• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.1-17
• /
• 2017

The deep learning framework is software designed to help develop deep learning models. Some of its important functions include "automatic differentiation" and "utilization of GPU". The list of popular deep learning framework includes Caffe (BVLC) and Theano (University of Montreal). And recently, Microsoft's deep learning framework, Microsoft Cognitive Toolkit, was released as open-source license, following Google's Tensorflow a year earlier. The early deep learning frameworks have been developed mainly for research at universities. Beginning with the inception of Tensorflow, however, it seems that companies such as Microsoft and Facebook have started to join the competition of framework development. Given the trend, Google and other companies are expected to continue investing in the deep learning framework to bring forward the initiative in the artificial intelligence business. From this point of view, we think it is a good time to compare some of deep learning frameworks. So we compare three deep learning frameworks which can be used as a Python library. Those are Google's Tensorflow, Microsoft's CNTK, and Theano which is sort of a predecessor of the preceding two. The most common and important function of deep learning frameworks is the ability to perform automatic differentiation. Basically all the mathematical expressions of deep learning models can be represented as computational graphs, which consist of nodes and edges. Partial derivatives on each edge of a computational graph can then be obtained. With the partial derivatives, we can let software compute differentiation of any node with respect to any variable by utilizing chain rule of Calculus. First of all, the convenience of coding is in the order of CNTK, Tensorflow, and Theano. The criterion is simply based on the lengths of the codes and the learning curve and the ease of coding are not the main concern. According to the criteria, Theano was the most difficult to implement with, and CNTK and Tensorflow were somewhat easier. With Tensorflow, we need to define weight variables and biases explicitly. The reason that CNTK and Tensorflow are easier to implement with is that those frameworks provide us with more abstraction than Theano. We, however, need to mention that low-level coding is not always bad. It gives us flexibility of coding. With the low-level coding such as in Theano, we can implement and test any new deep learning models or any new search methods that we can think of. The assessment of the execution speed of each framework is that there is not meaningful difference. According to the experiment, execution speeds of Theano and Tensorflow are very similar, although the experiment was limited to a CNN model. In the case of CNTK, the experimental environment was not maintained as the same. The code written in CNTK has to be run in PC environment without GPU where codes execute as much as 50 times slower than with GPU. But we concluded that the difference of execution speed was within the range of variation caused by the different hardware setup. In this study, we compared three types of deep learning framework: Theano, Tensorflow, and CNTK. According to Wikipedia, there are 12 available deep learning frameworks. And 15 different attributes differentiate each framework. Some of the important attributes would include interface language (Python, C ++, Java, etc.) and the availability of libraries on various deep learning models such as CNN, RNN, DBN, and etc. And if a user implements a large scale deep learning model, it will also be important to support multiple GPU or multiple servers. Also, if you are learning the deep learning model, it would also be important if there are enough examples and references.

