Information Maximizing Adaptation Network With Label Distribution Priors for Unsupervised Domain Adaptation

Article Automation & Control Systems

Two-Stage Selective Ensemble of CNN via Deep Tree Training for Medical Image Classification

Yun Yang et al.

Summary: The article introduces a novel deep tree training strategy and achieves better performance in medical image classification through a two-stage selective ensemble of CNN branches.

IEEE TRANSACTIONS ON CYBERNETICS (2022)

Add to Collection

Article Computer Science, Interdisciplinary Applications

Reservoir hosts prediction for COVID-19 by hybrid transfer learning model

Yun Yang et al.

Summary: In this study, a hybrid transfer learning model is proposed using machine learning techniques to more accurately predict the animal origins of COVID-19 in the face of limited training sets and class-imbalance issues. By utilizing the related virus domain as auxiliary to help build a robust model for the target virus domain, the model achieves accurate prediction of viral hosts.

JOURNAL OF BIOMEDICAL INFORMATICS (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Deep Subdomain Adaptation Network for Image Classification

Yongchun Zhu et al.

Summary: This study introduces a deep subdomain adaptation network (DSAN) that aligns relevant subdomain distributions across different domains based on the local maximum mean discrepancy (LMMD). DSAN is simple but effective, does not require adversarial training, and converges quickly. It can be easily integrated into feedforward network models to achieve efficient adaptation via backpropagation.

IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS (2021)

Add to Collection

Article Automation & Control Systems

Joint Learning of Multiple Latent Domains and Deep Representations for Domain Adaptation

Xinxiao Wu et al.

Summary: A novel unified framework based on deep neural networks is proposed for domain adaptation, which aims to predict latent domains from source data and learn deep representations from both source and target data. The framework utilizes an iterative algorithm to separate the source domain into latent clusters and train deep neural networks using domain membership as supervision, achieving better performance in latent domain discovery and feature learning tasks. The key idea behind this joint learning framework is that good representations can enhance prediction accuracy of latent domains, while domain prediction results provide useful supervisory information for feature learning, leading to semantically meaningful features among different classes.

IEEE TRANSACTIONS ON CYBERNETICS (2021)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

SENTRY: Selective Entropy Optimization via Committee Consistency for Unsupervised Domain Adaptation

Viraj Prabhu et al.

Summary: SENTRY is a UDA algorithm that selectively minimizes or maximizes predictive entropy based on the predictive consistency of target instances under a committee of random image transformations, leading to significant improvements over existing methods in various benchmarks when combined with pseudolabel-based approximate target class balancing.

2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021) (2021)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

Dynamic Weighted Learning for Unsupervised Domain Adaptation

Ni Xiao et al.

Summary: This paper proposes Dynamic Weighted Learning (DWL) to address the issue of negative transfer in unsupervised domain adaptation, caused by insufficient balance between domain alignment and class discrimination. By dynamically weighting the learning losses and considering the problem of sample imbalance across domains, DWL demonstrates excellent performance on several benchmark datasets.

2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021 (2021)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

Unsupervised Multi-source Domain Adaptation Without Access to Source Data

Sk Miraj Ahmed et al.

Summary: The research aims to achieve unsupervised domain adaptation without the need for access to source data, selecting the best source model. They proposed an efficient algorithm that, in most cases, not only reaches the best source accuracy but also outperforms it.

2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021 (2021)

Add to Collection

Proceedings Paper Acoustics

CROSS-DOMAIN SENTIMENT CLASSIFICATION WITH CONTRASTIVE LEARNING AND MUTUAL INFORMATION MAXIMIZATION

Tian Li et al.

Summary: This work proposes a novel model CLIM for cross-domain sentiment classification, which explores the potential of contrastive learning and mutual information maximization for learning domain-invariant and task-discriminative features, achieving significant results.

2021 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP 2021) (2021)

Add to Collection

Article Computer Science, Information Systems

Adversarial Network With Multiple Classifiers for Open Set Domain Adaptation

Tasfia Shermin et al.

Summary: Domain adaptation aims to transfer knowledge from a domain with adequate labeled samples to a domain with scarce labeled samples. This paper focuses on an open set domain adaptation setting where the target domain has both private and shared label space, while the source domain only has the shared label space. Traditional distribution-matching domain adaptation methods are inadequate for addressing this specific scenario.

IEEE TRANSACTIONS ON MULTIMEDIA (2021)

Add to Collection

Article Computer Science, Information Systems