检索结果-内蒙古大学图书馆

arXiv 2020年

作者： Grazzi, Riccardo Franceschi, Luca Pontil, Massimiliano Salzo, Saverio Computational Statistics and Machine Learning Istituto Italiano di Tecnologia Genoa Italy Department of Computer Science University College London London United Kingdom

We study a general class of bilevel problems, consisting in the minimization of an upper-level objective which depends on the solution to a parametric fixed-point equation. Important instances arising in machine learning include hyperparameter optimization, meta-learning, and certain graph and recurrent neural networks. Typically the gradient of the upper-level objective (hypergradient) is hard or even impossible to compute exactly, which has raised the interest in approximation methods. We investigate some popular approaches to compute the hypergradient, based on reverse mode iterative differentiation and approximate implicit differentiation. Under the hypothesis that the fixed point equation is defined by a contraction mapping, we present a unified analysis which allows for the first time to quantitatively compare these methods, providing explicit bounds for their iteration complexity. This analysis suggests a hierarchy in terms of computational efficiency among the above methods, with approximate implicit differentiation based on conjugate gradient performing best. We present an extensive experimental comparison among the methods which confirm the theoretical findings. Copyright © 2020, The Authors. All rights reserved.

关键词： computational efficiency

来源：评论

学校读者我要写书评

暂无评论

Multi-task averaging via task clustering

Multi-task averaging via task clustering

引用

2nd International Workshop on Similarity-Based Pattern Analysis and Recognition, SIMBAD 2013

作者： Martínez-Rego, David Pontil, Massimiliano Department of Computer Science University of a Coruña Campus de Elviña s/n 15071 A Coruña Spain Centre for Computational Statistics and Machine Learning Department of Computer Science University College London Malet Place Gower Street London WC1E 6BT United Kingdom

ISBN: (纸本)9783642391392

Multi-task averaging deals with the problem of estimating the means of a set of distributions jointly. It has its roots in the fifties when it was observed that leveraging data from related distributions can yield superior performance over learning from each distribution independently. Stein's paradox showed that, in an average square error sense, it is better to estimate the means of T Gaussian random variables using data sampled from all of them. This phenomenon has been largely disregarded and has recently emerged again in the field of multi-task learning. In this paper, we extend recent results for multi-task averaging to the n-dimensional case and propose a method to detect from data which tasks/distributions should be considered as related. Our experimental results indicate that the proposed method compares favorably to the state of the art. © 2013 Springer-Verlag.

关键词： Information theory

来源：评论

学校读者我要写书评

暂无评论

Conditional meta-learning of linear representations

arXiv

引用

arXiv 2021年

作者： Denevi, Giulia Pontil, Massimiliano Ciliberto, Carlo Department of Computer Science University College London London United Kingdom Istituto Italiano di Tecnologia Computational Statistics and Machine Learning Genova Italy

Standard meta-learning for representation learning aims to find a common representation to be shared across multiple tasks. The effectiveness of these methods is often limited when the nuances of the tasks' distribution cannot be captured by a single representation. In this work we overcome this issue by inferring a conditioning function, mapping the tasks' side information (such as the tasks' training dataset itself) into a representation tailored to the task at hand. We study environments in which our conditional strategy outperforms standard meta-learning, such as those in which tasks can be organized in separate clusters according to the representation they share. We then propose a meta-algorithm capable of leveraging this advantage in practice. In the unconditional setting, our method yields a new estimator enjoying faster learning rates and requiring less hyper-parameters to tune than current state-of-the-art methods. Our results are supported by preliminary experiments. Copyright © 2021, The Authors. All rights reserved.

关键词： machine learning

来源：评论

学校读者我要写书评

暂无评论

Meta Representation learning with Contextual Linear Bandits

arXiv

引用

arXiv 2022年

作者： Cella, Leonardo Lounici, Karim Pontil, Massimiliano Computational Statistics and Machine Learning Italian Institute of Technology Italy CMAP Ecole Polytechnique France Department of Computer Science University College London United Kingdom

Meta-learning seeks to build algorithms that rapidly learn how to solve new learning problems based on previous experience. In this paper we investigate meta-learning in the setting of stochastic linear bandit tasks. We assume that the tasks share a low dimensional representation, which has been partially acquired from previous learning tasks. We aim to leverage this information in order to learn a new downstream bandit task, which shares the same representation. Our principal contribution is to show that if the learned representation estimates well the unknown one, then the downstream task can be efficiently learned by a greedy policy that we propose in this work. We derive_an upper bound on the regret of this policy, which is, up to logarithmic factors, of order r√N(1 ∨ pd/T), where N is the horizon of the downstream task, T is the number of training tasks, d the ambient dimension and r d the dimension of the representation. We highlight that our strategy does not need to know r. We note that if T > d our bound achieves the same rate of optimal minimax bandit algorithms using the true underlying representation. Our analysis is inspired and builds in part upon previous work on meta-learning in the i.i.d. full information setting [Tripuraneni et al., 2021, Boursier et al., 2022]. As a separate contribution we show how to relax certain assumptions in those works, thereby improving their representation learning and risk analysis. Copyright © 2022, The Authors. All rights reserved.