• Radiation Oncology Journal
• /
• v.19 no.2
• /
• pp.181-189
• /
• 2001

Purpose : Changes in the balance between MMP and TIMP can have a profound effect on the composition in the extracellular matrix (ECM) and affect various cellular functions including adhesion, migration, differentiation of cells, and fibrosis and invasion and metastasis of cancer cells. Radiation therapy is a popular treatment modality for benign and malignant tumor, but the study for radiation effect on MMP and TIMP is scarce. In the current study, we have examined the expression of TIMP in fibrosis-prone (C57BL/6) mice after radiation. Methods and Materials : Adult female mice of $10\~12$ weeks were used. The whole body were irradiated using a Varian CL-4/100 with 2 and 10 Gy. Immunohistochemical staining was peformed according to Avidin Biotin complex method and evaluated by observing high power field. For TIMP-1, TIMP-2 antibodies, reactivity was assessed in the parenchymal cell and in the stromal cell. The scale of staining was assessed by combining the quantitative and qualiative intensity of staining. Results : TIMP-1 immunoreactivity did not change in lung. But, in liver, TIMP-1 immunoreactivity was localized in cytoplasm of hepatocyte and Kupffer cell. in kidney, TIMP-1 immunoreactivity was localized in cytoplasm of some tubular cell. Temporal variations were not seen. Dose-response relationship was not seen except kidney. TIMP-2 immunoreactivity in lung was a score (++) at 0 Gy and elevated to a score (+++) at 2 Gy. TIMP-2 immunoreactivity was a score (++) in liver at 0 Gy. TIMP-2 immunoreactivity was localized in cytoplasm of hepatocyte and Kupffer cell as same as patterns of TIMP-1 immunoreactivity. The TIMP-2 immunoreactivity in liver was elevated to (+++) at 2 Gy. Immunoreactivity to TIMP-2 in kidney was a score (+++) at 0 Gy and was not changed at 10 Gy. The score of TIMP-2 immunoreactivity was reduced to (++) at 2 Gy. TIMP-2 immunoreactivity was confined to tubules in kidney. Temporal variation of TIMP-2 immunoreactivity was irregular. Dose-response relationship of TIMP-2 immunoreactivity was not seen. Conclusions : Differences between intensity of expression of TIMP-1 and TIMP-2 in each organ was present. Expression of TIMP was localized to specific cell in each organ. Irradiation increased TIMP-1 immunoreactivity in the liver and the kidney. Irradiation increased TIMP-2 immunoreactivity in the lung. But, in the liver and the kidney, TIMP-2 expression to radiation was irregular. Temporal variation of TIMP-2 immunoreactivity was irregular. Dose-response relationship of TIHP-2 immunoreactivity was not seen. In the future, we expect that the study of immunohistochemical staining of longer period of postirradiation and quantitative analysis using western blotting and northern blotting could define the role of TIMP in the radiation induced tissue fibrosis.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.30 no.3
• /
• pp.271-276
• /
• 1985

Effects of media and its components on callus induction and plant regeneration were studied to increase the cultural efficiency in rice anther culture. The N$_{6}$ basic medium gave better results in callus induction than those of MS or Miller. The medium used for callus induction affects the plant regeneration. The frequency of plant regeneration from callus grown on Miller basic medium was lower than those of N$_{6}$ or MS. Most of calli derived from anthers, above 90%, were induced from 20 days to 40 days after anther inoculation. The cultural efficiency of modified N$_{6}$ basic medium which was composed of 31.5mM KNO$_3$ and 1.75mM(NH$_4$)$_2$SO$_4$ as nitrogen sources was higher than those of N$_{6}$ basic medium. Combination of NAA and Kinetin showed better results than that of 2, 4-D only in cultural efficiency. Effect of DL-alanine on callus induction in Indica variety, IR40, showed better response in the anthers pretreated for 6 days at 12$^{\circ}C$.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.14 no.2
• /
• pp.962-969
• /
• 2013

This study used Computational Fluid Dynamics(CFD) to assess the properties of the air current inflow and the flow velocity distribution in the fume hood. In order to verify the effect of improvement, it was also predicted the characteristics of the flow pattern in case the hood face is structurally improved. The assessment of the capture velocity with the existing fume hood confirmed maximum 23 to 30% difference as compared to the root mean square (RMS). And the hood face showed great difference in flow velocity, with the flow velocity in the upper part is 58 to 68% faster than that in the lower part of the hood. So, as a result of the improvement of the hood designed to maintain a steady exhaust at the hood face (that is, installing a baffle on the inner wall and designing the slot type face), a difference of maximum 7% as compared to RMS appeared while maximum 12% differentiation in flow velocity through sections was predicted, showing mitigation of much of the difference in control velocity as compared to the previous structure.

• Molecules and Cells
• /
• v.19 no.2
• /
• pp.185-190
• /
• 2005

The chicken oviduct is a dynamic organ that produces secretory proteins such as ovalbumin and its cells undergo cell proliferation and differentiation. There has been no study of the cellular mechanism involved in cell death in the chicken oviduct. Therefore, this study has focused on the study of apoptosis in primary oviduct cells. Because ceramide is known to activate apoptosis in tumor cells and is produced in the oviduct, we used an exogenous ceramide analog to induce cell death. The viability of ceramide-treated chicken oviduct cells decreased in a dose-dependent manner and apoptotic cells were detected by staining with annexin V. The expression of apoptosis-related genes was assessed by RT-PCR and bcl-2 mRNA was found to decrease after exposure to ceramide while Bcl-x mRNA increased 12 h post-treatment. In addition, caspase-3 was expressed strongly in the early stages of apoptosis, while caspase-1 and -9 transcripts increased at later times. We conclude that ceramide induces apoptosis in oviduct-derived primary cells via a caspase- and bcl-2-dependent pathway.

