Visual Tracking via Superpositional Matching – We propose a new framework for video prediction and visual comparison that combines deep learning and deep learning with convolutional neural network (CNN) using semi-supervised learning. Our framework aims to use CNNs to improve the accuracy of video prediction as well as improve the quality of comparisons. We show that Deep CNNs outperforms CNNs in the task of video ranking using supervised learning. Further, we show that CNNs can boost performance by utilizing the semantic relationship between videos. Finally, we provide a detailed analysis of how the proposed method converges to a new state of the art performance.
Anomaly detection provides a means for automatic and interpretable diagnosis of real-world events. We consider the problem of detecting anomalous systems in two aspects: (1) detecting the presence of anomalous devices and (2) detecting the presence of anomalous systems. Our approach proposes to first detect and assess any anomalous system and then apply a predictive model to infer the anomaly. Based on our proposed approach, we identify anomaly systems as well as a system that is the result of the system anomalous. To evaluate the predictive models, we develop a method of combining the predictive models with the hypothesis testing model and create a new anomaly detection model for each problem that is capable of detecting or recognizing anomalous systems with high probability. The method has been applied to a series of real-world datasets for which it showed similar or higher accuracy than the state-of-the-art methods.
Concise and Accurate Approximate Reference Sets for Sequential Learning
Estimating Linear Treatment-Control Variates from the Basis Function
Visual Tracking via Superpositional Matching
Fast FPGA and FPGA Efficient Distributed Synchronization
Towards a Deep Multitask Understanding of Task DynamicsAnomaly detection provides a means for automatic and interpretable diagnosis of real-world events. We consider the problem of detecting anomalous systems in two aspects: (1) detecting the presence of anomalous devices and (2) detecting the presence of anomalous systems. Our approach proposes to first detect and assess any anomalous system and then apply a predictive model to infer the anomaly. Based on our proposed approach, we identify anomaly systems as well as a system that is the result of the system anomalous. To evaluate the predictive models, we develop a method of combining the predictive models with the hypothesis testing model and create a new anomaly detection model for each problem that is capable of detecting or recognizing anomalous systems with high probability. The method has been applied to a series of real-world datasets for which it showed similar or higher accuracy than the state-of-the-art methods.