I am an ECE Ph.D. student at the University of Houston, currently working as a research assistant at HULA Lab and fellow at the Center for Advanced Computing and Data Science (CACDS). My research is mainly focused on Machine Learning, Deep Learning, Reinforcement Learning and their applications in Computer Vision and Natural Language Processing. To find more information about my research topics and other experiences, see my Resume.
Developing neural networks are likely to be endowed with functionally important variability across context, age, gender and other, as yet unknown, variables. To examine how neural responses vary across such factors, we assayed neural activity of children at baseline and while they played a videogame at the Children’s Museum of Houston.
Joint work with University of Houston Brain-Machine Interface Systems Team.
This paper investigates the use of capsule networks (CapsNets) as an alternative to CNNs. We show that CapsNets significantly outperforms CNNs when the number of training samples is small. To increase the computational efficiency, our paper proposes a consistent dynamic routing mechanism that results in 3x speedup of CapsNet. Finally, we show that the original image reconstruction method of CapNets performs poorly on lung nodule data. We propose an efficient alternative, called convolutional decoder, that yields lower reconstruction error and higher classification accuracy.
Joint work with Hien Van Nguyen.
we investigate the effectiveness of deep learning techniques for lung nodule classification in computed tomography scans. Our paper also proposes a novel framework for rapidly adapting deep networks to the radiologists' feedback, or change in the data due to the shift in sensor's resolution or patient population. Finally, we propose using inconsistency density function, computed by a recurrent network, as a way to discover potentially noisy labels in lung data.
In this paper, an architecture based on Long Short-Term Memory Networks has been proposed for the text-independent scenario which is aimed to capture the temporal speaker-related information by operating over traditional speech features. An LSTM architecture is trained to create a discrimination space for validating the match and non-match pairs for speaker verification.
Joint work with Mohammad Najarian.
Most of existing computer aided diagnosis (CAD) systems follow a rigid paradigm where the classifier’s decision function is optimized during the training phase, and fixed during the test phase. These systems are often perceived as unfriendly as they do not allow clinicians to provide input. They are also unable to cope with the perpetual changes in data distribution caused by different sensing technologies, imaging protocols, and patient populations. To address these shortcomings, this paper proposes a novel CAD model capable of incorporating expert domain knowledge in real-time to improve its decision function.
Immunotherapy is a type of cancer treatment that boosts the body's natural defenses to fight cancer. Detecting and Tracking the effector and target cells is a demanding task which can push the immunuterapy research forward.
In this project, we designed a pipeline to detect cells using Faster RCNN, and a mixture of Capsule Network and Bidirectional LSTM to classify the cells and check if each cell is dead/alive during time.
Joint work with Hengyang Lu.
We're working on a novel 3D fully-convolutional network for automatic segmentation of anatomical structures on 3D CT images.
The proposed architecture accomplishes an end-to-end, voxel-wise multiple-class classification to map each voxel in a CT image directly to an anatomical label using very few samples. This is often a necessity in biomedical domain where data collection and annotation is expensive and time-consuming.
Joint work with MD Anderson Cancer Center.
Popular deep learning models created today produce a point estimate but not an uncertainty value. In classification models, the probability vector obtained at the end of the pipeline (the softmax output) is often erroneously interpreted as model confidence. However, a model can be uncertain in its predictions even with a high softmax output. Understanding if the model is under-confident or falsely over-confident can help reasoning about the model.
In this project, we introduce "Sampling DropConnect" as a practical method to estimate model uncertainty: what the model doesn't know due to lack of training data.
Joint work with Hien Van Nguyen.