Biological underpinnings for lifelong learning machines DOI
Dhireesha Kudithipudi, Mario Aguilar-Simon,

Jonathan Babb

et al.

Nature Machine Intelligence, Journal Year: 2022, Volume and Issue: 4(3), P. 196 - 210

Published: March 23, 2022

Language: Английский

A survey of deep meta-learning DOI Creative Commons
Mike Huisman, Jan N. van Rijn, Aske Plaat

et al.

Artificial Intelligence Review, Journal Year: 2021, Volume and Issue: 54(6), P. 4483 - 4541

Published: April 19, 2021

Deep neural networks can achieve great successes when presented with large data sets and sufficient computational resources. However, their ability to learn new concepts quickly is limited. Meta-learning one approach address this issue, by enabling the network how learn. The field of Meta-Learning advances at speed, but lacks a unified, in-depth overview current techniques. With work, we aim bridge gap. After providing reader theoretical foundation, investigate summarize key methods, which are categorized into i)~metric-, ii)~model-, iii)~optimization-based In addition, identify main open challenges, such as performance evaluations on heterogeneous benchmarks, reduction costs meta-learning.

Language: Английский

Citations

313

Transfer Learning in Deep Reinforcement Learning: A Survey DOI
Zhuangdi Zhu, Kaixiang Lin, Anil K. Jain

et al.

IEEE Transactions on Pattern Analysis and Machine Intelligence, Journal Year: 2023, Volume and Issue: 45(11), P. 13344 - 13362

Published: July 4, 2023

Reinforcement learning is a paradigm for solving sequential decision-making problems. Recent years have witnessed remarkable progress in reinforcement upon the fast development of deep neural networks. Along with promising prospects numerous domains such as robotics and game-playing, transfer has arisen to tackle various challenges faced by learning, transferring knowledge from external expertise facilitate efficiency effectiveness process. In this survey, we systematically investigate recent approaches context learning. Specifically, provide framework categorizing state-of-the-art approaches, under which analyze their goals, methodologies, compatible backbones, practical applications. We also draw connections between other relevant topics perspective explore potential that await future research progress.

Language: Английский

Citations

311

Learning to Generate Novel Domains for Domain Generalization DOI
Kaiyang Zhou, Yongxin Yang, Timothy M. Hospedales

et al.

Lecture notes in computer science, Journal Year: 2020, Volume and Issue: unknown, P. 561 - 578

Published: Jan. 1, 2020

Language: Английский

Citations

301

Deep Long-Tailed Learning: A Survey DOI
Yifan Zhang, Bingyi Kang, Bryan Hooi

et al.

IEEE Transactions on Pattern Analysis and Machine Intelligence, Journal Year: 2023, Volume and Issue: 45(9), P. 10795 - 10816

Published: April 19, 2023

Deep long-tailed learning, one of the most challenging problems in visual recognition, aims to train well-performing deep models from a large number images that follow class distribution. In last decade, learning has emerged as powerful recognition model for high-quality image representations and led remarkable breakthroughs generic recognition. However, imbalance, common problem practical tasks, often limits practicality network based real-world applications, since they can be easily biased towards dominant classes perform poorly on tail classes. To address this problem, studies have been conducted recent years, making promising progress field learning. Considering rapid evolution field, article provide comprehensive survey advances specific, we group existing into three main categories (i.e., re-balancing, information augmentation module improvement), review these methods following taxonomy detail. Afterward, empirically analyze several state-of-the-art by evaluating what extent issue imbalance via newly proposed evaluation metric, i.e., relative accuracy. We conclude highlighting important applications identifying directions future research.

Language: Английский

Citations

287

Online continual learning in image classification: An empirical survey DOI
Zheda Mai, Ruiwen Li,

Jihwan Jeong

et al.

Neurocomputing, Journal Year: 2021, Volume and Issue: 469, P. 28 - 51

Published: Oct. 13, 2021

Language: Английский

Citations

272

Fine-Grained Image Analysis With Deep Learning: A Survey DOI
Xiu-Shen Wei, Yi-Zhe Song, Oisin Mac Aodha

et al.

