Statewide Users Group Conference Minutes - April 5, 2018

Wednesday, April 4th 

Thursday, April 5th 

Keynote Address: 
Andrew Siegel, Argonne Director of Application Development

Andrew Siegel is a senior scientist at Argonne National Laboratory, with appointments in both the Mathematics/Computer Science and Nuclear Engineering divisions. For the past decade, he has led Argonne's program in advanced reactor modeling and simulation. His research has focused on developing improved methods to model the physics of advanced reactors, including mixing, neutron/fluid coupling, and innovative computational approaches to stochastic methods for neutron transport. 

Andrew has taught more than 60 courses at the University of Chicago over the past 20 years, including high-performance computing, numerical methods, stochastic simulation, and computer architecture. He holds an undergraduate degree from the University of Chicago and a Ph.D. in astrophysics from the University of Colorado Boulder. For the past 9 years, Andrew has also held the position of Resident Dean of Burton-Judson at the University of Chicago.

Flash Talk and Poster Session Information

Flash Talk Session Winner:

Lifeng Jin, Graduate Student at The Ohio State University
"Unsupervised Depth-bounded Grammar Induction Model for PCFG with Inside-sampling"
Grammar acquisition or grammar induction for natural languages has been of interest to linguists and cognitive scientists for decades. Unfortunately, previous attempts converged to weak modes of a very multimodal distribution of grammars. Depth-bounding a grammar has been a popular technique for applying cognitively motivated restrictions to grammar induction to limit the search space of grammars. This work introduces a Bayesian depth-bounded grammar induction model with inside-sampling (DIMI) from raw text. Several analyses are performed in this work showing that depth-bounding is indeed effective in limiting the search space of the inducer. Results are also presented for successful unbounded PCFG induction as well as bounded induction on three different languages showing that our model is able to produce parse trees better than or competitive with state-of-the-art constituency grammar induction models in terms of parsing accuracy.
 

Flash Talk Session Runner-Up:

Jorge Torres, Graduate Student at The Ohio State University
"The role of HPC in the radio-detection of astrophysical neutrinos"
The radio-detection of neutrinos opens a window to unravel mysteries of the most energetic particles that arrive to Earth from space: astrophysical neutrinos, which can help us to answer open questions in fundamental physics, astrophysics and cosmology. Understanding the physics behind radio detection, designing experiments and analyzing experimental data often relies on the use of high-performance computing (HPC). In this talk, we will present how HPC has helped us, and will continue to play an important role, in our main goal: the detection of astrophysical neutrinos.
 

Poster Session Winner:

Adriaan Riet, Graduate Student at Case Western Reserve University
"Enhanced Diffusion in an MgO Grain Boundary Through Molecular Dynamics Simulations "
Grain boundary diffusion could be the driving mechanism by which metals in the core mix within the mantle. Understanding the magnitude and limits of diffusion can yield insight into the composition of the core mantle boundary, but the conditions present in the inner earth are difficult to achieve experimentally. We report the self-diffusion constant of magnesium in a magnesium oxide grain boundary as a function of temperature and pressure, obtained through molecular dynamics simulations.
 

Poster Session Runner-Ups:

Masood Delfarah, Graduate Student at The Ohio State University
"Recurrent Neural Networks for Cochannel Speech Separation in Reverberant Environments "
Speech separation is a fundamental problem in speech and signal processing. A particular challenge is monaural separation of cochannel speech, or a two-talker mixture, in a reverberant environment. In this paper, we study recurrent neural networks (RNNs) with long short-term memory (LSTM) in separating and enhancing speech signals in reverberant cochannel mixtures. Our investigation shows that RNNs are effective in separating reverberant speech signals. In addition, RNNs significantly outperform deep feedforward networks based on objective speech intelligibility and quality measures. We also find that the best performance is achieved when the ideal ratio mask (IRM) is used as the training target in comparison with alternative training targets. While trained using reverberant signals generated by simulated room impulse responses (RIRs), our model generalizes well to conditions where the signals are generated by recorded RIRs.

Rosario Distefano, Postdoctorate at The Ohio State University
"miREpiC: miRNA Editing Profiling in Cancer "
RNA editing is emerging as a new player in cancer biology (PubmedID:26439496;26440895).  Specifically, editing in microRNAs, established gene regulators, could aid in cancer prognosis and therapy (28411194). Expanding on our previous works (23044546;27298257), we leveraged on the NIH TCGA dataset (phs000178.v9.p8, comprising ~11K samples over 33 cancer types) and carried out a computational analysis to characterize and compare profiles of edited microRNAs, dysregulated across different cancer types. Our goal is to elucidate the potential of the RNA modification phenomena as potential marker for the classification of cancer subtypes as well as provide functional characterization of the shifted gene targetome.

SUG Press Release and Photos

Please see our press release and separate OSC Facebook album for additional photos.