ME-D2N: Multi-Expert Domain Decompositional Network for Cross-Domain Few-Shot Learning

Fu, Yuqian; Xie, Yu; Fu, Yanwei; Chen, Jingjing; Jiang, Yu‐Gang

doi:10.1145/3503161.3547995

Cited by 15 publications

(3 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Technically, a novel meta-FDMixup network is proposed to extract the disentangled domain-irrelevant and domain-specific features with a novel disentangle module and a domain classifier. And [23] follows this setup (introduce few labeled target domian data) and proposes a Multi-Expert Domain Decompositional Network (ME-D2N) to solve CDFSL. The loss function also include CE and KL loss.…”

Section: Hybrid Approachesmentioning

confidence: 99%

Deep Learning for Cross-Domain Few-Shot Visual Recognition: A Survey

Xu¹,

Zhi²,

Sun³

et al. 2023

Preprint

View full text Add to dashboard Cite

Deep learning has been highly successful in computer vision with large amounts of labeled data, but struggles with limited labeled training data. To address this, Few-shot learning (FSL) is proposed, but it assumes that all samples (including source and target task data, where target tasks are performed with prior knowledge from source ones) are from the same domain, which is a stringent assumption in the real world. To alleviate this limitation, Cross-domain few-shot learning (CDFSL) has gained attention as it allows source and target data from different domains and label spaces. This paper provides a comprehensive review of CDFSL at the first time, which has received far less attention than FSL due to its unique setup and difficulties. We expect this paper to serve as both a position paper and a tutorial for those doing research in CDFSL. This review first introduces the definition of CDFSL and the issues involved, followed by the core scientific question and challenge. A comprehensive review of validated CDFSL approaches from the existing literature is then presented, along with their detailed descriptions based on a rigorous taxonomy. Furthermore, this paper outlines and discusses several promising directions of CDFSL that deserve further scientific investigation, covering aspects of problem setups, applications and theories.

show abstract

Section: Hybrid Approachesmentioning

confidence: 99%

Deep Learning for Cross-Domain Few-Shot Visual Recognition: A Survey

Xu¹,

Zhi²,

Sun³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The Thirty-Eighth AAAI Conference on Artificial Intelligence (AAAI-24) become even more pronounced for cross-domain tasks (Fu et al 2023) where generalization to novel classes by finetuning has not been optimized by a simple pipeline such as PMF . In summary, we identify and address two problems in conventional deep learning models that stem from the few training data in fine-tuning: (1) the inability of conventional models to sufficiently extract informative features for downstream classification tasks, and (2) the suboptimal fine-tuning due to the insufficient training samples as well as the lack of local-global consistency between extracted features.…”

Section: Introductionmentioning

confidence: 99%

Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention

Chen,

Hsieh,

et al. 2024

AAAI

View full text Add to dashboard Cite

Due to the scarcity of training samples, Few-Shot Learning (FSL) poses a significant challenge to capture discriminative object features effectively. The combination of transfer learning and meta-learning has recently been explored by pre-training the backbone features using labeled base data and subsequently fine-tuning the model with target data. However, existing meta-learning methods, which use embedding networks, suffer from scaling limitations when dealing with a few labeled samples, resulting in suboptimal results. Inspired by the latest advances in FSL, we further advance the approach of fine-tuning a pre-trained architecture by a strengthened hierarchical feature representation. The technical contributions of this work include: 1) a hybrid design named Intra-Block Fusion (IBF) to strengthen the extracted features within each convolution block; and 2) a novel Cross-Scale Attention (CSA) module to mitigate the scaling inconsistencies arising from the limited training samples, especially for cross-domain tasks. We conducted comprehensive evaluations on standard benchmarks, including three in-domain tasks (miniImageNet, CIFAR-FS, and FC100), as well as two cross-domain tasks (CDFSL and Meta-Dataset). The results have improved significantly over existing state-of-the-art approaches on all benchmark datasets. In particular, the FSL performance on the in-domain FC100 dataset is more than three points better than the latest PMF (Hu et al. 2022).

show abstract

“…Few-shot learning is an outstanding experimentation topic that has been spaciously inspected in the past few years [6] [7]- [14]. Few-shot learning for image classification (few-shot image classification) is a part of machine learning but still needs a lot of research and improvement [15], [16].…”

Section: Introductionmentioning

confidence: 99%

Closer Look at Image Classification for Indonesian Sign Language with Few-Shot Learning Using Matching Network Approach

Sari

2023

JOIV : Int. J. Inform. Visualization

View full text Add to dashboard Cite

Huge datasets are important to build powerful pipelines and ground well to new images. In Computer Vision, the most basic problem is image classification. The classification of images may be a tedious job, especially when there are a lot of amounts. But CNN is known to be data-hungry while gathering. How can we build some models without much data? For example, in the case of Sign Language Recognition (SLR). One type of Sign Language Recognition system is vision-based. In Indonesian Sign Language dataset has a relatively small sample image. This research aims to classify sign language images using Computer Vision for Sign Language Recognition systems. We used a small dataset, Indonesian Sign Language. Our dataset is listed in 26 classes of alphabet, A-Z. It has loaded 12 images for each class. The methodology in this research is few-shot learning. Based on our experiment, the best accuracy for few-shot learning is Mnasnet1_0 (85.75%) convolutional network model for Matching Networks, and loss estimation is about 0,43. And the experiment indicates that the accuracy will be increased by increasing the number of shots. We can inform you that this model's matching network framework is unsuitable for the Inception V3 model because the kernel size cannot be greater than the actual input size. We can choose the best algorithm based on this research for the Indonesian Sign Language application we will develop further.

show abstract

ME-D2N: Multi-Expert Domain Decompositional Network for Cross-Domain Few-Shot Learning

Cited by 15 publications

References 32 publications

Deep Learning for Cross-Domain Few-Shot Visual Recognition: A Survey

Deep Learning for Cross-Domain Few-Shot Visual Recognition: A Survey

Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention

Closer Look at Image Classification for Indonesian Sign Language with Few-Shot Learning Using Matching Network Approach

Contact Info

Product

Resources

About