247 research outputs found
ShmCaffe: A Distributed Deep Learning Platform with Shared Memory Buffer for HPC Architecture
No full textOne of the reasons behind the tremendous success of deep learning theory and applications in the recent days is advances in distributed and parallel high performance computing (HPC). This paper proposes a new distributed deep learning platform, named ShmCaffe, which utilizes remote shared memory for communication overhead reduction in massive deep neural network training parameter sharing. ShmCaffe is designed based on Soft Memory Box (SMB), a virtual shared memory framework. In the SMB framework, the remote shared memory is used as a shared buffer for asynchronous massive parameter sharing among many distributed deep learning processes. Moreover, a hybrid method that combines asynchronous and synchronous parameter sharing methods is also discussed in this paper for improving scalability. As a result, ShmCaffe is 10.1 times faster than Caffe and 2.8 times faster than Caffe-MPI for deep neural network training when Inception\-v1 is trained with 16 GPUs. We verify the convergence of the Inception\-v1 model training using ShmCaffe-A and ShmCaffe-H by varying the number of workers. Furthermore, we evaluate scalability of ShmCaffe by analyzing the computation and communication times per one iteration of deep learning training in four convolutional neural network (CNN) models
Replication Data for: Can Foreign Aid Improve the Donor Country’s Image among a Third-party Country’s Public? The Case of a World Heritage Site Restoration Project
No full textThis study explores whether and under what conditions foreign aid can help improve the donor country’s image in countries that did not receive aid. We identified a world heritage site restoration project that is visible and localized, has no political attachments, and deals with the global public good as the type of foreign aid most likely to generate this positive effect. In light of the literature suggesting that tensions with the target country undermine the effectiveness of public diplomacy, we expect the positive effect will be more pronounced in non-recipient countries with which the donor country has a more amicable relationship. To empirically investigate our argument, we conduct a survey experiment in Australia, which is a developed non-aid-receiving country. We provide information to the Australian public about an aid project to restore the Angkor Monument in Cambodia conducted either by China or South Korea. We find that information on Korea’s aid to Cambodia improves the image of Korea and the willingness to cooperate with the Korean government among Australians. No such effect, however, is observed for similar aid provided by China, whose relations with Australia have been strained in multiple domains. Our findings have policy implications for donor countries seeking to utilize the soft power element of foreign aid as a public diplomacy tool
Modulation of saliva pattern and accurate detection of ovulation using an electrolyte pre-deposition-based method: a pilot study
No full textWe developed an electrolyte pre-deposition-based saliva pattern modulation method to detect ovulation with high accuracy and reliability. Ovulation tests using human saliva have advantages in terms of the earlier ovulation detection and more convenient sample collection procedure; however, accuracy is low, which is a critical limitation given that the concentrations of salivary constituents can vary depending on the health status of the tested individual and subjective user judgement of the test result. In this study, we quantitatively analyzed saliva patterns according to the concentrations of electrolytes and proteins in the ovulation test and found that changes in the saliva pattern during the ovulatory period can be controlled by sodium chloride (NaCl) pre-deposition, which directly affects the accuracy of ovulation detection. The 100 nmol NaCl pre-deposition condition proved optimal, being two-fold more sensitive to changes in saliva pattern versus the non-pre-deposition condition (accuracy of ovulation detection = 66.6% and 33.3%, respectively). Although accuracy remained insufficient for actual applications compared to the urine-based ovulation detection method, we expect that the electrolyte pre-deposition method will greatly contribute to enhancing the performance of saliva-based ovulation detection tests, toward a commercially satisfactory level of accuracy.11Nsciescopu
Designing for separation anxiety: Engaging connection and building trust
No full textDSpace SAF Submission Ingestion Package generated from Vireo submission #17132 on 2022-04-29 at 16:08:58Made available in DSpace on 2022-04-29T21:58:14Z (GMT). No. of bitstreams: 2
KIM-THESIS-2021.pdf: 4578155 bytes, checksum: f7b14847faa55a6cadffdf00855e67f5 (MD5)
LICENSE.txt: 4206 bytes, checksum: 5be11982733f84f2e4b673a8724a3035 (MD5)
Previous issue date: 2021-12-10Embargo set by: Seth Robbins for item 123408
Lift date: 2024-04-29T21:58:46Z
Reason: Author requested closed access (OA after 2yrs) in Vireo ETD systemAuthor requested closed access (OA after 2yrs) in Vireo ETD systemLimitedSubmission published under a 24 month embargo labeled 'Closed Access', the embargo will last until 2023-12-01The student, Eunji Kim, accepted the attached license on 2021-12-08 at 09:46.The student, Eunji Kim, submitted this Thesis for approval on 2021-12-08 at 09:49.This Thesis was approved for publication on 2021-12-10 at 10:02."""The bird fights its way out of the egg. The egg is the world, and whoever will be born must destroy the world."" - Demian, Hermann Hesse.
As they grow old, babies develop attachment and bonding to their guardians, particularly their mothers. It is a natural human tendency to be excessively reliant on one's parents because they are incapable of surviving without their assistance. In being helped, and their instinctive needs are satisfied, the baby develops attachment and bonding to the guardian who helps them. While an infant develops a connection to a guardian, they may show anxiety and be extra clingy when separation occurs. Many parents and children suffer from these problems, as the parents cannot always be alongside the children. Separation Anxiety Disorder (SAD) could lead to perpetual psychological issues throughout the children's adulthood. Ultimately the goal of a child is to be self-sufficient and independent from their guardian. Along the way, the guardian must be aware of the children's needs and help them cope with the problem.
Within this study, separation anxiety is covered following the infant's cognitive development and how it can be possibly treated. Also, while referencing the traditionally applied or accepted treatment methods, I explored the tangible and intangible design options that could connect children to their parents to soothe their anxiety and help them form a secure attachment to their guardians, and eventually help them be independent.
