Learning to Rank for Passive Perception in Unlabeled Data – We propose a novel method for classification tasks, by first finding a score of items that the target (or a subset of items) is interested in. This is a very challenging task, and our method is motivated by the following question: how to predict the target of an item? The goal of this work is to infer its value of a set of items and use that value to generate a ranking metric. We propose an algorithm that learns a rank-based value which serves as a baseline to improve classification accuracy. The method is applied to two challenging categories, namely, text classification and video analysis. Our experiments demonstrate the effectiveness of using the rank-based value to improve classification performance.
We propose an effective and robust method for the face recognition task at hand. It extends the approach from the face recognition task of a user to that of a human. Our method uses a deep neural network to discover the region of interest and the global context of the region. Unlike previous approaches for recognizing faces in images, we focus on the region of interest in order to learn how to predict the identity function in the region. Our method learns the global context by learning a new face identity function that maps a set of the face instances together. We use this new identity function to predict the pose for a given face instance as well as pose prediction metrics such as SVMs. Our method outperforms state-of-the-art human-level face recognition methods on the BLEU dataset with an accuracy of 97% that is comparable to human experts, which are more challenging to achieve in the face recognition community.
A Novel Approach for Automatic Image Classification Based on Image Transformation
A Deep Interactive Deep Learning Framework for Multi-Subject Crossdiction with Object Segmentation
Learning to Rank for Passive Perception in Unlabeled Data
Clustering with Missing Information and Sufficient Sampling Accuracy
Deep Learning Approach to Robust Face Recognition in Urban EnvironmentWe propose an effective and robust method for the face recognition task at hand. It extends the approach from the face recognition task of a user to that of a human. Our method uses a deep neural network to discover the region of interest and the global context of the region. Unlike previous approaches for recognizing faces in images, we focus on the region of interest in order to learn how to predict the identity function in the region. Our method learns the global context by learning a new face identity function that maps a set of the face instances together. We use this new identity function to predict the pose for a given face instance as well as pose prediction metrics such as SVMs. Our method outperforms state-of-the-art human-level face recognition methods on the BLEU dataset with an accuracy of 97% that is comparable to human experts, which are more challenging to achieve in the face recognition community.
Leave a Reply