Deep learning (DL) is one of the most prominent branches of machine learning. Due to the immense computational cost of DL workloads, industry and academia have developed DL libraries with ...highly-specialized kernels for each workload/architecture, leading to numerous, complex code-bases that strive for performance, yet they are hard to maintain and do not generalize. In this work, we introduce the batch-reduce GEMM kernel and show how the most popular DL algorithms can be formulated with this kernel as the basic building-block. Consequently, the DL library-development degenerates to mere (potentially automatic) tuning of loops around this sole optimized kernel. By exploiting our new kernel we implement Recurrent Neural Networks, Convolution Neural Networks and Multilayer Perceptron training and inference primitives in just 3K lines of high-level code. Our primitives outperform vendor-optimized libraries on multi-node CPU clusters, and we also provide proof-of-concept CNN kernels targeting GPUs. Finally, we demonstrate that the batch-reduce GEMM kernel within a tensor compiler yields high-performance CNN primitives, further amplifying the viability of our approach.
Objective: CX
3CR1 is a novel chemokine receptor located on monocytes. Recently, two polymorphisms were linked to coronary artery disease (CAD), V249I and T280M. Carriers of at least one I-allele or ...one M-allele were found less frequently among patients with CAD compared to controls. The aim of the present study was to investigate the influence of these polymorphisms on the development of peripheral arterial disease (PAD).
Methods: 522 human subjects with documented PAD and 522 age and sex matched controls were genotyped by polymerase chain reaction followed by restriction digestion.
Results: Adjusted odds ratio (OR) of carriers of the I-allele for PAD was 1.34 (95% confidential interval (CI) from 0.86 to 2.09;
P=0.19). The OR associated with the M-allele for PAD was 0.65 (95% CI from 0.41 to 1.04;
P=0.07), when tested in the same regression analysis with the V249I genotypes. The genotypes were not linked to age at onset or severity of the disease. A subgroup of 137 CAD patients of whom 131 could be genotyped and who did not differ in baseline parameters from the remaining PAD patients, showed VV-genotype in 52.0%, VI in 42.7% and II in 5.3% CAD (OR associated with the I-allele for CAD: 1.29; 95% CI: 0.66–2.51;
P=0.46). The distribution of the T280M genotypes was 67.1, 29.8, 3.1% (TT, TM, MM) also showing no association with CAD (OR=0.77; 95% CI 0.36–1.46;
P=0.37).
Conclusion: In this study we could not detect a difference in genotype frequencies of the V249I and T280M polymorphisms in CX
3CR1 between PAD patients and controls. CAD concomitant with PAD was also not affected by the I- or the M-allele.
Skull Rommel, Sentiel A.; Pabst, D. Ann; McLellan, William A. ...
Encyclopedia of Marine Mammals,
2018, 20180000
Book Chapter, Reference
This chapter compares the skull anatomy of a diverse group of seven marine mammals to the basic anatomy of terrestrial mammals. It pays particular attention to the relation between feeding and diet ...and skull morphology, and discusses telescoping in cetacean skulls in detail. It also discusses the cranial nerves and their link to the foramina in the skull they travel through.
The transcription factor hypoxia-inducible factor 1 (HIF1) is the crucial regulator of
genes that are involved in metabolism under hypoxic conditions, but information regarding the
transcriptional ...activity of HIF1 in normoxic metabolism is limited. Different tumor cells were treated
under normoxic and hypoxic conditions with various drugs that affect cellular metabolism. HIF1ff
was silenced by siRNA in normoxic/hypoxic tumor cells, before RNA sequencing and bioinformatics
analyses were performed while using the breast cancer cell line MDA-MB-231 as a model. Differentially
expressed genes were further analyzed and validated by qPCR, while the activity of the metabolites
was determined by enzyme assays. Under normoxic conditions, HIF1 activity was significantly
increased by (i) glutamine metabolism, which was associated with the release of ammonium, and
it was decreased by (ii) acetylation via acetyl CoA synthetase (ACSS2) or ATP citrate lyase (ACLY), respectively, and (iii) the presence of L-ascorbic acid, citrate, or acetyl-CoA. Interestingly, acetylsalicylic
acid, ibuprofen, L-ascorbic acid, and citrate each significantly destabilized HIF1ff only under normoxia.
The results from the deep sequence analyses indicated that, in HIF1-siRNA silenced MDA-MB-231
cells, 231 genes under normoxia and 1384 genes under hypoxia were transcriptionally significant
deregulated in a HIF1-dependent manner. Focusing on glycolysis genes, it was confirmed that HIF1
significantly regulated six normoxic and 16 hypoxic glycolysis-associated gene transcripts. However,
the results from the targeted metabolome analyses revealed that HIF1 activity affected neither the
consumption of glucose nor the release of ammonium or lactate; however, it significantly inhibited
the release of the amino acid alanine. This study comprehensively investigated, for the first time,
how normoxic HIF1 is stabilized, and it analyzed the possible function of normoxic HIF1 in the
transcriptome and metabolic processes of tumor cells in a breast cancer cell model. Furthermore, these
data imply that HIF1 compensates for the metabolic outcomes of glutaminolysis and, subsequently,
theWarburg effect might be a direct consequence of the altered amino acid metabolism in tumor cells.