Colloquia Announcements

Wednesday, 22nd February 2017, 5:00pm, MEA202

Faisal Sikder
Department of Computer Science
University of Miami

will present

Semi-Automatic Extraction of Training Examples from Sensor Readings
for Fall and Activity Identification

While inexpensive wearable motion-sensing devices have shown great promise for fall detection and human activity monitoring, two major challenges still exist and have to be solved: 1) a framework for the development of firmware, and 2) software to make intelligent decisions.  We address both the problems. In this talk, we show our proposed generic framework for developing firmware. We also demonstrate that the k-means clustering algorithm can semi-automatically extract training examples from motion data. Moreover, we discuss about several one- and two level classification networks combinations of neural networks and softmax regression to monitor non-fall activities and to detect fall events. We also illustrate how the datasets for training and testing have been collected using the devices we assembled with four off-the-shelf components. This work advances the state-of the-art for development and training of wearable devices for monitoring non-fall activities and detecting fall events.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Friday, 17th February 2017, 2:30pm, Cox 42

Dr. Houssam Nassif

will present

Diversifying Amazon Recommendation

Dr. Nassif will be presenting two recently published papers on diversifying Amaz on recommendations: "Diversifying Music Recommendations" (ICML'16 Workshop) uses submodular diversity to significantly improve Prime Music App recommendations quality and user engagement.  "Adaptive, Personalized Diversity for Visual Discovery" (RecSys'16 Best Short Paper Award) describes Amazon Stream's seasonal, personalized and diversified recommendation framework. Amazon Stream (, a new website for fashion discovery, uses Bayesian regression to score products, balances exploration and exploitation, applies submodularity to diversify recommendations, and learns seasonal and personalized weights to produce the final recommended personalized stream.

This is another in the Department of Computer Science Seminar series, presented jointly with the Department of Neuroscience and the Center for Computational Science.

Wednesday, 15th February 2017, 5:00pm, UB230

Dr. Athena Hadjixenofontos
Center for Computational Science
University of Miami

will present

The Many Forces of Nature: Lessons on the Genetic Architecture of Multiple Sclerosis
from the Isolated Population of Sardinia

Complex diseases are the most prevalent causes of death and disability in developed countries. Susceptibility to complex diseases is determined by the cumulative results of hundreds to thousands of genetic variants, as well as environmental exposures. In the last twenty years, numerous large scale projects have been undertaken to identify the genetic variants that underlie susceptibility to a number of complex diseases, including multiple sclerosis. Success has been limited: for multiple sclerosis, 110 genetic variants have been associated with the disease in outbred Caucasian populations, and it is estimated that hundreds to thousands more remain in the dark. The parameters that define the genetic architecture of a complex disease extend beyond the number of variants that underlie susceptibility, and include their effect sizes, population frequencies and whether or not they act additively. The limited success in forming a complete picture has led to the exploration of alternative hypotheses. Some of the alternative hypotheses target areas of the genetic architecture that traditional experiments are not designed to address. In this talk I will recount our contributions to understanding the genetic architecture of multiple sclerosis through the study of the isolated population of Sardinia. I will then lay out a set of future directions that are designed to challenge our assumptions about the hidden areas of genetic susceptibility, through forward-in-time genetic simulations. The completion of the proposed experiments will explain the reasons for our limited success in mapping disease variants and position us to choose the methods that can be effective in uncovering the remaining effects.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 8th February 2017, 5:00pm, UB230

Dr. Fabian Soto
Department of Psychology
Florida International University

will present

Extending Multidimensional Signal Detection Theory to Study the Independence of Brain Representation

A common goal in visual neuroscience is to determine whether some stimulus dimensions (e.g., shape and spatial information, different shape properties, face expression and identity) are processed and represented independently from others. Such representation can be extremely useful; for example, if most objects can be recognized on the basis of a few shape dimensions, then representing those shape dimensions independently from any other visual information would allow fast object learning, by focusing attention only on the relevant dimensions and ignoring the irrelevant information. Furthermore, such learning would generalize broadly to any new object image from which the relevant shape dimensions can be extracted, regardless of how different this new image is from the training images. Such fast, generalizable learning is easily observed in people, but poorly understood.  Unfortunately, analytical tools and models tailored specifically to study the independence of brain representations have not been developed.  Without such tools, cognitive neuroscientists have resorted to proposing a multiplicity of operational tests of independence, each a small modification of traditional analyses adapted to measure a vaguely defined construct. Unsurprisingly, this research strategy has yielded contradictory results in most areas. This talk summarizes recent work aimed at solving this problem by extending General Recognition Theory, a multidimensional version of signal detection theory, to the study of independence of brain representations, including neuroimaging and neurophysiology.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Wednesday, 1st February 2017, 5:00pm, UB506

