检索结果-内蒙古大学图书馆

Channel Dependence, Limited Lookback Windows, and the Simplicity of datasets: How Biased is Time Series Forecasting?

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Abdelmalak, Ibram Madhusudhanan, Kiran Choi, Jungmin Stubbemann, Maximilian Schmidt-Thieme, Lars Information Science and Machine Learning Lab VWFS Data Analytics Research Center University of Hildesheim Niedersachsen Hildesheim Germany

Time-series forecasting research has converged to a small set of datasets and a standardized collection of evaluation scenarios. Such a standardization is to a specific extent needed for comparable research. However, the underlying assumption is, that the considered setting is a representative for the problem as a whole. In this paper, we challenge this assumption and show that the current scenario gives a strongly biased perspective on the state of time-series forecasting research. To be more detailed, we show that the current evaluation scenario is heavily biased by the simplicity of the current datasets. We furthermore emphasize, that when the lookback-window is properly tuned, current models usually do not need any information flow across channels. However, when using more complex benchmark data, the situation changes: Here, modeling channel-interactions in a sophisticated manner indeed enhances performances. Furthermore, in this complex evaluation scenario, Crossformer, a method regularly neglected as an important baseline, is the SOTA method for time series forecasting. Based on this, we present the Fast Channel-dependent Transformer (FaCT), a simplified version of Crossformer which closes the runtime gap between Crossformer and TimeMixer, leading to an efficient model for complex forecasting datasets. © 2025, CC BY.

关键词：

Robust and Efficient Writer-Independent IMU-Based Handwriting Recognization

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Li, Jindong Hamann, Tim Barth, Jens Kaempf, Peter Zanca, Dario Eskofier, Bjoern Machine Learning and Data Analytics Lab Friedrich-Alexander-Universität Erlangen-Nürnberg Germany STABILO International GmbH Germany

Online handwriting recognition (HWR) using data from inertial measurement units (IMUs) remains challenging due to variations in writing styles and the limited availability of high-quality annotated datasets. Traditional models often struggle to recognize handwriting from unseen writers, making writer-independent (WI) recognition a crucial but difficult problem. This paper presents an HWR model with an encoder-decoder structure for IMU data, featuring a CNN-based encoder for feature extraction and a BiLSTM decoder for sequence modeling, which supports inputs of varying lengths. Our approach demonstrates strong robustness and data efficiency, outperforming existing methods on WI datasets, including the WI split of the OnHW dataset and our own dataset. Extensive evaluations show that our model maintains high accuracy across different age groups and writing conditions while effectively learning from limited data. Through comprehensive ablation studies, we analyze key design choices, achieving a balance between accuracy and efficiency. These findings contribute to the development of more adaptable and scalable HWR systems for real-world applications. The code is available at https://***/jindongli24/REWI. Copyright © 2025, The Authors. All rights reserved.

关键词： Units of measurement

Sublinear Variational Optimization of Gaussian Mixture Models with Millions to Billions of Parameters

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Salwig, Sebastian Kahlke, Till Hirschberger, Florian Forster, Dennis Lücke, Jörg Carl von Ossietzky University Oldenburg Faculty VI Machine Learning Oldenburg Germany Frankfurt University of Applied Sciences Faculty 3 Data Analytics and AI Frankfurt am Main Germany

Gaussian Mixture Models (GMMs) range among the most frequently used machine learning models. However, training large, general GMMs becomes computationally prohibitive for datasets with many data points N of high-dimensionality D. For GMMs with arbitrary covariances, we here derive a highly efficient variational approximation, which is integrated with mixtures of factor analyzers (MFAs). For GMMs with C components, our proposed algorithm significantly reduces runtime complexity per iteration from O(NCD2) to a complexity scaling linearly with D and remaining constant w.r.t. C. Numerical validation of this theoretical complexity reduction then shows the following: the distance evaluations required for the entire GMM optimization process scale sublinearly with NC. On large-scale benchmarks, this sublinearity results in speed-ups of an order-of-magnitude compared to the state-of-the-art. As a proof of concept, we train GMMs with over 10 billion parameters on about 100 million images, and observe training times of approximately nine hours on a single state-of-the-art CPU. Copyright © 2025, The Authors. All rights reserved.

