Multi-target tracking without line complementation – The object detection framework for multi-target tracking has received a lot of attention in the past years. One of the applications that has been adopted in this work is multi-object tracking, which relies on a large number of target locations. However, most of the existing multi-object tracking methods treat the object locations as a feature descriptor of the target locations. In this work, we consider the task where each point with an object is seen as having a similar pose to those with a different pose. The pose of each region has to be known beforehand to be used for tracking. We propose a deep learning framework that uses a recurrent neural network (RNN) to jointly learn to learn a pose and target location descriptors. We provide two benchmark datasets, namely, a real-world database and an online and real-world dataset for the state-of-the-art and demonstrate that the network learned correctly on both datasets. The approach is evaluated in the COCO database and our method performs favorably compared to state-of-the-art systems even though our approach is very expensive, especially for the same pose.
A large body of research shows how to optimize training and testing of classification models. The aim of this work is to improve the performance of state-of-the-art classification models. Two approaches were proposed for the purpose of the research: one that uses a reinforcement learning-based model, and another one that uses a Bayesian model in a two-dimensional space. In the second approach, we are interested in learning and training a Bayesian classifier using multiple-parameter Bayesian network training. Our objectives are twofold: (1) To reduce the computational time while training a learning model, and (2) To learn a Bayesian model with a very large weight. We present a method for this purpose in the framework of two-parameter Bayesian network training. We also propose a method to learn Bayes’ density-weighted representation of the Bayes’ model.
High-performance hybrid neural network for Alzheimer disease biomarker generation
Multi-target tracking without line complementation
An Expectation-Propagation Based Approach for Transfer Learning of Reinforcement Learning Agents
A Unified Approach to Evaluating the Fitness of ClassifiersA large body of research shows how to optimize training and testing of classification models. The aim of this work is to improve the performance of state-of-the-art classification models. Two approaches were proposed for the purpose of the research: one that uses a reinforcement learning-based model, and another one that uses a Bayesian model in a two-dimensional space. In the second approach, we are interested in learning and training a Bayesian classifier using multiple-parameter Bayesian network training. Our objectives are twofold: (1) To reduce the computational time while training a learning model, and (2) To learn a Bayesian model with a very large weight. We present a method for this purpose in the framework of two-parameter Bayesian network training. We also propose a method to learn Bayes’ density-weighted representation of the Bayes’ model.