Dr. Geoff Sutcliffe
Department of Computer Science
University of Miami

will present

Automated Reasoning for the Dialetheic Logic RM3

A dialiethic logic allows formulae to be true, or false, or (differently from classical logic) both true and false, and the connectives are interpreted in terms of these three truth values. Consequently some inferences rules of classical logic are invalid in RM3, and some theorems of classical logic are not theorems of RM3. An automated theorem prover for RM3 has been developed, based on translations of RM3 formulae to classical first-order order logic, and use of an existing first-order theorem prover to reason over the translated formulae. Examples and results are provided to highlight the differences between reasoning in classical logic and in RM3.

This is another in the Department of Computer Science Pizza Seminar Series. Refreshments will be served at 4:30 p.m. in the reception area of the 3rd floor of Ungar.

Friday, 27th January 2017,2:30pm, Ungar 230

Dr. Hosna Jabbari
Ingenuity Lab
University of Alberta

will present

In Silico Tool to Improve Efficacy of Gene Therapy

With the amount of genomic data produced every day, advances in medical sciences and development of new gene modification tools, a revolution in medicine is expected.Recent approval of the first gene therapies by FDA is a leap towards this revolution. Computational methods provide unique opportunities to realize this revolution by providing both an inexpensive framework (in terms of cost, time and safety) to explore the complex biological systems of diseases, and a reduced search space for treatment options. In this talk, I will describe an example of such framework for treatment of Duchenne muscular dystrophy through gene therapy, and highlight some of the challenges and future opportunities.

This is another in the Department of Computer Science Seminar series.

Wednesday, 25th January 2017, 5:00pm, Ungar 230

Dr. Tatiana Engel
Department of Bioengineering
Stanford University

will present

Discovering Dynamic Computations in the Brain from Large-Scale Neural Recordings

Neuronal responses and behavior are influenced by internal brain states, such as arousal, vigilance, or task context. Ongoing variations of these internal states affect global patterns of neural activity, giving rise to apparent variability of neuronal responses to sensory stimuli, from trial to trial and across time within single trials. Demultiplexing these endogenously generated and externally driven signals proved difficult with traditional techniques based on trial-averaged responses of single neurons, which dismiss neural variability as noise. In this talk, I will describe my recent work leveraging multi-electrode neural activity recordings and computational models to uncover how internal brain states interact with goal-directed behavior. I will show that ensemble neural activity within single columns of the primate visual cortex spontaneously fluctuates between phases of vigorous (On) and faint (Off) spiking. These endogenous On-Off dynamics, reflecting global changes in arousal, are also modulated at a local scale during spatial attention and predict behavioral performance. I will also demonstrate that these On-Off dynamics provide a single unifying mechanism that explains general features of correlated variability classically observed in cortical responses (e.g., changes in neural correlations during attention). I will conclude by sketching out a roadmap for developing a general theory that will allow us to discover dynamic computations from large-scale neural recordings and to link these computations to behavior.

This is another in the Department of Computer Science Seminar series.

Monday, 23rd January 2017, 2:15pm, Ungar 230

Dr. Catie Chang
National Institute of Health

will present

Uncovering New Dimensions of Human Brain Function from fMRI Data

Functional magnetic resonance imaging (fMRI) is a powerful technique for human neuroscience. The richness and complexity of fMRI data present exciting challenges at the interface between computation and neuroscience, and require innovative data analysis methods together with deeper understanding of the neural and physiological basis of fMRI signals. I will describe my studies revealing features of brain function embedded in the dynamics of intrinsic brain networks. I will also discuss how, by integrating fMRI with electrophysiological, behavioral, and heart rate data, we uncovered components of fMRI dynamics related to vigilance and autonomic activity and developed a data-driven approach for detecting vigilance fluctuations in fMRI scans. These studies highlight ways in which previously unexplored dimensions of systems-level brain activity may be extracted from fMRI signals, and open new directions for neuroimaging biomarkers in health and disease.

