Colloquia Announcements

Wednesday, 24th April 2019, 5:00pm, MM217

Multi-view stereo algorithms tackle a classic computer vision problem: reconstructing 3D geometry of objects using a set of images. The 3D output has applications in mapping and navigation, online shopping, 3D printing, augmented and virtual reality, video games, cultural heritage documentation, to name a few. This talk briefly describes the multi view stereo problem and provides an overview of some established and successful MVS pipelines as well as some new developments in this field.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 17th April 2019, 5:00pm, MM217

The genetic make-up of admixed populations is not homogenous. Admixture creates mosaic chromosomes of distinct ancestry that lead to a variation of allele frequencies. Due to this structure, admixed populations may have more than one ancestral background across the samples at a particular genomic region. This could contribute to the cohort-level test differences often seen in association studies. For example, the strongest risk gene identified for late-onset Alzheimer disease is ApoE. However, the risk for Alzheimer disease due to ApoE is not consistent across the populations. Individuals with the African ancestry experience less risk from ApoE than individuals of European or Asian ancestry. Thus genetic ancestral methods may play an essential role in understanding the factors contributing to the lower/higher risk effect across the populations by using admixed populations. In particular, identification of population-specific variation that influences disease could inform precision medicine initiatives, and lead to the development of ancestry-specific disease treatments. This would improve treatments, and help reduce health disparities.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 10th April 2019, 5:00pm, MM217

Deep Learning puts forward some of the most sophisticated machine learning solutions to the current state-of-art Artificial Intelligence (AI) applications including natural language processing (NLP), audio processing, computer vision, etc. In spite of all recent achievements of deep learning, it has yet to achieve semantic learning required to reason about the data. This lack of reasoning is partially imputed to the boorish memorization of patterns and characteristics from millions of training samples and ignoring the underlying, often hidden, relationships. This is the reason why mere rotations or color-inversions can easily confound very deep neural networks even though the modified images share the same semantics and structures as the original images. Over the last decade, the Database Data-mining and Multimedia (DDM) research lab has been extensively working on Knowledge Representation and Reasoning (KRR) and reasoning based deep learning problems that focus on bridging the semantic gap between low-level features and their high-level contextual meanings. We believe that the next generation of AI systems will need to have the ability to understand problems at a deeper level rather than just based on memorization of data. This talk is about one of the core methods, called Variational Autoencoders, that we use to uncover the latent relationships in the observed data. Variational autoencoders work on the principle of variational inferencing that has gained lots of attention as being the models of choice for generative models. Variational autoencoders help us uncover deep representations by mapping the observed data to latent spaces and use the latent relationships and generative modeling to infer hidden information about data.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 28th March 2019, 5:00pm, MM217

Recent advances in machine learning and artificial intelligence have dramatically redefined our everyday experience. From medical diagnosis to criminal justice, machine learning algorithms start to play an important role in our life. Unlike most machine learning applications that focus solely on data, this presentation discusses human-centered research to build personalized, interdisciplinary, creative applications. In this presentation, I will present my projects in three domains: computational music research, natural language processing and American Sign Language recognition, and machine learning for wellbeing. I will also briefly discuss the courses that I teach, including Augmented Reality and Design with AI.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 21st March 2019, 5:00pm, MM217

As Augmented Reality is gaining popularity in all fields of life, the specialized field of brain and spine surgery is sure enough to be one of the first to take advantage of this incredible technology. Imaging modalities that are currently used in the operating room are expensive, bulky, and produce harmful radiation. AR has the potential to alleviate this burden and early studies have shown its feasibility in doing so. However, there is still a tremendous amount of work that needs to be done. Current AR systems are not completely optimized for the operating room use. Collaboration and understanding between the end-user operators and the developers will ensure a productive growth of the technology. Closing the feedback loop and identifying caveats that need improvement needs to be a part of every discussion.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 14th March 2019, 5:00pm, MM217

