检索结果-内蒙古大学图书馆

Large Scale Multi-Task Bayesian Optimization with Large Language Models

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Zeng, Yimeng Maus, Natalie Jones, Haydn Thomas Tao, Jeffrey Wan, Fangping Torres, Marcelo Der Torossian de la Fuente-Nunez, Cesar Marcus, Ryan Bastani, Osbert Gardner, Jacob R. Department of Computer and Information Science University of Pennsylvania PhiladelphiaPA United States Machine Biology Group Departments of Psychiatry and Microbiology Institute for Biomedical Informatics Institute for Translational Medicine and Therapeutics Perelman School of Medicine University of Pennsylvania PhiladelphiaPA United States Departments of Bioengineering and Chemical and Biomolecular Engineering School of Engineering and Applied Science University of Pennsylvania PhiladelphiaPA United States Department of Chemistry School of Arts and Sciences University of Pennsylvania PhiladelphiaPA United States Penn Institute for Computational Science University of Pennsylvania PhiladelphiaPA United States

In multi-task Bayesian optimization, the goal is to leverage experience from optimizing existing tasks to improve the efficiency of optimizing new ones. While approaches using multi-task Gaussian processes or deep kernel transfer exist, the performance improvement is marginal when scaling to more than a moderate number of tasks. We introduce a novel approach leveraging large language models (LLMs) to learn from, and improve upon, previous optimization trajectories, scaling to approximately 2000 distinct tasks. Specifically, we propose an iterative framework in which an LLM is fine-tuned using the high quality solutions produced by BayesOpt to generate improved initializations that accelerate convergence for future optimization tasks based on previous search trajectories. We evaluate our method on two distinct domains: database query optimization and antimicrobial peptide design. Results demonstrate that our approach creates a positive feedback loop, where the LLM’s generated initializations gradually improve, leading to better optimization performance. As this feedback loop continues, we find that the LLM is eventually able to generate solutions to new tasks in just a few shots that are better than the solutions produced by "from scratch" by Bayesian optimization while simultaneously requiring significantly fewer oracle calls. © 2025, CC BY-SA.

关键词： Structured Query Language

Integrated investigation of the role of 3D architecture in the mechanical behavior of knitted textiles 18

学校读者我要写书评

暂无评论

Integrated investigation of the role of 3D architecture in t...

18th European Conference on Composite Materials, ECCM 2018

作者： Tekerek, Emine Liu, Dani Wisner, Brian Matthew, Melvin Breen, David Kontsos, Antonios Theoretical and Applied Mechanics Group Department of Mechanical Engineering and Mechanics Drexel University PhiladelphiaPA United States Department of Computer Science Drexel University United States

ISBN: (纸本)9781510896932

The mechanical behavior of 3D architectured knitted textiles is investigated experimentally and computationally. In this aspect, different imaging methods including microscopy and digital image correlation (DIC) are used to investigate the multiscale behavior of such structures. Local and global kinematics are observed in-situ as displacement is applied providing a link between behavior at the two scales. Additionally, Finite Element Analysis (FEA) is used to simulate the geometrical details of the entangled yarns in the 3D models developed. Parallelized Direct Numerical Simulations (DNS) conducted in high performance computing clusters and are shown to be capable of investigating the influence of local architecture at the yarn level on the global response. The effect of local contact evolution on kinematic behavior of knitted textiles is also investigated. The results obtained both computationally and experimentally were found to be in good agreement. © CCM 2020 - 18th European Conference on Composite Materials. All rights reserved.

关键词： Yarn

Anytime diagnosis for reconfiguration

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Felfernig, Alexander Walter, Rouven Galindo, José A. Benavides, David Erdeniz, Seda Polat Atas, Müslüm Reiterer, Stefan Applied Software Engineering Group Institute for Software Technology TU Graz Austria Symbolic Computation Group WSI Informatics Universität Tübingen Germany Computer Languages and Systems Department University of Sevilla Spain SelectionArts Austria