关键词： Risk analysis

来源：评论

学校读者我要写书评

暂无评论

Anomaly Detection in Offshore Wind Turbine Structures using Hierarchical Bayesian Modelling 14

Anomaly Detection in Offshore Wind Turbine Structures using ...

引用

14th International Workshop on Structural Health Monitoring: Designing SHM for Sustainability, Maintainability, and Reliability, IWSHM 2023

作者： Smith, Simon M. Hughes, Aidan J. Dardeno, Tina A. Bull, Lawrence A. Dervilis, Nikolaos Worden, Keith Dynamics Research Group Department of Mechanical Engineering The 2University of Sheffield Western Bank SheffieldS10 2TN United Kingdom Computational Statistics and Machine Learning Group Department of Engineering University of Cam- bridge CambridgeCB3 0FA United Kingdom

ISBN: (纸本)9781605956930

Population-based structural health monitoring (PBSHM), aims to share information between members of a population. An offshore wind (OW) farm could be considered as a population of nominally-identical wind-turbine structures. However, benign variations exist among members, such as geometry, sea-bed conditions and temperature differences. These factors could influence structural properties and therefore the dynamic response, making it more difficult to detect structural problems via traditional SHM techniques. This paper explores the use of a hierarchical Bayesian model to infer expected soil stiffness distributions at both population and local levels, as a basis to perform anomaly detection, in the form of scour, for new and existing turbines. To do this, observations of natural frequency will be generated as though they are from a small population of wind turbines. Differences between individual observations will be introduced by postulating distributions over the soil stiffness and measurement noise, as well as reducing soil depth (to represent scour), in the case of anomaly detection. © 2023 by DEStech Publi cations, Inc. All rights reserved

关键词： Anomaly detection

来源：评论

学校读者我要写书评

暂无评论

Sensor placement and coordination via distributed multi-agent cooperative control 10

Sensor placement and coordination via distributed multi-agen...

引用

3rd International Conference on PErvasive Technologies Related to Assistive Environments, PETRA 2010

作者： Papangelis, Alexandros Metsis, Vangelis Shawe-Taylor, John Makedon, Fillia Heracleia Human-Centered Computing Laboratory Dept. of Computer Science and Engineering University of Texas Arlington 416 Yates St. Arlington TX 76019-0015 United States Centre for Computational Statistics and Machine Learning University College London London WC1E 6BT United Kingdom

ISBN: (纸本)9781450300711

This paper examines the problem of sensor placement and coordination to maximize the sensor utilization when mon-itoring different types of environments. Our assumption is that the sensors are mobile and each sensor can have more than one type of sensing capabilities which can be active or not at each specific moment. The goal is to maximize the amount of information collected from the environment, given the limited amount of resources that the total of the avail-able sensors can provide, and at the same time to be fault tolerant in failures of individual sensors by using a decen-tralized approach that re-organizes their placement in case of failures. We tackle this problem by employing a decentralized multi-agent coordination framework using message passing and the Max-Sum algorithm for building and maintaining a common picture of the area to be monitored. We show that by repre-senting each sensor as an independent agent which can take decisions individually and at the same time can affect the decisions of its neighbouring sensor-agents we can provide a robust and efficient system for the monitoring of life-critical environments such as assistive environments or governmen-tal infrastructures. Copyright 2010 ACM.

