Background and Obejctives: Multi-Task learning is a widespread mechanism to improve the learning of Multiple objectives with a shared representation in one deep neural network. In Multi-Task learning, it is critical to determine how to combine the Tasks loss functions. The straightforward way is to optimize the weighted linear sum of Multiple objectives with equal weights. Despite some studies that have attempted to solve the realtime Multi-person pose estimation problem from a 2D image, major challenges still remain unresolved. Methods: The prevailing solutions are two-stream, learning two Tasks simultaneously. They intrinsically use a Multi-Task learning approach for predicting the confidence maps of body parts and the part affinity fields to associate the parts to each other. They optimize the average of the two Tasks loss functions, while the two Tasks have different levels of difficulty and uncertainty. In this work, we overcome this problem by applying a Multi-Task objective that captures Task-based uncertainties without any additional parameters. Since the estimated poses can be more certain, the proposed method is called “CertainPose”. Results: Experiments are carried out on the COCO keypoints data sets. The results show that capturing the Task-dependent uncertainty makes the training procedure faster and causes some improvements in human pose estimation. Conclusion: The highlight advantage of our method is improving the realtime Multi-person pose estimation without increasing computational complexity.

Named Entity Recognition is a challenging Task, specially for low resource languages, such as Persian, due to the lack of massive gold data. As developing manually-annotated datasets is time consuming and expensive, we use a MultiTask learning (MTL) framework to exploit different datasets to enrich the extracted features and improve the accuracy of recognizing named entities in Persian news articles. Highly motivated auxiliary Tasks are chosen to be included in a deep learning based structure. Additionally, we investigate the effect of chosen datasets on performance of the model. Our best model significantly outperformed the state of the art model by, according to F1 score in the phrase level.

Emotions are a part of everyday communications of people and one of the important elements of human nature. We can distinguish a person’s emotions from some outcome behaviors such as speech, facial expression, body movements, and gestures. Another outcome behavior is his/her grammar and written method that reflects the inner states of the person. Since people are nowadays more likely to use textual tools to make the connection, emotion extraction from the text has attracted much attention. The majority of methods in this regard consider emotion extraction from the text as a classification problem. Therefore, most studies depend on a huge number of handcrafted features and are done on feature engineering to enhance the classification performance. Considering that a text may include more than one emotion that only one of them is text dominant emotion, we model the emotion extraction problem as a Multi-label classification problem by removing the fixed boundaries of emotions. Next, we recognize all the existing emotions in the sentence and in dominant emotion. Our goal is to achieve a better performance only with minimal feature engineering. To this end, we propose a hybrid deep learning model that benefits both CNN and RNN deep models. The experiments are done on a Multi-label dataset including 629 sentences with eight emotional categories. Based on the results, our proposed method shows a better performance (about 0.12%) compared with available Multi-label learning methods (e.g., BR, RAKEL, and MLkNN).

With the increasing amount of accessible textual information via the internet, it seems necessary to have a summarization system that can generate a summary of information for user demands. Since a long time ago, summarization has been considered by natural language processing researchers. Today, with improvement in processing power and the development of computational tools, efforts to improve the performance of the summarization system is continued, especially with utilizing more powerful learning algorithms such as deep learning method. In this paper, a novel Multi-lingual Multi-document summarization system is proposed that works based on deep learning techniques, and it is amongst the first Persian summarization system by use of deep learning. The proposed system ranks the sentences based on some predefined features and by using a deep artificial neural network. A comprehensive study about the effect of different features was also done to achieve the best possible features combination. The performance of the proposed system is evaluated on the standard baseline datasets in Persian and English. The result of evaluations demonstrates the effectiveness and success of the proposed summarization system in both languages. It can be said that the proposed method has achieve the state of the art performance in Persian and English.