Deep neural network training with hidden panels for nonlinear adaptive filtering – We present a novel network-model-guided approach to learning to-watch video data. Through a deep learning method that learns an encoding function for each frame of the video sequence, the network is trained with an eye-tracking strategy on the sequence, which is then used to predict future frames of the relevant sequence. Our model uses a multi-sensor convolutional neural network that can learn the visual attribute of the input video. We propose a novel framework, called ConvNet-CNN, to learn the visual attribute of the input video from multi-view regression. We show that our method outperforms three state-of-the-art CNN architectures on various datasets.

We present a computationally efficient algorithm that approximates the Fisher information density using maximum likelihood. The method is applicable to any Bayesian Nonparametric model and is scalable. The proposed method achieves state-of-the-art accuracies on a range of MNIST benchmark data sets. We use it to evaluate its performance in the modeling of natural images, which show it can be used for efficient estimation of the Fisher information density.

Automated segmentation of the human brain from magnetic resonance images using a genetic algorithm

Generative Autoencoders for Active Learning

Deep neural network training with hidden panels for nonlinear adaptive filtering

Variational Approximation via Approximations of Approximate Inference

The Bayesian Nonparametric model in Bayesian NetworksWe present a computationally efficient algorithm that approximates the Fisher information density using maximum likelihood. The method is applicable to any Bayesian Nonparametric model and is scalable. The proposed method achieves state-of-the-art accuracies on a range of MNIST benchmark data sets. We use it to evaluate its performance in the modeling of natural images, which show it can be used for efficient estimation of the Fisher information density.