Two essays in finance
No full textThe student, Eunji Oh, accepted the attached license on 2019-07-03 at 09:31.The student, Eunji Oh, submitted this Dissertation for approval on 2019-07-03 at 09:40.This Dissertation was approved for publication on 2019-07-03 at 13:11.DSpace SAF Submission Ingestion Package generated from Vireo submission #14146 on 2019-11-26 at 13:04:22Made available in DSpace on 2019-11-26T20:49:20Z (GMT). No. of bitstreams: 2
OH-DISSERTATION-2019.pdf: 1199904 bytes, checksum: 64b5174294342284960453cff82dae53 (MD5)
LICENSE.txt: 4205 bytes, checksum: cd41ec9f20030d48d5a2be940081acfb (MD5)
Previous issue date: 2019-07-03The first essay examines whether systematic equity risk of firms reflects the risk of their R&D strategies at various angles. More novel R&D strategy is risky because it can be related to more extreme outcome. This risk could indirectly affect the firm’s systematic risk. In the case of success of the strategy, the productivity of technologies developed by novel R&D strategy could be procyclical; thus public firms with more novel technologies could be more subject to the aggregate risk. To investigate this problem, I devise an ex-ante measure for the novelty of innovation, Tech Synthesis Level (below TSL), which quantifies the degree that new technology is drawn from prior technologies in the far different technological fields, using patent citations. I find that patents with high TSL are associated with more extreme technological outcomes both at patent-level and startup-level. At public firm analyses, I find that high TSL is associated with high abnormal returns by 2.532 percent (annualized) and high systematic volatility. These findings support the hypothesis that a failure probability of the R&D project increases systematic risk. I also find evidence that high TSL patents are technologically more productive when aggregate innovation is very active, so firms with high TSL patents are subject to high systematic equity risk.
The second essay studies the effect of intangible collateral, which has gradually increased since the ’90s, by testing hypotheses inspired by [Ai et al., 2018]’s collateralizability premium. Firms with more collateralizable capital have lower stock returns due to the insurance effect of the capital during economic recessions when financial constraints get tighter. If intangible collateral also can relax financial constraint, firms with intangible collateral are expected to have lower stock returns than the other similar firms without collateralizable intangible capital. I add empirical evidence by using Dealscan data of US-originated secured long-term loans. I find that firms using intangibles as collateral in addition to traditional collateralizable assets have higher stock returns than the other firms pledging only tangible assets to secure corporate loans. Also, they could achieve the similar or even slightly higher level of leverage, implying intangible collateral also can relax financial constraint. This is not assumed possible in many theoretical and empirical studies. Even with matching analysis I find that firms pledging intangible capital as collateral still have higher stock returns than the other similar firms without intangible collateral. The empirical evidence I find does not fully support the collateralizability premium hypothesis.Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2021-08-01Embargo set by: Seth Robbins for item 112924
Lift date: 2021-11-26T20:49:41Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemU of I Only Restriction Lifted for Item 112924 on 2021-11-27T10:15:30Z
On convergence rates of convex regression in multiple dimensions
No full textWe consider a least squares estimator for estimating a convex function f*: [0,1]d → ℝ with bounded sub-gradients. A rate at which the sum of squared differences between the estimator and the true function f* converges to zero is computed. This work sheds light on computing the convergence rate of the multidimensional convex regression estimator. © 2014 INFORMS
On the Convergence Rate for Stochastic Approximation in the Nonsmooth Setting
No full text We consider a stochastic approximation (SA) method for finding the minimizer of a function f, which is convex but nondifferentiable at the minimizer. Due to the nondifferentiability at the minimizer, f is allowed to increase at a positive rate in a neighborhood of the minimizer. From this property, we show that the nth estimate for the minimizer generated by the SA procedure converges at a rate of 1/n in the mean, which is significantly faster than the classical convergence rates for differentiable functions f. We also discuss an example from an inventory control system that exhibits a convex but nondifferentiable cost function. </jats:p
A resilient MXene/CNT nano-accordion framework for anode-free sodium-metal batteries with exceptional cyclic stability
No full textAnode-free sodium-metal batteries are considered one of the most promising alternatives for developing high-end batteries due to their high theoretical capacity, low cost, and high natural abundance. However, they have severe drawbacks in the form of inferior long-term cyclic stability. We engineered mechanically resilient MXene/CNT nano-accordion frameworks (NAFs) to host significant Na without dendrite development, even at high currents. The microcellular structures of MXene/CNT-NAFs possess numerous micro-sized pores and sodium nucleation sites. The synergetic effects of strong adhesion and charge transfer between MXene and CNT reduce overpotential during plating/stripping and facilitate uniform Na deposition. Resilient nano-accordion structures are compressed by capillary of Na nucleation during plating and expand during stripping, allowing long-term plating/ stripping with little volume change. These benefits allow the MXene/CNT-NAFs/Na asymmetric cell to maintain its average CE at 99.7 % with capacities of 1.0 mAh & sdot;cm- 2 at 1.0 mA & sdot;cm- 2 for 900 h. Furthermore, MXene/CNTNAFs symmetric cell exhibits a very low overpotential of 12.0 mV after 1,500 h with a capacity of 3.0 mAh & sdot;cm- 2 at 3.0 mA & sdot;cm- 2 and stores high capacities of 20.0 mAh & sdot;cm- 2 at 5.0 mA & sdot;cm-2 for 1,200 h. The anode-free MXene/ CNT-NAFs//Na3V2(PO4)3@C full-cell demonstrates exceptional long-term cyclic stability over 5,000 cycles at 5.0 C and 10.0 C without cell failure.
- …
