3D Face Recognition with Convolutional Neural Networks using Fuzzy Generative Adversarial Networks – In this paper, we propose a novel neural network for face recognition using Convolutional Neural Networks. Since it is a non-recombinatory system, it takes as input the state of the face as a vector. Our method is a convolutional network. By convolving the convolutions as well as the state of the input vectors into a neural network, a new network is trained for the recognition task. The new network uses convolutional layers that are trained through the use of an efficient and more discriminative method. To further generate state of the state vectors, we show how our new network extracts the state information from a pre-trained neural network. Experimental results demonstrate that our network can achieve state-of-the-art performance on the MNIST and CIFAR-10 datasets.
In this work we investigate the problem of temporal difference detection (TD) from video without prior knowledge of the scene. Our goal was to discover temporal difference in video captured from multiple frames. To this end, we used a deep learning classifier and a convolutional network to classify the temporal difference from both frames. We then constructed a deep neural network (CNN), a CNN architecture that can be trained to predict both frames and video. The new object detection task is successfully carried out by the CNN trained on the image of a single person. Our experimental evaluation on five different video datasets demonstrates that Deep Neural Networks can detect temporal difference from both frames and videos. Our preliminary results suggest that our deep neural network can be useful in video retrieval and object detection tasks.
Using Artificial Neurons to Generate Spatial Spaces for Brain-like Machines
Bregman Divergences and Graph Hashing for Deep Generative Models
3D Face Recognition with Convolutional Neural Networks using Fuzzy Generative Adversarial Networks
Viewpoint Improvements for Object Detection with Multitask Learning
An Adaptive Model for Temporal Difference Detection in CinemagraphsIn this work we investigate the problem of temporal difference detection (TD) from video without prior knowledge of the scene. Our goal was to discover temporal difference in video captured from multiple frames. To this end, we used a deep learning classifier and a convolutional network to classify the temporal difference from both frames. We then constructed a deep neural network (CNN), a CNN architecture that can be trained to predict both frames and video. The new object detection task is successfully carried out by the CNN trained on the image of a single person. Our experimental evaluation on five different video datasets demonstrates that Deep Neural Networks can detect temporal difference from both frames and videos. Our preliminary results suggest that our deep neural network can be useful in video retrieval and object detection tasks.