关键词： Expectation maximization algorithm

Noise filtering approach to improve handwritten digit recognition using customized CNN for Cerebral Palsy individuals

学校读者我要写书评

暂无评论

The European Physical Journal Special Topics 2025年 1-19页

作者： Muthureka, K. Srinivasulu Reddy, U. Janet, B. Machine Learning and Data Analytics Lab Department of Computer Applications National Institute of Technology-Tiruchirappalli Tiruchirappalli India Centre of Excellence (CoE) in Artificial Intelligence National Institute of Technology-Tiruchirappalli Tiruchirappalli India Information Processing Lab Department of Computer Applications Tiruchirappalli India

Automatic recognition of handwritten digits in people with Cerebral Palsy (CP) is a serious challenge that requires advancements in data preparation for improved predictive accuracy. The disorganized digit pixel patterns caused by CP people’ innate motor coordination problems result in the classification errors due to the noise and outlier. Despite the fact that many handwritten digit databases for normal people are available, none expressly cater to those with shaky hands/CP dataset. This highlights the importance of an efficient pre-processing method for mitigating variances in CP individuals’ digit pixel patterns. A total of 43,000 images from 43 CP people were collected and organized into 10 different folders for data labeling. The authors used the K-Means Based Noise Removal (KMNR) approach to remove noise and outliers from a CP and tested it on a customized CNN. The KMNR model achieved an accuracy of 98.86%, compared to 93.54% without it. The F1-score values for state-of-the-art models without KMNR ranged from 85.00% to 97.00%, but after integrating KMNR, they increased to 92.00% to 98.00%. The precision values ranged from 85.00% to 97.00% in the absence of KMNR, but increased to 91.00% to 98.00% in the presence of KMNR. The recall values ranged from 85.00% to 97.00% without KMNR, but increased to 87.00% to 98.00% with KMNR.

关键词：

CROSS-TECHNOLOGY GENERALIZATION IN SYNTHESIZED SPEECH DETECTION: EVALUATING AST MODELS WITH MODERN VOICE GENERATORS

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Ustinov, Andrew Yordanov, Matey Kuchma, Andrei Bychkov, Mikhail Department of Machine Learning and Data Analytics ITMO University Russia AI Innovation Hub ANTEI Limited Hong Kong Department of Integrative Systems and Design Hong Kong University of Science and Technology AI Innovation Hub ANTEI Limited Hong Kong

This paper evaluates the Audio Spectrogram Transformer (AST) architecture for synthesized speech detection, with focus on generalization across modern voice generation technologies. Using differentiated augmentation strategies, the model achieves 0.91% EER overall when tested against ElevenLabs, NotebookLM, and Minimax AI voice generators. Notably, after training with only 102 samples from a single technology, the model demonstrates strong cross-technology generalization, achieving 3.3% EER on completely unseen voice generators. This work establishes benchmarks for rapid adaptation to emerging synthesis technologies and provides evidence that transformer-based architectures can identify common artifacts across different neural voice synthesis methods, contributing to more robust speech verification systems. © 2025, CC BY-NC-ND.

关键词： Spectrographs

DISCO: Internal Evaluation of Density-Based Clustering

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Beer, Anna Krieger, Lena Weber, Pascal Ritzert, Martin Assent, Ira Plant, Claudia Faculty of Computer Science University of Vienna Vienna Austria IAS-8: Data Analytics and Machine Learning Forschungszentrum Jülich Jülich Germany UniVie Doctoral School Computer Science University of Vienna Vienna Austria Data Science @ Uni Vienna University of Vienna Vienna Austria Institute of Computer Science and Campus Institute Data Science University of Göttingen Göttingen Germany Department of Computer Science Aarhus University Aarhus Denmark

In density-based clustering, clusters are areas of high object density separated by lower object density areas. This notion supports arbitrarily shaped clusters and automatic detection of noise points that do not belong to any cluster. However, it is challenging to adequately evaluate the quality of density-based clustering results. Even though some existing cluster validity indices (CVIs) target arbitrarily shaped clusters, none of them captures the quality of the labeled noise. In this paper, we propose DISCO, a Density-based Internal Score for Clustering Outcomes, which is the first CVI that also evaluates the quality of noise labels. DISCO reliably evaluates density-based clusters of arbitrary shape by assessing compactness and separation. It also introduces a direct assessment of noise labels for any given clustering. Our experiments show that DISCO evaluates density-based clusterings more consistently than its competitors. It is additionally the first method to evaluate the complete labeling of density-based clustering methods, including noise labels. © 2025, CC BY.

