Few-shot activity learning by dual Markov logic networks

Zhang, Zhimin; Zhu, Tao; Gao, Dazhi; Xu, Jiabo; Hong, Lu; Ning, Huansheng

doi:10.1016/j.knosys.2022.108158

Cited by 2 publications

(2 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These algorithms can generalize some unseen data giving only a few labeled training examples to the network. The learning to learn approach, which includes meta training 9,10,22 was introduced to make the FSL algorithm an effective domain. It divides the problem into two parts one is the base level in which standard supervised learning is performed, another part is the meta-level that extracts the information from the base level.…”

Section: Literature Reviewmentioning

confidence: 99%

“…The advantage of using the FSL technique is that a network with a small set of labeled images is possible to train in less time while in many machine learning methods, a huge amount of data and time are required to train a network so that it can accurately predict. These days many approaches are used for FSL like learning to learn which includes meta training, [9][10][11] gradient-based approaches, 12,13 matric-based approaches, 14,15 and transductive learning. 16,17 Matric-based approaches have shown much more improvements as compared to the traditional meta-learning and matching algorithms.…”

mentioning

confidence: 99%

See 1 more Smart Citation

An intelligent medical image classification system using few‐shot learning

Abbas

2022

Concurrency and Computation

View full text Add to dashboard Cite

Summary Due to the wide range of diseases and imaging modalities, a retrieving system is a challenging task to access the corresponding clinical cases from a large medical repository on time. Several intelligent systems are developed to recognize medical images based on various machine learning algorithms such as deep learning and transfer learning methods that have outperformed compared to traditional methods. However, they are time‐consuming and required more labeled data for training the network models. To overcome these issues, an intelligence medical imaging recognition (IMR‐FSL) system is developed based on three phases of the few‐short learning (FSL) concept, which required few labeled examples to train the model and they are capable to generalize on huge and unseen medical datasets. To test and train the performance of the proposed IMR‐FSL system, many imaging modalities and multiclass datasets such as TCIA and Kvasir are utilized. Statistical analysis is also performed to measure the performance of the IMR‐FSL system. On average, the IMR‐FSL system achieved SE of 95%, SP of 96.5%, ACC of 97.5%, and AUC of 0.96 on both datasets. The performance of the IMR‐FSL system demonstrates the effectiveness of other state‐of‐the‐art systems.

show abstract

Section: Literature Reviewmentioning

confidence: 99%