• The Journal of the Korean dental association
• /
• v.46 no.9
• /
• pp.563-573
• /
• 2008

Purpose : The purpose of this study is to investigate the effects of Simvastain, which is HMG-CoA reductase inhibitor, on proliferation and differentiation of osteoblast. Materials & Methods : Twenty-four cell culture plates containing essential medium were seeded with UMR-106 cell lines, at density of 5 x $10^4$ cells per plate. Each plates were incubated with 5% $CO^2$incubator $37^{\circ}C$. Starting from 2 days after incubation, cell culture medias were replaced with Osteogenesis induction media every 2 days, for 12 days. In some plates, 0.01, 0.1, 1, 10, $100\muM$ of Simvastatin were added with Osteogenesis induction media, and classified as "test group". Those not added with Simvastatin were classified as "control group". Results : 1. When Alrizarin Red staining was observed with naked eye, control group showed normal deep red color, but test group show rapid decrease of red color as Simvastatin concentration increased more than $0.1\muM$. 2, When observed with microscope, compared to control group, amount of osteo matrix stained with Alrizarin Red decreased rapidly in Simvastatin concentration more than $0.1\muM$. 3. In optical density analysis, regarding control group as a basis, mineral deposition decreased rapidly when Simvastatin concentration increased more than $0.1\muM$. 4. In flow cytometry analysis, survival rate of UMR-106 cell showed no changes in both control group and test group. Conclusion : From the above results, we were able to identify that Simvastatin inhibited osteogenesis without effecting survival or cell number of osteoblasts.

• Reproductive and Developmental Biology
• /
• v.29 no.2
• /
• pp.69-73
• /
• 2005

The hematopoietic growth factor erythropoietin (EPO) is required for the maintenance, proliferation, and differentiation of the stem cells that produce erythrocytes. To analyse the biological activity of the recombinant human EPO (rec-hEPO), we have cloned the EPO cDNA and genomic DNA and produced rec-hEPO in the CHO cell lines. The growth and differentiation of EPO-dependent human leukemic cell line (F36E) were used to measure cytokine dependency and in vitro bioactivity of rec-hEPO. MIT assay values were increased by survival of F36E cells at 24h or 72h. The hematocrit and RBC values were increased by subcutaneous injection of 20 IU (in mice) and 100IU(in rats) rec-hEPO. Hematocrit values remarkably increased at $13.2\%$ (in mice) and $12.2\%$ (in rats). The pharmacokinetic behavior with injection of 6 IU of rec-hEPO remained detectable after 24 h in all mice tested. The highest peat appeared at 2h after injection. The long half-life of rec-hEPO is likely to confer clinical advantages by allowing less frequent dosing in patients treated for anemia. These data demonstratethat ree-hEPO produced in this study has a potent activity in vivo and in vitro. The results also suggest that biological activity of ree-hEPO could be remarkably enhanced by genetic engineering that affects the potential activity, including mutants with added oligosaccharide chain and designed to produce EPO-EPO fusion protein.

• Clinical and Experimental Reproductive Medicine
• /
• v.37 no.3
• /
• pp.207-216
• /
• 2010

Objective: This study was performed to clarify the role of HomeoboxA (HOXA) and its related signaling molecules in the decidualization of primary cultured endometrial cells. Methods: Human endometrial tissues were obtained by curettage of hysterectomy specimens from patients with conditions other than endometrial diseases. Tissues were minced and digested with Trypsin-EDTA for 20 min, $37^{\circ}C$. Cells were cultured with DMEM/F12 medium in $37^{\circ}C$, 5% $CO_2$ incubator for 24 hrs. Cells were treated with HOXA10 siRNA and added transforming growth factor (TGF)-${\beta}1$ (10 ng/mL) for 48 hrs to induces decidualization in vitro. Reverse transcription polymerase chain reaction analysis was accomplished to observe the expression of HOXA10, prolactin, cyclooxygenase (COX)-2, peroxisome proliferatoractivated receptor (PPAR)-$\gamma$, and wingless-type MMTV integration site family (Wnt). Results: HOXA10 expression was increased (1.8 fold vs. non-treated control) in TGF-${\beta}1$ treated cells. Decidualization marker, prolactin, was significantly increased in TGF-${\beta}1$ treated cells compared with HOXA10 siRNA treated cells. Endometrial cell differentiation marker, COX-2 was down-regulated by HOXA10 siRNA even if cells were treated with TGF-${\beta}1$. Wnt4 was down-regulated by treated with HOXA10 siRNA, this expression patters was not changed by TGF-${\beta}1$. Expression of PPAR-$\gamma$ was down regulated by TGF-${\beta}1$ in regardless of HOXA10 siRNA treatment. Conclusion: TGF-${\beta}1$ which is induced by progesterone in endometrial epithelial cells may induces stromal cell decidualization via HOXA10 and Wnt signaling cascade.