This is another in the Department of Computer Science Seminar series.

Thursday, 19th January 2017, 8:30am, Ungar 230

Dr. Xuan Guo
Oak Ridge National Laboratory

will present

Multi-omics Data Analyses via High-performance Computing for
Complex Biological Systems

Systems biology aims to model complex biological interactions at the system level by integrating information from interdisciplinary fields using a holistic perspective approach. Emerging high-throughput omics technologies promote current biology research into the age of systems biology and also raise huge challenges in multi-omics data integration, modeling, and systems-level analyses. High-performance computing techniques are promising to overcome the limits posed by conventional methods to the mining and exploration of large amounts of multi-omics data. In this talk, I will explore how to analyze high-throughput multi-omics data to better understand complex diseases and microbial communities. I will present several parallel algorithms and high-performance computing framework that are broadly applicable for the analyses of large data and complex biological systems.

This is another in the Department of Computer Science Seminar series.

Thursday, 12th January 2017, 8:30am, Ungar 230

Dr. Zheng Wang
University of Southern Mississippi

will present

The Complex Systems of Protein Domain Co-Occurrences and
the Three-Dimensional Structure of the Genome

This presentation will start from a biological network named protein domain co-occurrence network that has not been fully studied but important, in which each node represents a protein domain, and if two domains co-exist in a protein, an edge is created to connect them. After repeating this procedure for all of the proteins of a species, a network of the species is created that contains species-specific biological signatures. Dr. Wang's research verifies that this type of network is scale-free network. Other topological properties including the shortest path distribution and the clustering coefficient distribution will also be shown in the presentation. A robustness test shows that this network is vulnerable to attack (remove nodes starting from the ones with the highest degree value) and robust to failure (remove nodes randomly).  Dr. Wang has successfully used this type of network to predict protein and domain functions using neighbor-counting, biological function enrichment, and a machine learning algorithm based on the network topology. Another experiment of Dr. Wang has achieved an accuracy of 93.43% when applying this network to infer the phylogenetic relationships of 398 single-chromosome prokaryotic species using a graph alignment algorithm.

After showing these research outcomes, the presentation will continue to a newly-emerged and potentially ground-breaking research topic, that is, the three-dimensional (3D) structure of the genome. The 3D conformations of healthy human chromosomes will be shown, followed by the comparisons of the intra-chromosomal spatial proximities of healthy, leukemia, and lymphoma human B-cells or cell-lines. The inter-chromosomal (between different chromosomes) spatial proximities and the chromosome translocation between chromosome 11 and 14 (a segment of chr.11 is exchanged with a segment of chr.14) in leukemia will also be illustrated in the presentation. Furthermore, a novel type of biological complex network will be introduced, which can indicate the 3D spatial proximities among biological components including protein coding genes, transcription factors, and lncRNAs. Using this type of novel complex network, Dr. Wang's lab has recently reconstructed the 3D structure of the chromosome X of mouse embryonic stem cells mapped with the localization intensities of Xist transcripts (an lncRNA that can inactivate the entire chromosome X by altering its 3D structure). Dr.  Wang will present their algorithms and the corresponding reconstructed 3D structures in the resolutions of 500K base pair and, more excitingly, 40K base pair that has reached the gene level.

After that, Dr. Wang's latest research in topological domains in mammalian genomes, which are recently believed to be the structural and functional units of the genome, will be briefly discussed. Particularly, Dr. Wang will show a novel structural measurement for topological domains and its correlations with genetic and epigenetic features.  Together with Dr. Wang's internationally-recognized research in protein function prediction and protein model quality assessment using deep networks (stacked denoising autoencoders, SdAs), a newly-designed bioinformatics course proposed in a pending NSF CAREER proposal will also be briefly mentioned. Grant proposal ideas with computer science, physics, biological sciences, biostatistics, and medical school will be proposed throughout the presentation.

This is another in the Department of Computer Science Seminar series.

Previous Colloquia Announcements