Department of Computer Engineering (2014 - Present)
Postdoc at German Research Center for AI (DFKI)
, Bremen University, Bremen, Germany
Computer Engineering - Artificial Intelligence
, Welsh Cardiff Card, England
Artificial Intelligence and Robotics
Electrical and Computer Engineering, University of Tehran, Tehran, Iran
Computer Hardware Engineering
Computer Engineering, Sharif University of Technology, Tehran, Iran
Video contents have variations in temporal and spatial dimensions, and recognizing human actions requires considering the changes in both directions. To this end, convolutional neural networks (CNNs) and recurrent neural networks (RNNs) and their combinations have been used to tackle the video dynamics. However, a hybrid architecture usually results in a more complex model and hence a greater number of parameters to be optimized. In this study, we propose to use a stack of gated recurrent unit (GRU) layers on top of a two-stream inflated convolutional neural network. Raw frames and optical flow of the video are processed in the first and second streams, respectively. We first segment the video frames in order to be able to track the video c
Introduction: Epilepsy is one of the mos t common brain disorders that greatly affect patients’ life. However, early detection of seizure attacks can significantly improve their quality of life. In this s tudy, we evaluated a deep neural network to learn robus t features from electroencephalography (EEG) signals to automatically detect and predict seizure attacks. Materials and Methods: The architecture consis ts of convolutional neural networks and long short-term memory networks. It is designed to simultaneously capture spectral, temporal, and spatial information. Moreover, the architecture does not rely on explicit channel selection algorithms. The method is applied to the Children’s Hospital of Bos ton-Massachusetts Ins titute of Te
In this paper, a new technique has been designed to capture the outline of 2D shapes using cubic Bezier curves. The proposed technique avoids the traditional method of optimizing the global squared fitting error and emphasizes the local control of data points. A maximum error has been determined to preserve the absolute fitting error less than a criterion and it administers the process of curve subdivision. Depending on the specified maximum error, the proposed technique itself subdivides complex segments, and curve fitting is done simultaneously. A comparative
Despite considerable enhances in recognizing hand gestures from still images, there are still many challenges in the classification of hand gestures in videos. The latter comes with more challenges, including higher computational complexity and arduous task of representing temporal features. Hand movement dynamics, represented by temporal features, have to be extracted by analyzing the total frames of a video. So far, both 2D and 3D convolutional neural networks have been used to manipulate the temporal dynamics of the video frames. 3D CNNs can extract the changes in the consecutive frames and tend to be more suitable for the video classification task, however, they usually need more time. On the other hand, by using techniques like tiling
Hand gesture recognition from videos has more challenges compared with still images due to the difficulty of representing temporal features and longer training times especially in real-time applications. In this paper, we investigated the potential of 2D over 3D CNN's for representing temporal features and classification of hand gestures in videos. We mapped the frame sequence of hand gestures to a chronological tiled pattern in order to capture the dynamics of the hand movement in a single frame. Then, using 2D CNN's, we generated feature vectors containing both special and temporal features. Additionally, we proposed a new approach for fusing data and predictions through a two-stream architecture to exploit depth information. The effects
Deep networks have recently achieved great success in feature learning problem on various computer vision applications. Among different approaches in deep learning, unsupervised methods have attracted a lot of attention particularly to problems with limited training data. However, compared with supervised methods, unsupervised deep learning methods usually suffer from lower accuracy and higher computational time. To deal with these problems, we aim to restrict the number of connections between successive layers while enhancing discriminatory power using a data-driven approach. To this end, we propose a novel deep multi-view ensemble model. The structure of each layer is composed of an ensemble of encoders or decoders and mask operations. Th
Introduction: Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that occurs in the early years of life and is characterized by social impairment, verbal and non-verbal communication difficulties as well as stereotypical behaviors. Rehabilitating autistic children at the early stages of growth, in which their brain is highly flexible, yields to enhanced treatment process and provides the chance of utilizing their talents. In other words, late detection and treatment will leave these children's behavior unchanged until adulthood. Considering the role of eyes, as one of the most valuable sources of information in social interactions and the different patterns of eye behaviors in autistic children in response to social stimuli, th
Sequencing large number of candidate disease genes which cause diseases in order to identify the relationship between them is an expensive and time-consuming task. To handle these challenges, different computational approaches have been developed. Based on the observation that genes associated with similar diseases have a higher likelihood of interaction, a large class of these approaches relay on analyzing the topological properties of biological networks. However, the incomplete and noisy nature of biological networks is known as an important challenge in these approaches. In this paper, we propose a two-step framework for disease gene prioritization:(1) construction of a reliable human FLN using sequence information and machine learning
The structure of online social networks such as Facebook is continuously changing. Phenomena such as birth, growth, contraction, split, dissolution, and merging with other communities are issues that occur in the communities of online social networks over time. However, characteristics of the consecutive time slots of these networks depend on each other, and independent investigation of each time slot is not efficient for detecting communities in terms of execution time due to the big size of data in each time slot. In order to detect the changes in communities over time, there is a need for algorithms that can detect communities incrementally with proper precision. In this paper, we propose an unsupervised machine learning algorithm for in
no record found