Many domains require scalable algorithms that help to determine diagnoses efficiently and often within predefined time limits. Anytime diagnosis is able to determine solutions in such a way and thus is especially useful in real-time scenarios such as production scheduling, robot control, and communication networks management where diagnosis and corresponding reconfiguration capabilities play a major role. Anytime diagnosis in many cases comes along with a trade-off between diagnosis quality and the efficiency of diagnostic reasoning. In this paper we introduce and analyze FLEXDIAG which is an anytime direct diagnosis approach. We evaluate the algorithm with regard to performance and diagnosis quality using a configuration benchmark from the domain of feature models and an industrial configuration knowledge base from the automotive domain. Results show that FLEXDIAG helps to significantly increase the performance of direct diagnosis search with corresponding quality tradeoffs in terms of minimality and accuracy. Copyright © 2021, The Authors. All rights reserved.

关键词： Scheduling

BiLSTM with Data Augmentation using Interpolation Methods to Improve Early Detection of Parkinson Disease 15

学校读者我要写书评

暂无评论

BiLSTM with Data Augmentation using Interpolation Methods to...

15th Federated Conference on computer science and Information Systems, FedCSIS 2020

作者： Abayomi-Alli, Olusola O. Damasevicius, Robertas Maskeliunas, Rytis Abayomi-Alli, Adebayo Kaunas University of Technology Department of Software Engineering Kaunas Lithuania Vytautas Magnus University Department of Applied Informatics Kaunas Lithuania Federal University of Agriculture Department of Computer Science Abeokuta Nigeria

ISBN: (纸本)9788395541674

The lack of dopamine in the human brain is the cause of Parkinson disease (PD) which is a degenerative disorder common globally to older citizens. However, late detection of this disease before the first clinical diagnosis has led to increased mortality rate. Research effort towards the early detection of PD has encountered challenges such as: small dataset size, class imbalance, overfitting, high false detection rate, model complexity, etc. This paper aims to improve early detection of PD using machine learning through data augmentation for very small datasets. We propose using Spline interpolation and Piecewise Cubic Hermite Interpolating Polynomial (Pchip) interpolation methods to generate synthetic data instances. We further investigate on reducing dimensionality of features for effective and real-time classification while considering computational complexity of implementation on real-life mobile phones. For classification we use Bidirectional LSTM (BiLSTM) deep learning network and compare the results with traditional machine learning algorithms like Support Vector Machine (SVM), Decision Tree, Logistic regression, KNN and Ensemble bagged tree. For experimental validation we use the Oxford Parkinson disease dataset with 195 data samples, which we have augmented with 571 synthetic data samples. The results for BiLSTM shows that even with a holdout of 90%, the model was still able to effectively recognize PD with an average accuracy for ten rounds experiment using 22 features as 82.86%, 97.1%, and 96.37% for original, augmented (Spline) and augmented (Pchip) datasets, respectively. Our results show that proposed data augmentation schemes have significantly (p © 2020 Polish Information Processing Society - as it is since 2011.

关键词： Interpolation

HelioCubed: A High Order Inner Heliosphere Simulation Code with a Mapped Cubed Sphere Grid and Adaptive Mesh Refinement 16

学校读者我要写书评

暂无评论

HelioCubed: A High Order Inner Heliosphere Simulation Code w...

16th International Conference on Numerical Modeling of Space Plasma Flows, ASTRONUM 2024

作者： Singh, Talwinder Colella, Phillip Van-Straalen, Brian Bozhart, Chris Pogorelov, Nikolai V. Department of Physics & Astronomy Georgia State University AtlantaGA30303 United States Applied Numerical Algorithms Group Lawrence Berkeley National Laboratory BerkeleyCA94720 United States Electrical Engineering and Computer Science Department University of California Berkeley BerkeleyCA94720 United States Department of Space Science The University of Alabama in Huntsville HuntsvilleAL35805 United States Center for Space Plasma and Aeronomic Research The University of Alabama in Huntsville HuntsvilleAL35805 United States

