Supervised Learning with Evolving Tasks and Performance Guarantees
Basque Center for Applied Mathematics (BCAM)(巴斯克应用数学中心(BCAM));IKERBASQUE-Basque Foundation for Science(巴斯克科学基金会(IKERBASQUE));University of the Basque Country UPV/EHU(巴斯克大学(UPV/EHU))
Verónica Álvarez, Santiago Mazuelas, Jose A. Lozano
备注:
arXiv admin note: text overlap with arXiv:2310.15974
Multiple supervised learning scenarios are composed by a sequence of classification tasks. For instance, multi-task learning and continual learning aim to learn a sequence of tasks that is either fixed or grows over time. Existing techniques for learning tasks that are in a sequence are tailored to specific scenarios, lacking adaptability to others. In addition, most of existing techniques consider situations in which the order of the tasks in the sequence is not relevant. However, it is common that tasks in a sequence are evolving in the sense that consecutive tasks often have a higher similarity. This paper presents a learning methodology that is applicable to multiple supervised learning scenarios and adapts to evolving tasks. Differently from existing techniques, we provide computable tight performance guarantees and analytically characterize the increase in the effective sample size. Experiments on benchmark datasets show the performance improvement of the proposed methodology in multiple scenarios and the reliability of the presented performance guarantees.
Minimax Forward and Backward Learning of Evolving Tasks with Performance Guarantees
Basque Center of Applied Mathematics (BCAM)(巴斯克应用数学中心);IKERBASQUE-Basque Foundation for Science(伊克尔巴斯克科学基金会);University of the Basque Country UPV/EHU(巴斯克大学)
Adv-SSL: Adversarial Self-Supervised Representation Learning with Theoretical Guarantees
School of Mathematics and Statistics, Wuhan University(武汉大学数学与统计学学院);School of Artificial Intelligence, Wuhan University(武汉大学人工智能学院);National Center for Applied Mathematics in Hubei, Wuhan University(湖北省应用数学中心);Hubei Key Laboratory of Computational Science, Wuhan University(湖北省计算科学重点实验室);Center of Statistical Research, Southwestern University of Finance and Economics(西南财经大学统计研究中心);School of Statistics and Data Science, Southwestern University of Finance and Economics(西南财经大学统计与数据科学学院);New Cornerstone Science Laboratory, Southwestern University of Finance and Economics(西南财经大学新基石科学实验室);Institute for Math & AI, Wuhan University(武汉大学数学与人工智能研究院)
EGFC: Evolving Gaussian Fuzzy Classifier from Never-Ending Semi-Supervised Data Streams -- With Application to Power Quality Disturbance Detection and Classification
Federal University of Lavras(拉夫拉斯联邦大学);University of Bologna(博洛尼亚大学);INFN Bologna(意大利国家核物理研究所博洛尼亚分部);Federal Center for Technological Education CEFET-MG(米纳斯吉拉斯联邦技术教育中心)
Computing the Shattering Coefficient of Supervised Learning Algorithms
University of São Paulo(圣保罗大学)
统计学习理论为基于经验风险最小化原则的有监督机器学习提供了理论保证。该原则定义了一个上限,以确保经验风险Remp(f)的一致收敛,即,在给定的数据样本上测量的误差,到风险R(f)的期望值(a.k.a.实际风险),这取决于将X中的输入示例x映射到Y中的类别标签y的联合概率分布P(X x Y)。只有当破碎系数N(F,2n)具有多项式增长行为时,才能确保一致收敛。本文证明了包含输入空间X的任何Hilbert空间H的Shattering系数,并讨论了其在监督机器算法的学习保证方面的影响。
Weakly Supervised Deep Learning Approach in Streaming Environments
Nanyang Technological University(南洋理工大学)
现有的数据流算法的可行性往往受到数据流的弱监督条件的阻碍。提出了一种自进化的深度神经网络,即Parsimonious Network(ParsNet),用于解决各种弱监督数据流问题。提出了一种带套期保值的自标记策略(SLASH),它的自校正机制处理了影响模型泛化能力的\textit{the accumulation of mistakes}。ParsNet是从自进化生成和判别训练过程的闭环配置开发的,利用共享参数,其结构灵活地增长和收缩,以克服有/无标签的概念漂移问题。数值评估已经进行了两个具有挑战性的问题,即零星访问地面真理和无限延迟访问地面真理。我们的数值研究显示了ParsNet的优势,在高维数据流和无限延迟模拟协议方面与同行有很大的差距。为了支持可重复的研究计划,ParsNet的源代码以及补充材料可在https://bit.ly/2qNW7p4上获得。
Generalized Maximum Entropy for Supervised Classification
BCAM-Basque Center for Applied Mathematics(巴斯克应用数学中心);IKERBASQUE-Basque Foundation for Science(巴斯克科学基金会);Department of Electronic Engineering, Tsinghua University(清华大学电子工程系)
Simulating Human Grandmasters: Evolution and Coevolution of Evaluation Functions
Bar-Ilan University(巴伊兰大学);Tilburg University(蒂尔堡大学);Gonda Brain Research Center at Bar-Ilan University(巴伊兰大学贡达脑科学研究中心);Center for Automation Research, University of Maryland(马里兰大学自动化研究中心)