• International Journal of Oral Biology
• /
• v.34 no.1
• /
• pp.21-28
• /
• 2009

Mutations in DLX3 are associated with both autosomal dominant hypoplastic hypomaturation amelogenesis imperfecta (ADHHAI) and tricho-dento-osseous (TDO) syndrome. ADHHAI is caused by a c.561_562delCT (2bp-del DLX3) mutation whereas TDO syndrome is associated with a c.571_574delGGGG (4bp-del DLX3) mutation. However, although the causal relationships between DLX3 and an enamel phenotype have been established, the pathophysiological role of DLX3 mutations in enamel development has not yet been clarified. In our current study, we prepared expression vectors for wild type and deletion mutant DLX3 products (4bp-del DLX3, 2bp-del DLX3) and examined the effects of their overexpression on the expression of the enamel matrix proteins and proteases. Wild type DLX3 enhanced the expression of matrix metalloprotease 20 (MMP20) mRNA and protein in murine ameloblast-like cells. However, neither a 4bp-del nor 2bp-del DLX3 increased MMP20 expression. Wild type DLX3, but not the above DLX3 mutants, also increased the activity of reporters containing 1.5 kb or 0.5 kb of the MMP20 promoter. An examination of protein stability showed that the half-life of wild type DLX3 protein was less than 12 h whilst that of both deletion mutants was longer than 24 h. Endogenous Dlx3 was also found to be continuously expressed during ameloblast differentiation. Since inactivating mutations in the gene encoding MMP20 are associated with amelogenesis imperfecta, the inability of 4bp-del or 2bp-del DLX3 to induce MMP20 expression suggests a possible involvement of such mutations in the enamel phenotype associated with TDO syndrome or ADHHAI.

• Tuberculosis and Respiratory Diseases
• /
• v.63 no.4
• /
• pp.353-361
• /
• 2007

Background: Malignancies are a common and important causes of exudative pleural effusions. Several tumor markers have been studied because the pleural fluid cytology and pleural biopsy specimens do not provide a diagnosis in a high percentage of malignant effusions. In an attempt to overcome this limitation, procalcitonin and C-reactive protein (CRP) in pleural effusions and serum, which are known to be inflammation markers, were measured to determine if they can differentiate an exudate from trasndate as well as the diverse causes of exudative pleural effusion. Methods: 178 consecutive patients with pleural effusion (malignant 57, tuberculous 51, parapneumonic 31, empyema 5, miscellaneous benign 7, transudative 27)were studied prospectively. The standard parameters of pleural effusion and measured serum and pleural procalcitonin were examined using in immunoluminometric assay. The level of CRP in serum and pleural fluid was determined by turbidimetric immunoassay. Results: The pleural procalcitonin levels in the exudate were significantly higher than those in the transudate, $0.81{\pm}3.09ng/mL$ and $0.12{\pm}0.12ng/mL$, respectively (p=0.007). The pleural CRP levels were significantly higher in the exudate than the transudate, $2.83{\pm}3.31mg/dL$ and $0.74{\pm}0.67mg/dL$, respectively (p<0.001). The pleural procalcitonin levels in the benign effusion were significantly higher than those in the malignant effusion, $1.15{\pm}3.82ng/mL$ and $0.25{\pm}0.92ng/mL$, respectively (p=0.032). The pleural CRP levels were significantly higher in the benign effusion than in the malignant effusion, $3.68{\pm}3.78mg/dL$ and $1.42{\pm}1.54mg/dL$, respectively (p<0.001). The pleural procalcitonin levels in the non-tuberculous effusion were significantly higher than those in the tuberculous effusion, $1.16{\pm}3.75ng/mL$ and $0.13{\pm}0.37ng/mL$, respectively (p=0.008). Conclusion: Measuring the level of procalcitonin and CRP in the pleural fluid is helpful for differentiating between transudates and exudates. In addition, it is useful for differentiating between benign and malignant pleural effusions.