We present the HelioCubed, a high-order magnetohydrodynamic (MHD) code designed for modeling the inner heliosphere. The code is designed to achieve 4th order accuracy both in space and in time. In addition, HelioCubed can perform simulations on mapped grids, such as those based on cubed spheres, which makes it possible to overcome stability limitations caused by the geometrical singularity at the polar axis of a spherical grid, thus enabling substantially larger time steps. HelioCubed has been developed using the high-level Proto library, ensures performance portability across CPU and GPU architectures, and supports back-end implementations, e.g., CUDA, HIP, OpenMP, and MPI. The code is compatible with the HDF5 library, which facilitates seamless data handling for simulations and boundary conditions derived from semi-empirical and MHD models of the solar corona. While presenting the results of preliminary simulations, we demonstrate that our simulations are indeed performed with 4th order of accuracy. Our approach ensures that HelioCubed solves the MHD equations preserving the radial flow to machine round-off error even on cubed-sphere grids. Solar wind simulations are performed using the boundary conditions provided by the Wang-Sheeley-Arge coronal model of the ambient solar wind. It also allows us to to simulate coronal mass ejections using observation-driven flux rope models. These capabilities make HelioCubed a versatile and powerful tool to advance heliophysics research and space weather forecasting. © 2025 Institute of Physics Publishing. All rights reserved.

关键词： Mesh generation

Technical Debt Management: The Road Ahead for Successful software Delivery

学校读者我要写书评

暂无评论

Technical Debt Management: The Road Ahead for Successful Sof...

software engineering: Future of software engineering (ICSE-FoSE), IEEE/ACM International Conference on

作者： Paris Avgeriou Ipek Ozkaya Alexander Chatzigeorgiou Marcus Ciolkowski Neil A. Ernst Ronald J. Koontz Eltjo Poort Forrest Shull Dept. of Computing Science University of Groningen Groningen The Netherlands Software Engineering institute Carnegie Mellon University Pittsburgh PA USA Dept. of Applied informatics University of Macedonia Thessaloniki Greece QAware GmbH München Germany Department of Computer Science University of Victoria Victoria Canada The Boeing Company Mesa AZ USA CGI Rotterdam The Netherlands

Technical Debt, considered by many to be the ‘silent killer’ of software projects, has undeniably become part of the everyday vocabulary of software engineers. We know it compromises the internal quality of a system, either deliberately or inadvertently. We understand Technical Debt is not all derogatory, often serving the purpose of expediency. But, it is associated with a clear risk, especially for large and complex systems with extended service life: if we do not properly manage Technical Debt, it threatens to “bankrupt” those systems. software engineers and organizations that develop software-intensive systems are facing an increasingly more dire future state of those systems if they do not start incorporating Technical Debt management into their day to day practice. But how? What have the wins and losses of the past decade of research and practice in managing Technical Debt taught us and where should we focus next? In this paper, we examine the state of the art in both industry and research communities in managing Technical Debt; we subsequently distill the gaps in industrial practice and the research shortcomings, and synthesize them to define and articulate a vision for what Technical Debt management looks like five years hence.

关键词：

A Video-based Study on Perceived Intelligence, Subjective Performance and Trust under Variation of Prior Information given to Users in Autonomous Driving

学校读者我要写书评

暂无评论

A Video-based Study on Perceived Intelligence, Subjective Pe...

IEEE Workshop on Advanced Robotics and its Social Impacts, ARSO

作者： Kolja Kü hnlenz Ute Schmid Barbara Kü hnlenz Department of Electrical Engineering and Computer Science Robotics Research Lab Coburg University of Applied Sciences and Arts Coburg Germany Cognitive Systems Group University of Bamberg Bamberg Germany Academic Center for Sciences and Humanities Coburg University of Applied Sciences and Arts Coburg Germany

This paper presents results from a video-based study on the impact of prior information on the user experience dimensions perceived intelligence, subjective performance, and trust in autonomous driving. A simulated autonomous driving situation is presented to test participants, while they are given different prior information in terms of descriptions of the autonomous vehicle and its integration in a network of intelligent vehicles. Significant differences of the tested user experience dimensions are revealed underlining the importance of information provided to the potential users and towards the criticality of information regarding the introduction of intelligent technology in general.

