联邦持续学习(federated continual learning)中存在多个客户端, 每个客户端不断获取新的数据. 由于存储空间的限制, 客户端以持续学习的方式进行模型更新, 同时客户端之间通过共享知识提升本地模型性能. Yoon等 [9]最早给出了联邦持续学习的定义, 其提出的方法将模型参数拆分为通用参数和任务特定参数, 客户端之间
Thus, learning the global representation of all class data together can alleviate the problem of model overfitting on the base classes like machine learning and deep learning. The novel class data is learned at the beginning, and the metric novel class data and the global representation strengthen...
Current zero shot learning methods mostly focus on learning the mapping function from image feature space to semantic space which is extremely important. However, these methods assume the seen and unseen class prototypes are fixed. A class prototype is referred to the semantic representation of a ...
We identify shortcut learning as the key limiting fac- tor for online CL, where the learned features may be bi- ased, not generalizable to new tasks, and may have an ad- verse impact on knowledge distillation. To tackle this issue, we present the online prototype learning (...
The features are reconstructed using basis vectors in sub- prototype space, and since all majority and minority classes share these vectors, the rich knowledge of the majority classes can assist the minority classes in learning a more robust representation. Duri...
To give an intuition of open set recognition problem, Fig.1shows the difference between closed set recognition and open set recognition from the perspective of representation learning. Assuming that a data set contains four known categories, and with the open set condition, the data set might also...
{Unsupervised Voice-Face Representation Learning by Cross-Modal Prototype Contrast}, author={Zhu, Boqing and Xu, Kele and Wang, Changjian and Qin, Zheng and Sun, Tao and Wang, Huaimin and Peng, Yuxing}, booktitle={Proceedings of the Thirty-First International Joint Conference on Artificial ...
To build the dissimilarity measure, we use a vectorial representation of each observation with the 365 daily temperatures. We measure the dissimilarity between n and i as the Euclidean distance between the corresponding vectors of temperatures. In both the covering and the partitioning models, we ...
Deep joint- semantics reconstructing hashing for large-scale unsuper- vised cross-modal retrieval. In ICCV, 2019. 1 4109 [61] Chen Sun, Austin Myers, Carl Vondrick, Kevin Murphy, and Cordelia Schmid. Videobert: A joint model for video and language representat...
To enhance the expression ability of distributional word representation learning model, many researchers tend to induce word senses through clustering, and learn multiple embedding vectors for each word, namely multi-prototype word embedding model. However, most related work ignores the relatedness among ...