Clinicians and researchers studying children for autism spectrum disorder (ASD) have long noticed the atypical head postures and movement patterns of children with ASD. They have used the difference in the head movement patterns between the children with ASD and those without ASD as supplementary diagnostic cues based on manual interaction and analysis. Our work provides an automatic head movement tracking, analysis, and interpretation framework for ASD diagnosis and research. Specifically, we implemented a computational framework for extracting the temporal patterns of head movement and utilizing the imbalance of temporal pattern distribution between diagnostic categories (children with or without ASD) as potential diagnostic cues. The temporal patterns are extracted from multiple motion feature dimensions and time resolutions to form a high-dimensional "big-data" feature array. Then our proposed framework identifies the temporal patterns with imbalance distributions between contrasting diagnostic categories as potential clinical diagnostic cues. Our proposed automatic framework aims to improve clinical operational efficiency while allowing human experts to focus on essential diagnostic decisions or to tackle datasets of larger scales. The proposed analysis and interpretation framework is also useful for exploring the motor movement processes of ASD-related atypical head movement for their translational integration towards therapeutic interventions.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 6th March 2019, 5:00pm, MM217

Tensor as a multi-dimensional generalization of matrix has received increasing attention in industry due to its success in personalized recommendation systems. Traditional recommendation systems are mainly based on the user-item matrix, whose entry denotes each user's preference for a particular item. To incorporate additional information into the analysis, such as the temporal behavior of users, we encounter a user-item-time tensor. Existing tensor decomposition methods are mostly established in the non-sparse regime where the decomposition components include all features. In online advertising, the ad-click tensor is usually sparse due to the rarity of ad clicks. In this talk, I will discuss a new sparse tensor decomposition method that incorporates the sparsity of each latent component to the CP tensor decomposition. In theory, in spite of the non-convexity of the optimization problem, it is proven that an alternating updating algorithm attains an estimator whose rate of convergence significantly improves those shown in non-sparse decomposition methods. The potential business impact of our method is demonstrated via an application of click-through rate prediction for personalized advertising. In the second part of the talk, I will discuss an extension of the proposed sparse tensor decomposition to handle multiple sources of tensor data. In online advertising, the users' click behavior on different ads from multiple devices forms a user-ad-device tensor, and the ad characteristics data forms an ad-feature matrix. We propose a unified learning framework to extract latent features embedded in both tensor data and matrix data. We conduct cluster analysis of advertisements based on the extracted latent features and provide meaningful insights in linking different ad industries.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 27th February 2019, 5:00pm, MM217

Developmental psychology is concerned with transformative change processes that occur over the human lifespan. Much early development occurs in the context of social interaction. New sensing technologies, combined with machine learning, may provide insight into the social and emotional development of typically developing children and those with communication disorders such as autism and deafness. Automated detection of smiling and vocal turn-taking is shedding light on the diagnosis of autism. Automated analysis of classroom movement and vocal interaction is suggesting how language development occurs among peers. The talk will consider the strengths, challenges, and future of objective measurement and modeling of child behavior.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 20th February 2019, 5:00pm, MM217

Medical Physics integrates computer science, mathematics, physics, engineering and medicine into an interdisciplinary profession in serving diagnostic imaging, radiation therapy, nuclear medicine imaging, and radiation control/protection. This presentation provides an application example of image-guided radiosurgery that demonstrates medical physics research in cancer treatment. Tracking tumor movement during the treatment is crucially important for radiation therapy. The gold standard of real-time tracking of abdominal tumors requires the use of fiducial markers, however, which usually induces complications. We hypothesized that the two-dimensional (2D) location of the lung-diagram border can be used to determine the three-dimensional (3D) location of the tumor with clinically acceptable accuracy in comparison to using fiducial markers. Using a simple two-layer ANN architecture, we built correlation models that link the lung-diaphragm border's location with the corresponding 3D location of the tumor volume. Both simulation study (proof of concept) and clinical study (validation of concept) show the feasibility of accurately predicting the tumor volume's position with the use of widely available kV imagers through machine learning without fiducial markers. This patented innovative technique has the potential to eliminate fiducial markers in the tracking of liver or other abdominal tumors.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 13th February 2019, 5:00pm, MM217

