Most previous works only consider simple or weak interactions, thereby failing to model complex correlations among three or more tasks. In this paper, we propose a multi-task learning architecture with four types of recurrent neural layers to fuse information across multiple related tasks. The architecture is structurally flexible and considers various interactions among tasks, which can be regarded as a generalized case of many previous works.

In this paper, we propose a multi-task convolutional neural network that is able to share the features among tasks in a selective way. This advantage is achieved by two cross-task modules that can filter the helpful features.

In this paper, we introduce gate mechanism into multi-task CNN and propose a new Gated Sharing Unit, which can filter the feature flows between tasks and greatly reduce the interference.

N-gram based CNNs are inherently limited to proactively adapt to the transformations of features. In this paper, we introduce two modules, giving CNNs the adaptability for transformation, namely, transformable convolution and transformable pooling. Both modules combine the thoughts of two mainstream end-to-end transformable methods by adding dynamic and static deviations to sampling locations to offset the transformation of features, which is learned from current task without extra supervision. Our modules can be easily employed by other CNN models to generate new transformable convolutional networks.