be mapped to the left and right hand.
  • Konferenzbeitrag
    Inductive Learning of Concept Representations from Library-Scale Bibliographic Corpora
    (INFORMATIK 2019: 50 Jahre Gesellschaft für Informatik – Informatik für Gesellschaft, 2019) Galke, Lukas; Melnychuk, Tetyana; Seidlmayer, Eva; Trog, Steffen; Förstner, Konrad U.; Schultz, Carsten; Tochtermann, Klaus
    Automated research analyses are becoming more and more important as the volume of research items grows at an increasing pace. We pursue a new direction for the analysis of research dynamics with graph neural networks. So far, graph neural networks have only been applied to small-scale datasets and primarily supervised tasks such as node classification. We propose to use an unsupervised training objective for concept representation learning that is tailored towards bibliographic data with millions of research papers and thousands of concepts from a controlled vocabulary. We have evaluated the learned representations in clustering and classification downstream tasks. Furthermore, we have conducted nearest concept queries in the representation space. Our results show that the representations learned by graph convolution with our training objective are comparable to the ones learned by the DeepWalk algorithm. Our findings suggest that concept embeddings can be solely derived from the text of associated documents without using a lookup-table embedding. Thus, graph neural networks can operate on arbitrary document collections without re-training. This property makes graph neural networks useful for the analysis of research dynamics, which is often conducted on time-based snapshots of bibliographic data.
  • Textdokument
    The Influence of Training Parameters on Neural Networks' Vulnerability to Membership Inference Attacks
    (INFORMATIK 2022, 2022) Bouanani,Oussama; Boenisch,Franziska
    With Machine Learning (ML) models being increasingly applied in sensitive domains, the related privacy concerns are rising. Neural networks (NN) are vulnerable to, so-called, membership inference attacks (MIA) which aim at determining whether a particular data sample was used for training the model. The factors that render NNs prone to this privacy attack are not yet fully understood. However, previous work suggests that the setup of the models and the training process might impact a model's risk to MIAs. To investigate these factors more in detail, we set out to experimentally evaluate the influence of the training choices in NNs on the models' vulnerability. Our analyses highlight that the batch size, the activation function, and the application and placement of batch normalization and dropout have the highest impact on the success of MIAs. Additionally, we applied statistical analyses to the experiment results and found a highly positive correlation between a model's ability to resist MIAs and its generalization capacity. We also defined a metric to measure the difference in the distributions of loss values between member and non-member data samples and observed that models scoring higher values on that metric were consistently more exposed to the attack. The latter observation was further confirmed by manually generating predictions for member and non-member samples producing loss values within specific distributions and launching MIAs on them.
  • Konferenzbeitrag
    Intelligent System for Computer-assisted Clinical Cancer Image Analysis
    (Information systems technology and its applications – 6th international conference – ISTA 2007, 2007) Bondarenko, Anatoly N.; Katsuk, Andrei V.
    We present CaDiS - a new multimedia medical workstation, which helps early and precise diagnosis and treatment of cervical cancer. The workstation is developed with the close participation of medical staff of Novosibirsk Clinical Center. CaDiS is an advanced tool for medical image recognition, designed on the basis of new algorithms for medical image analysis. It is under intensive development and will include history of illness, analysis and prediction, based on perceptual and conceptual semantics. The detection accuracy of the proposed system has reached to 91%, providing thus an indication that such intelligent schemes could be used as a supplementary diagnostic tool in cervical cancer detection.
  • Zeitschriftenartikel
    MoneyBee: Aktienkursprognose mit künstlicher intelligenz bei hoher rechenleistung
    (Wirtschaftsinformatik: Vol. 45, No. 3, 2003) Bohn, Andreas; Güting, Thomas; Mansmann, Till; Selle, Stefan
    The company i42 GmbH, Mannheim, developed MoneyBee: a system to predict stock market values, basing on artificial intelligence (neural networks), distributed computing and different applications to optimize the input- and output-data (e.g. genetic algorithms, statistical methods). More than 200 values (especially from German stock market) are observed by this system continuously, with daily updated predictions. The information technology product is an innovation — not by its basic technology, but by its cooperation of different program groups on high level.
  • Textdokument
    Towards probabilistic multiclass classification of gamma-ray sources
    (INFORMATIK 2022, 2022) Malyshev,Dmitry; Bhat,Aakash
    Machine learning algorithms have been used to determine probabilistic classifications of unassociated sources. Often classification into two large classes, such as Galactic and extra-galactic, is considered. However, there are many more physical classes of sources (23 classes in the latest Fermi-LAT 4FGL-DR3 catalog). In this note we subdivide one of the large classes into two subclasses in view of a more general multi-class classification of gamma-ray sources. Each of the three large classes still encompasses several of the physical classes. We compare the performance of classifications into two and three classes. We calculate the receiver operating characteristic curves for two-class classification, where in case of three classes we sum the probabilities of the sub-classes in order to obtain the class probabilities for the two large classes. We also compare precision, recall, and reliability diagrams in the two- and three-class cases.
  • Workshopbeitrag
    UX Design Pattern für Datenschutz und Vertrauen
    (Mensch und Computer 2022 - Workshopband, 2022) Hoth, Veronica; Ivanova, Maria; Brandenburg, Stefan
    Cookie-Banner sind für die meisten Websites aufgrund der Allgemeinen Datenschutzverordnung (DSGVO) der EU obligatorisch. In einer Studie mit 52 Teilnehmenden wurde untersucht, ob persuasive UI-Elemente wie Dark und White Pattern die User Experience und das Vertrauen der User beeinflussen. Die verwendeten Cookie- Banner hatten entweder keine persuasiven UI Elemente (N), zwei unterschiedlich starke Dark Pattern (Dp1 und Dp2) oder ein White Pattern (Wp). Das starke Dark Pattern (Dp2) verringerte die User Experience- und Usability-Bewertung der Nutzenden, im Vergleich zum neutralen Design (N). Cookie Banner mit leichten Dark Pattern (Dp1) führten zu höheren positiven Emotionen als das neutrale Design (N). Wenn die Affinität zur Technologie der Teilnehmenden berücksichtigt wurde, dann erhöhte sich das wahrgenommene Vertrauen bei der Interaktion mit starkem Dark Pattern (Dp2) im Vergleich zum neutralen Design (N). Die Ergebnisse implizieren, dass User das manipulative Design nicht unbedingt als solches wahrnehmen. Dies wirft die Frage auf, ob es ethisch vertretbar ist, User durch Interface Design zum Zweck der Datengewinnung zu manipulieren.