关键词：

Deep deterministic uncertainty for semantic segmentation

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Mukhoti, Jishnu van Amersfoort, Joost Torr, Philip H.S. Gal, Yarin Oxford Applied & Theoretical Machine Learning Group Department of Computer Science University of Oxford Oxford United Kingdom Torr Vision Group Department of Engineering Science University of Oxford Oxford United Kingdom

We extend Deep Deterministic Uncertainty (DDU) (Mukhoti et al., 2021), a method for uncertainty estimation using feature space densities, to semantic segmentation. DDU enables quantifying and disentangling epistemic and aleatoric uncertainty in a single forward pass through the model. We study the similarity of feature representations of pixels at different locations for the same class and conclude that it is feasible to apply DDU location independently, which leads to a significant reduction in memory consumption compared to pixel dependent DDU. Using the DeepLab-v3+ architecture on Pascal VOC 2012, we show that DDU improves upon MC Dropout and Deep Ensembles while being significantly faster to compute. Copyright © 2021, The Authors. All rights reserved.

关键词： Pixels

Awareness in robotics: An early perspective from the viewpoint of the EIC Pathfinder Challenge "Awareness Inside"

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Santina, Cosimo Della Corbato, Carlos Hernandez Sisman, Burak Leiva, Luis A. Arapakis, Ioannis Vakalellis, Michalis Vanderdonckt, Jean D’Haro, LuisFernando Manzi, Guido Becchio, Cristina Elamrani, Aïda Alirezaei, Mohsen Castellano, Ginevra Dimarogonas, Dimos V. Ghosh, Arabinda Haesaert, Sofie Soudjani, Sadegh Stroeve, Sybert Verschure, Paul Bacciu, Davide Deroy, Ophelia Bahrami, Bahador Gallicchio, Claudio Hauert, Sabine Sanz, Ricardo Lanillos, Pablo Iacca, Giovanni Sigg, Stephan Gasulla, Manel Steels, Luc Sierra, Carles Department of Cognitive Robotics Delft University of Technology Delft Netherlands Oberpfaffenhofen Germany Department of Computer Science University of Luxembourg Luxembourg 4 Telefónica I+D Spain AEGIS IT Research GmbH Germany Université catholique de Louvain Belgium Speech Technology and Machine Learning Group ETSI de Telecomunicación Universidad Politécnica de Madrid Madrid Spain Alien Venture Studio Alien Technology Transfer Rome Italy Hamburg Germany Institut Jean Nicod École Normale Supérieure Paris France Research Technology and Development Department Digital Industry Software Helmond Netherlands Department of Information Technology Uppsala University Uppsala Sweden Division of Decision and Control Systems School of Electrical Engineering and Computer Science KTH Royal Institute of Technology Stockholm Sweden Max Planck Institute for Software Systems Kaiserslautern Germany Department of Electrical Engineering Control Systems Group Eindhoven University of Technology Eindhoven Netherlands Amsterdam Netherlands Donders Centre for Neuroscience Radboud University Nijmegen Netherlands Computer Science Department University of Pisa Pisa Italy Faculty of Philosophy Philosophy of Science The Study of Religion Ludwig Maximilian University of Munich Munich Germany School of Engineering Mathematics and Technology University of Bristol Bristol United Kingdom Autonomous Systems Laboratory Universidad Politécnica de Madrid Spain Madrid Spain Department of Information Engineering and Computer Science University of Trento Trento Italy Department of Information and Communications Engineering Aalto University Espoo Finland Department of Electronic Engineering Universitat Politècnica de Catalunya Barcelona Spain Studio Stelluti Brussels Belgium Barcelona Spain

Consciousness has been historically a heavily debated topic in engineering, science, and philosophy. On the contrary, awareness had less success in raising the interest of scholars in the past. However, things are changing as more and more researchers are getting interested in answering questions concerning what awareness is and how it can be artificially generated. The landscape is rapidly evolving, with multiple voices and interpretations of the concept being conceived and techniques being developed. The goal of this paper is to summarize and discuss the ones among these voices that are connected with projects funded by the EIC Pathfinder Challenge called "Awareness Inside", a nonrecurring call for proposals within Horizon Europe that was designed specifically for fostering research on natural and synthetic awareness. In this perspective, we dedicate special attention to challenges and promises of applying synthetic awareness in robotics, as the development of mature techniques in this new field is expected to have a special impact on generating more capable and trustworthy embodied systems. © 2024, CC BY.

关键词： Robotics