Visual-semantic Fuzzy Interaction Network for Zero-shot Learning

IEEE Transactions on Artificial Intelligence, Год журнала: 2025, Номер 6(5), С. 1345 - 1359

Опубликована: Янв. 3, 2025

Zero-shot learning (ZSL) aims to recognize unseen class image objects using manually defined semantic knowledge corresponding both seen and images. The key of ZSL lies in building the interaction between precise data fuzzy knowledge. fuzziness is attributed difficulty quantifying human However, existing methods ignore inherent treat it as during visual-semantic interaction. This not good for transferring from classes classes. In order solve this problem, we propose a Visual-semantic Fuzzy Interaction Network (VSFIN) ZSL. VSFIN utilize an effective encoder-decoder structure, including prototype encoder (SPE) visual feature decoder (VFD). SPE VFD enable features interact with via cross-attention. To achieve VFD, introduce concept membership function set theory design loss function. allows certain degree imprecision interaction, thereby enabling becomingly given Moreover, rank sum test distribution alignment alleviate bias towards Extensive experiments on three widely used benchmarks have demonstrated that outperforms current state-of-the-art under conventional (CZSL) generalized (GZSL) settings.

Язык: Английский

---

A transformer-based dual contrastive learning approach for zero-shot learning

Neurocomputing, Год журнала: 2025, Номер unknown, С. 129530 - 129530

Опубликована: Фев. 1, 2025

Язык: Английский

---

Vision transformer-based generalized zero-shot learning with data criticizing

Applied Intelligence, Год журнала: 2025, Номер 55(6)

Опубликована: Фев. 7, 2025

Язык: Английский

---

ReKon3D: Relation-knowledge aware multi-modal embedding and contrastive GAN for zero-shot 3D recognition

Neurocomputing, Год журнала: 2025, Номер 640, С. 130264 - 130264

Опубликована: Апрель 28, 2025

Язык: Английский

---

META-EEG: Meta-learning-based class-relevant EEG representation learning for zero-calibration brain–computer interfaces

Expert Systems with Applications, Год журнала: 2023, Номер 238, С. 121986 - 121986

Опубликована: Окт. 10, 2023

Transfer learning for motor imagery-based brain-computer interfaces (MI-BCIs) struggles with inter-subject variability, hindering its generalization to new users. This paper proposes an advanced implicit transfer framework, META-EEG, designed overcome the challenge arising from variability. By incorporating gradient-based meta-learning intermittent freezing strategy, META-EEG ensures efficient feature representation learning, providing a robust zero-calibration solution. A comparative analysis reveals that significantly outperforms all baseline methods and competing on three different public datasets. Moreover, we demonstrate efficiency of proposed model through neurophysiological feature-representational analysis. With robustness superior performance challenging datasets, provides effective solution calibration-free MI-EEG classification, facilitating broader usability.

Язык: Английский

---

Zero-shot text classification with knowledge resources under label-fully-unseen setting

Neurocomputing, Год журнала: 2024, Номер unknown, С. 128580 - 128580

Опубликована: Сен. 1, 2024

Язык: Английский

---

Consistent penalizing field loss for zero-shot image retrieval

Expert Systems with Applications, Год журнала: 2023, Номер 236, С. 121287 - 121287

Опубликована: Авг. 25, 2023

Язык: Английский

---

Advancing zero-shot semantic segmentation through attribute correlations

Neurocomputing, Год журнала: 2024, Номер 594, С. 127829 - 127829

Опубликована: Май 10, 2024

Язык: Английский

---

Generative-based hybrid model with semantic representations for generalized zero-shot learning

Signal Image and Video Processing, Год журнала: 2024, Номер 19(1)

Опубликована: Дек. 2, 2024

Язык: Английский

---

A nuclide identification method of γ spectrum and model building based on the transformer

Nuclear Science and Techniques, Год журнала: 2024, Номер 36(1)

Опубликована: Дек. 18, 2024

Язык: Английский

---