• 

  Automating Wavefront Parallelization for Sparse Matrix Computations

  Venkat, Anand; Mohammadi, Mahdi Soltan; Park, Jongsoo; Rong, Hongbo; Barik, Rajkishore; Strout, Michelle Mills; Hall, Mary; Univ Arizona, Dept Comp Sci (IEEE, 2016)

  This paper presents a compiler and runtime framework for parallelizing sparse matrix computations that have loop-carried dependences. Our approach automatically generates a runtime inspector to collect data dependence information and achieves wavefront parallelization of the computation, where iterations within a wavefront execute in parallel, and synchronization is required across wavefronts. A key contribution of this paper involves dependence simplification, which reduces the time and space overhead of the inspector. This is implemented within a polyhedral compiler framework, extended for sparse matrix codes. Results demonstrate the feasibility of using automatically-generated inspectors and executors to optimize ILU factorization and symmetric Gauss-Seidel relaxations, which are part of the Preconditioned Conjugate Gradient (PCG) computation. Our implementation achieves a median speedup of 2.97x on 12 cores over the reference sequential PCG implementation, significantly outperforms PCG parallelized using Intel's Math Kernel Library (MKL), and is within 6% of the median performance of manually-parallelized PCG.
• 

  Galician coda restrictions and plural clusters

  Colina, S.; Simonet, M.; Univ Arizona, Dept Linguist (WALTER DE GRUYTER, 2014-11)

  The present study investigates the phonology and phonetics of Galician post-vocalic velar nasals. Galician has very strict coda restrictions – it does not allow for complex codas. One exception to this restriction is found in the plurals of words ending in a nasal consonant, which add /s/ to the “right” of a noun or adjective: man ‘hand’, mans ‘hands’; pan ‘bread’, pans ‘breads’. The present study puts forward a proposal, initially based on synchronic, formal phonological grounds, according to which post-vocalic, pre-/s/ nasals in plural forms are not nasal stops, but nasal glides. Their nature as nasal glides allows for their syllabification in the nucleus rather than in the coda, thus preserving (i.e., not violating) the restriction on complex codas. This proposal is then tested with a production experiment based on quantitative acoustic data. The acoustic study reveals indeed a difference in the degree of weakening of post-vocalic nasals, with pre-/s/ nasals in the plural forms showing a significantly higher degree of weakening than pre-/s/ nasals in the singular forms. The article concludes with an Optimality-Theoretic analysis of the phonological facts.
• 

  Epigenetic reactivation of LINE-1 retrotransposon disrupts NuRD corepressor functions and induces oncogenic transformation in human bronchial epithelial cells

  Bojang, Pasano, Jr.; Ramos, Kenneth S.; Univ Arizona, Coll Med, Div Pulm Allergy Crit Care & Sleep Med; Univ Arizona Hlth Sci, Ctr Appl Genet & Genom Med (WILEY, 2018-08)

  Long interspersed nuclear element-1 (LINE-1 or L1) reactivation is linked to poor prognosis in non-small-cell lung carcinoma (NSCLC), but the molecular bases of this response remain largely unknown. In this report, we show that challenge of human bronchial epithelial cells (HBECs) with the lung carcinogen, benzo(a)pyrene (BaP), shifted the L1 promoter from a heterochromatic to euchromatic state through disassembly of the nucleosomal and remodeling deacetylase (NuRD) complex. Carcinogen challenge was also associated with partial displacement of constituent proteins from the nuclear to the cytoplasmic compartment. Disruption of NuRD corepression by genetic ablation or carcinogen treatment correlated with accumulation of L1 mRNA and proteins. Mi2 bound directly to the L1 promoter to effect retroelement silencing, and this response required the DNA- and ATPase-binding domains of Mi2. Sustained expression of L1 in HBECs was tumorigenic in a human-SCID mouse xenograft model, giving rise to tumors that regressed over time. Together, these results show that functional modulation of the NuRD constituent proteins is a critical molecular event in the activation of L1 retrotransposon. L1 expression creates a microenvironment in HBECs that is conducive to neoplasia and malignant transformation.
• 

  A Multicellular Vascular Model of the Renal Myogenic Response

  Ciocanel, Maria-Veronica; Stepien, Tracy; Sgouralis, Ioannis; Layton, Anita; Univ Arizona, Dept Math (MDPI, 2018-07)

  The myogenic response is a key autoregulatory mechanism in the mammalian kidney. Triggered by blood pressure perturbations, it is well established that the myogenic response is initiated in the renal afferent arteriole and mediated by alterations in muscle tone and vascular diameter that counterbalance hemodynamic perturbations. The entire process involves several subcellular, cellular, and vascular mechanisms whose interactions remain poorly understood. Here, we model and investigate the myogenic response of a multicellular segment of an afferent arteriole. Extending existing work, we focus on providing an accurate-but still computationally tractable-representation of the coupling among the involved levels. For individual muscle cells, we include detailed Ca2+ signaling, transmembrane transport of ions, kinetics of myosin light chain phosphorylation, and contraction mechanics. Intercellular interactions are mediated by gap junctions between muscle or endothelial cells. Additional interactions are mediated by hemodynamics. Simulations of time-independent pressure changes reveal regular vasoresponses throughout the model segment and stabilization of a physiological range of blood pressures (80-180 mmHg) in agreement with other modeling and experimental studies that assess steady autoregulation. Simulations of time-dependent perturbations reveal irregular vasoresponses and complex dynamics that may contribute to the complexity of dynamic autoregulation observed in vivo. The ability of the developed model to represent the myogenic response in a multiscale and realistic fashion, under feasible computational load, suggests that it can be incorporated as a key component into larger models of integrated renal hemodynamic regulation.
• 

  Consistency of modularity clustering on random geometric graphs

  Davis, Erik; Sethuraman, Sunder; Univ Arizona, Dept Math (INST MATHEMATICAL STATISTICS, 2018-08)

  Given a graph, the popular "modularity" clustering method specifies a partition of the vertex set as the solution of a certain optimization problem. In this paper, we discuss scaling limits of this method with respect to random geometric graphs constructed from i.i.d. points X-n = {X-1, X-2,..., X-n}, distributed according to a probability measure nu supported on a bounded domain D subset of R-d. Among other results, we show, via a Gamma convergence framework, a geometric form of consistency: When the number of clusters, or partitioning sets of X-n is a priori bounded above, the discrete optimal modularity clusterings converge in a specific sense to a continuum partition of the underlying domain D, characterized as the solution to a "soap bubble" or "Kelvin"-type shape optimization problem.

View more