IEEE Transactions on Pattern Analysis and Machine Intelligence, Journal Year: 2021, Volume and Issue: 44(12), P. 8927 - 8948

Published: Nov. 9, 2021

Fine-grained image analysis (FGIA) is a longstanding and fundamental problem in computer vision pattern recognition, underpins diverse set of real-world applications. The task FGIA targets analyzing visual objects from subordinate categories, e.g., species birds or models cars. small inter-class large intra-class variation inherent to fine-grained makes it challenging problem. Capitalizing on advances deep learning, recent years we have witnessed remarkable progress learning powered FGIA. In this paper present systematic survey these advances, where attempt re-define broaden the field by consolidating two research areas - recognition retrieval. addition, also review other key issues FGIA, such as publicly available benchmark datasets related domain-specific We conclude highlighting several directions open problems which need further exploration community.

Language: Английский

Citations

242

Machine learning for a sustainable energy future DOI Open Access
Zhenpeng Yao, Yanwei Lum, Andrew Johnston

et al.

Nature Reviews Materials, Journal Year: 2022, Volume and Issue: 8(3), P. 202 - 215

Published: Oct. 18, 2022

Language: Английский

Citations

232

The Role of AI in Hospitals and Clinics: Transforming Healthcare in the 21st Century DOI Creative Commons
Shiva Maleki Varnosfaderani, Mohamad Forouzanfar

Bioengineering, Journal Year: 2024, Volume and Issue: 11(4), P. 337 - 337

Published: March 29, 2024

As healthcare systems around the world face challenges such as escalating costs, limited access, and growing demand for personalized care, artificial intelligence (AI) is emerging a key force transformation. This review motivated by urgent need to harness AI’s potential mitigate these issues aims critically assess integration in different domains. We explore how AI empowers clinical decision-making, optimizes hospital operation management, refines medical image analysis, revolutionizes patient care monitoring through AI-powered wearables. Through several case studies, we has transformed specific domains discuss remaining possible solutions. Additionally, will methodologies assessing solutions, ethical of deployment, importance data privacy bias mitigation responsible technology use. By presenting critical assessment transformative potential, this equips researchers with deeper understanding current future impact on healthcare. It encourages an interdisciplinary dialogue between researchers, clinicians, technologists navigate complexities implementation, fostering development AI-driven solutions that prioritize standards, equity, patient-centered approach.

Language: Английский

Citations

201

Towards Personalized Federated Learning DOI
Alysa Ziying Tan, Han Yu, Lizhen Cui

et al.

IEEE Transactions on Neural Networks and Learning Systems, Journal Year: 2022, Volume and Issue: 34(12), P. 9587 - 9603

Published: March 28, 2022

In parallel with the rapid adoption of artificial intelligence (AI) empowered by advances in AI research, there has been growing awareness and concerns data privacy. Recent significant developments regulation landscape have prompted a seismic shift interest toward privacy-preserving AI. This contributed to popularity Federated Learning (FL), leading paradigm for training machine learning models on silos manner. this survey, we explore domain personalized FL (PFL) address fundamental challenges heterogeneous data, universal characteristic inherent all real-world datasets. We analyze key motivations PFL present unique taxonomy techniques categorized according personalization strategies PFL. highlight their ideas, challenges, opportunities, envision promising future trajectories research new architectural design, realistic benchmarking, trustworthy approaches.

Language: Английский

Citations

185

Learned Initializations for Optimizing Coordinate-Based Neural Representations DOI
Matthew Tancik, Ben Mildenhall, Terrance Wang

et al.

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Journal Year: 2021, Volume and Issue: unknown

Published: June 1, 2021

Coordinate-based neural representations have shown significant promise as an alternative to discrete, array-based for complex low dimensional signals. However, optimizing a coordinate-based network from randomly initialized weights each new signal is inefficient. We propose applying standard meta-learning algorithms learn the initial weight parameters these fully-connected networks based on underlying class of signals being represented (e.g., images faces or 3D models chairs). Despite requiring only minor change in implementation, using learned enables faster convergence during optimization and can serve strong prior over modeled, resulting better generalization when partial observations given are available. explore benefits across variety tasks, including representing 2D images, reconstructing CT scans, recovering shapes scenes image observations.

Language: Английский

Citations

184