关键词： Cluster analysis

BoTier: Multi-Objective Bayesian Optimization with Tiered Composite Objectives

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Haddadnia, Mohammad Grashoff, Leonie Strieth-Kalthoff, Felix Harvard University Department of Biological Chemistry & Molecular Pharmacology BostonMA United States Harvard University Dana-Farber Cancer Institute Department of Cancer Biology BostonMA United States University of Wuppertal School of Mathematics and Natural Sciences Wuppertal Germany University of Wuppertal Interdisciplinary Center for Machine Learning and Data Analytics Wuppertal Germany

Scientific optimization problems are usually concerned with balancing multiple competing objectives, which come as preferences over both the outcomes of an experiment (e.g. maximize the reaction yield) and the corresponding input parameters (e.g. minimize the use of an expensive reagent). Typically, practical and economic considerations define a hierarchy over these objectives, which must be reflected in algorithms for sample-efficient experiment planning. Herein, we introduce BoTier, a composite objective that can flexibly represent a hierarchy of preferences over both experiment outcomes and input parameters. We provide systematic benchmarks on synthetic and real-life surfaces, demonstrating the robust applicability of BoTier across a number of use cases. Importantly, BoTier is implemented in an auto-differentiable fashion, enabling seamless integration with the BoTorch library, thereby facilitating adoption by the scientific community. © 2025, CC BY.

关键词： Bot (Internet)

Adapting Agricultural Virtual Environments in Game Engines to Improve HPC Accessibility 6th

学校读者我要写书评

暂无评论

Adapting Agricultural Virtual Environments in Game Engines ...

6th Nordic e-Infrastructure Collaboration Conference, NeIC 2024

作者： Baker, Dirk Norbert Bauer, Felix Schnepf, Andrea Scharr, Hanno Riedel, Morris Göbbert, Jens Henrik Hvannberg, Ebba School of Engineering and Natural Sciences University of Iceland Reykjavik Iceland Jülich Supercomputing Centre Forschungszentrum Jülich GmbH Jülich Germany Institute of Bio- and Geosciences 3: Agrosphere Forschungszentrum Jülich GmbH Jülich Germany Institute for Advanced Simulation 8: Data Analytics and Machine Learning Forschungszentrum Jülich GmbH Jülich Germany

ISBN: (纸本)9783031862397

E-infrastructures deliver basic supercomputing and storage capabilities but can benefit from innovative higher-level services that enable use-cases in critical domains, such as environmental and agricultural science. This work describes methods to distribute virtual scenes to the GPU nodes of a modular supercomputer for data generation. High information density virtual scenes, containing >100k geometries, typically cannot be rendered in real-time without techniques that change the information content, such as level-of-detail or culling approaches. Our work enables the concurrent and partitioned coupling to the image analysis in such a way that the data generation is dynamic and can be allocated to GPU nodes on demand, resulting in the possibility of moving through a continuous virtual scene rendered on multiple nodes. Within agricultural data analysis, the approach is especially impactful as virtual fields contain many individual geometries that coexist in one continuous system. Our work facilitates the generation of high-quality image data sets, which has the potential to solve the challenge of scarcity of well-annotated data in agricultural science. We use real-time communication standards to couple the data production with the image analysis training. We demonstrate how the use-case rendering impacts effective use of the compute nodes and furthermore develop techniques to distribute the workload to improve the data production. © The Author(s) 2025.

关键词： Virtual environments

VRoot: A VR-Based application for manual root system architecture reconstruction

学校读者我要写书评

暂无评论

Plant Phenomics 2025年第2期7卷

作者： Baker, Dirk N. Selzner, Tobias Göbbert, Jens Henrik Scharr, Hanno Riedel, Morris Hvannberg, Ebba þóra Schnepf, Andrea Zielasko, Daniel Jülich Supercomputing Centre Forschungszentrum Jülich GmbH Jülich Germany School of Engineering and Natural Sciences University of Iceland Reykjavík Iceland Institute of Bio- and Geosciences 3: Agrosphere Forschungszentrum Jülich GmbH Jülich Germany Institute of Advanced Simulation 8: Data Analytics and Machine Learning Forschungszentrum Jülich GmbH Jülich Germany Human-Computer Interaction Trier University Trier Germany

This article describes an immersive virtual reality reconstruction tool for root system architectures from 3D scans of soil columns. In practical scenarios, experimental conditions will be adapted to fit the need of the data analysis pipeline, including sieving and drying the soil before scanning. Based on previous reports of automatic systems that do not represent what experts would annotate, we developed a virtual reality system to assist with the extraction of root systems in cases in which automated approaches fall short of expert knowledge. The aim of the present study is to evaluate whether our immersive method is superior to classical annotation approaches when tested on synthetic data sets using untrained participants. Our laboratory user study consists of evaluating the root extractions of participants, along with their rating on central user experience and usability measures. We show significant improvement in F1 score across conditions (noisy or clear data) as well as an improved usability. Our study highlights that using virtual reality in root extraction improves accuracy, and we perform an in-depth evaluation of biases that occur when users trace roots in soil volumes. © 2025

关键词： Photointerpretation