In this talk I will first provide an overview of cognitive neuroscience approaches by comparing and contrasting various invasive and non-invasive cognitive neuroscience techniques. I will then talk about using MRI to map large-scale structural and functional networks of the brain and how such approaches may be used to quantify developmental changes and identify differences in clinical conditions such as autism spectrum disorder.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 6th February 2019, 5:00pm, MM217

In the past 5 decades, rapidly increasing of computing power and development of computational fluid dynamics enable virtual simulations of various aerospace systems. Such simulations make many proof of concepts must faster at a much lower cost. This seminar will talk about the impact that high performance computing brings to aerospace engineering and the future perspective.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 30th January 2019, 5:00pm, MM217

Intervertebral disc (IVD) is the largest avascular structure in the human body and its main function is to support mechanical loading and to provide the flexibility for the spine system.Degenerative disc disease (DDD) is related to low back pain which affects more than 600 million people worldwide. One of the challenges in modeling DDD is that biological, chemical, electrical, and mechanical events in IVD are nonlinearly coupled. It is important to understand the biophysics and pathophysiology in IVD in order to develop a model successfully. A multiscale and multi-physics model for IVD has been developed recently. In this model, nonlinear interactions among biological (cell activity), chemical (osmolarity and pH), electrical (charges on matrix and solutes), and mechanical (loading and tissue swelling) events in the IVD are considered. Numerical results are obtained by solving a dozen of partial differential equations using a finite element method. Applications of this model to simulating degenerative progression of IVDs due to poor nutrition supply will be presented. In-silico clinical trials of cellular therapies for dis repair will also be discussed. The study provides not only new insights into the mechanisms of disc degeneration, but also new diagnostic means for disc degeneration.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Friday, 25th January 2019, 10:00am, UB330D

Incorporating a dynamic kick engine that is both fast and effective is pivotal to be competitive in one of the world'???'s biggest AI and robotics initiative: RoboCup. Using the NAO robot as a testbed, we developed a dynamic kick engine that can generate a kick trajectory with an arbitrary direction without prior input or knowledge of the parameters of the kick. The trajectories are generated using cubic splines, sextic polynomials, or cubic Hermite splines, and are executed while the robot is dynamically balancing on one foot. When the robot swings the leg for the kick motion, unprecedented forces might be applied on the robot. To compensate for these forces, we developed a Zero Moment Point (ZMP) based preview controller that minimizes the ZMP error. Although a variety of kick engines have been implemented by others, there are only a few papers on how kick engine parameters have been optimized to give an effective kick. Parameters such as kick configuration, limit of the robot, or shape of the polynomial can be optimized. We propose an optimization framework based on the Webots simulator to optimize these parameters. Experiments of the physical robot show promising results.

This is a Department of Computer Science MS Defence.

Wednesday, 23rd January 2019, 5:00pm, MM217

This presentation will discuss our three-layer dynamic data-driven application systems-based framework for the control and planning of smart energy systems. We will highlight the issues of network topologies in smart grids and then analyze the operations management of smart energy systems. Next, we will provide the details of our multi-fidelity simulation model for timely monitoring and system assessment. Last, we will briefly discuss the real-time decision-making module that adapts the operation plan from the optimization model using robust demand response, network reconfiguration and/or operation rescheduling models based on dynamically changing data from the smart grid. We will explain the results of the investigated framework based on our computational tests on IEEE-9, IEEE-30, IEEE-118 bus systems and the designed smart grid for the City of Coral Gables. The presentation will conclude with a discussion on the future for the smarter grids.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm outside MM217.

Wednesday, 16th January 2019, 5:00pm, MM217

Automation means reliability. Reducing human factor in software development increases project's overall quality. In this presentation we will introduce ways to increase automation on 7 different levels iteratively by adding conceptual git "lanes" (branches). It is a concept, that can be applied widely regardless the size of the project, number of programmers, language or environment. The only question is how much automation and thus reliability does your project need.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30pm in MM217.

Previous Colloquia Announcements