关键词： Multi agent systems

来源：评论

学校读者我要写书评

暂无评论

Neighborhood Component Feature Selection for Multiple Instance learning Paradigm

SSRN

引用

SSRN 2023年

作者： Romeo, Luca Turri, Giacomo Department of Economics and Law Università degli Studi di Macerata Macerata Italy Computational Statistics and Machine Learning Fondazione Istituto Italiano di Tecnologia Genova Italy

In a Multiple instance learning (MIL) scenario, the outcome annotation is usually only reported at the bag level. Considering simplicity, interpretability, and convergence criteria, the lazy learning approach (k-Nearest Neighbors [KNN]) also plays a crucial role in predicting bag labels in the MIL domain. Indeed, two extensions of the KNN algorithm to the MIL framework were proposed based on the Hausdorff metric and Bayesian/citation approaches. However, when irrelevant and redundant features influence the neighborhood search to the same degree as highly relevant features, the model’s generalization decreases. To overcome this limitation of KNN in the singleinstance learning scenario, a feature weighting algorithm named neighborhood component feature selection (NCFS) is often applied to find each feature’s optimal degree of influence. However, there is a relevant gap in the literature on how the NCFS model can be introduced in the MIL framework. No feature selection methodologies are integrated explicitly into Bayesian-KNN (BKNN) and Citation-KNN (CKNN) to discard redundant and irrelevant features and deal with the weakly supervised setting of the MIL framework. In this paper, we introduce the NCFS for the MIL framework to fill this gap. The proposed methodologies, NCFS-BKNN, NCFS-CKNN, and NCFSBayesian Citation-KNN (NCFS-BCKNN), learn the optimal features weighting vector by minimizing the regularized leave-one-out error of the training bags. Hence, the prediction of unseen bags was computed by combining the Bayesian and citation ap-proaches with the optimal weighted minimal Hausdorff distance. We demonstrated the statistical significance of the proposed methodologies with respect to state-of-the-art MIL algorithms that do not employ any a priori feature weighting strategy on different benchmark MIL datasets in the biomedical informatics and affective computing domains. © 2023, The Authors. All rights reserved.

关键词： Feature Selection

来源：评论

学校读者我要写书评

暂无评论

Neural Architecture Search for Genomic Sequence Data

Neural Architecture Search for Genomic Sequence Data

引用

IEEE Symposium on computational Intelligence and Bioinformatics and computational Biology (CIBCB)

作者： Amadeu Scheppach Hüseyin Anil Gündüz Emilio Dorigatti Philipp C. Münch Alice C. McHardy Bernd Bischl Mina Rezaei Martin Binder Department of Statistics LMU Munich Munich Germany Department of Statistics LMU Munich and Munich Center for Machine Learning Munich Germany Department for Computational Biology of Infection Research Helmholtz Centre for Infection Research Braunschweig Germany

Deep learning has enabled outstanding progress on bioinformatics datasets and a variety of tasks, such as protein structure prediction, identification of regulatory regions, genome annotation, and interpretation of the noncoding genome. The layout and configuration of neural networks used for these tasks have mostly been developed manually by human experts, which is a time-consuming and error-prone process. Therefore, there is growing interest in automated neural architecture search (NAS) methods in bioinformatics. In this paper, we present a novel search space for NAS algorithms that operate on genome data, thus creating extensions for existing NAS algorithms for sequence data that we name Genome-DARTS, Genome-P-DARTS, Genome-BONAS, Genome-SH, and Genome-RS. Moreover, we introduce two novel NAS algorithms, CWP-DARTS and EDPDARTS, that build on and extend the idea of P-DARTS. We evaluate the presented methods and compare them to manually designed neural architectures on a widely used genome sequence machine learning task to show that NAS methods can be adapted well for bioinformatics sequence datasets. Our experiments show that architectures optimized by our NAS methods outperform manually developed architectures while having significantly fewer parameters.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Operator World Models for Reinforcement learning

arXiv

引用

arXiv 2024年

作者： Novelli, Pietro Pratticò, Marco Pontil, Massimiliano Ciliberto, Carlo Computational Statistics and Machine Learning - Istituto Italiano di Tecnologia Genova16100 Italy AI Centre Computer Science Department University College London London United Kingdom

Policy Mirror Descent (PMD) is a powerful and theoretically sound methodology for sequential decision-making. However, it is not directly applicable to Reinforcement learning (RL) due to the inaccessibility of explicit action-value functions. We address this challenge by introducing a novel approach based on learning a world model of the environment using conditional mean embeddings. Leveraging tools from operator theory we derive a closed-form expression of the action-value function in terms of the world model via simple matrix operations. Combining these estimators with PMD leads to POWR, a new RL algorithm for which we prove convergence rates to the global optimum. Preliminary experiments in finite and infinite state settings support the effectiveness of our method. Copyright © 2024, The Authors. All rights reserved.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Subspace, Latent Structure and Feature Selection 1

引用

丛书名： Lecture Notes in Computer Science

1000年

作者： Craig Saunders Marko Grobelnik Steve Gunn John Shawe-Taylor

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：