Multitask Learning In Deep Neural Network And Social Media

In recent years,the ability of human collecting,storing,transmitting and processing data has been improved rapidly.Deep learning is a computer algorithm which can analyze and use data in the era of artificial intelligence eruption.Among them,the convolutional neural network has been achieved a leap in computer vision.However,on the one hand,as the depth and breadth of convolutional neural networks continue to expand,model training becomes an arduous task.On the other hand,traditional model training for object detection and image classification task depends heavily on manual labeling in computer vision.As we enter the era of big data,labeling for millions of images is very time-consuming and laborious.Therefore,this paper proposes a multitask learning algorithm based on deep network and webly supervision,which is devoted to solving the above two problems.In this paper,a visual concept learning algorithm with cardinality guided instance mining and clustered multitask refinement is proposed to detect the object of interest in the test image.In order to reduce manual annotation of data sets,this paper addresses the problem of weblysupervised visual concept learning,and develops a fully automatic algorithm using parallel text and visual corpora to mine information on the Internet.Based on the captions and object proposals,a cardinality-guided multiple instance learning algorithm is designed to establish the link between the image regions and the literal concepts.Furthermore,due to the diversity of visual concepts,we perform clustered multitask refinement on the learned concept classifiers to enhance their generalization via clustered regularization.Experiments on the challenging dataset show that the proposed method outperforms conventional supervised and weakly supervised approaches,and is even comparable to many state-of-the-art deep learning based approaches.In this paper,a layer-wise supervised neural network is proposed for face alignment.Although convolutional neural networks have achieved prominent performance in facial landmark detection in recent years.However,the training of such deep network is non-trivial due to the over-fitting problem caused by the insufficient training data and the diminishing gradients problem occurred in the backpropagation.To address these problems,we propose a multitask learning framework with supervised neural networks to jointly detect facial landmarks with a set of related tasks.On the one hand,to handle the over-fitting problem,the proposed method takes the advantage of additional task labels to train the model in a multitask learning fashion to generate a shared feature representation for high-level recognition tasks.On the other hand,in order to tackle the transparency and diminishing gradients problem,the proposed method enforces supervision to the intermediate layers of the network,augmenting the gradient signal propagated from the final layer.Experiments on public benchmarks validate the effectiveness of the proposed method.