Search

Hardware and software technologies are being developed to efficiently run various deep learning models from embedded systems to servers. Moreover, multiple deep learning models compete for system resources such as CPU and memory to meet the diverse needs of users and improve recognition accuracy. MLPerf was developed to objectively evaluate these deep learning model execution devices. However, the current MLPerf executes DNN models in a serialized manner, which does not reflect the real-world environment where multiple DNN models are running. In this paper, we improve MLPerf to enable the simultaneous execution needs of multiple DNNs through multi-threading-based query parallelization, and to enable multiple DNN models to be executed on the target system similar to the real environment. The experimental results show that the query processing performance is about two times higher than the existing one, which is similar to the actual execution environment. [ABSTRACT FROM AUTHOR]

This study was performed to develop a Kdiet-index based on the definitions and characteristics of traditional Korean food. The Korean Food Pattern Index (Kdiet-index) for diseases was applied using dietary data from the 2017 National Health and Nutrition Examination Survey (KNHANES) and the association between Kdiet-index and metabolic disease was confirmed. The study subjects were 1,971 Korean adults aged >40, and the components were classified according to 14 criteria used to establish the Kdiet-index. Main ingredients up to the top 3 were designated using dish names, ingredients, and contents using KNHANES dietary data. Kdiet-index was classified using scores of ≤3 points, 4-7 points, and ≥8 points and total Kdiet-index were calculated by summing dietary scores. Correlations between Kdiet-index and obesity, hypertension, hyperlipidemia, and diabetes were analyzed. The odds ratio (OR) and 95% confidence interval (CI) for obesity were 0.531, 0.385~0.732 and for elevated cholesterol (≥240 mg/dL) indices were 0.471, 0.282~0.788, respectively, which showed a significant decreased in the risk for each disease for Kdiet-index of ≥8 points. This study confirms that metabolically related clinical results improved significantly as Kdiet-index increased and that higher Korean food pattern indices are associated with lower risks of metabolic disease [ABSTRACT FROM AUTHOR]

With the development of deep learning technology, there are many cases of using DNNs in embedded systems such as unmanned vehicles, drones, and robotics. Typically, in the case of an autonomous driving system, it is crucial to run several DNNs which have high accuracy results and large computation amount at the same time. However, running multiple DNNs simultaneously in an embedded system with relatively low performance increases the time required for the inference. This phenomenon may cause a problem of performing an abnormal function because the operation according to the inference result is not performed in time. To solve this problem, the solution proposed in this paper first reduces the computation by applying the Tucker decomposition to DNN models with big computation amount, and then, make DNN models run in parallel as much as possible in the unit of hidden layer inside the GPU. The experimental result shows that the DNN inference time decreases by up to 75.6% compared to the case before applying the proposed technique. [ABSTRACT FROM AUTHOR]

With the recent development of deep learning technology, autonomous things technology is attracting attention, and DNNs are widely used in embedded systems such as drones and autonomous vehicles. Embedded systems that can perform large-scale operations and process multiple DNNs for high recognition accuracy without relying on the cloud are being released. DNNs with various levels of priority exist within these systems. DNNs related to the safety-critical applications of autonomous vehicles have the highest priority, and they must be handled first. In this paper, we propose a priority-based scheduling framework for DNNs when multiple DNNs are executed simultaneously. Even if a low-priority DNN is being executed first, a high-priority DNN can preempt it, guaranteeing the fast response characteristics of safety-critical applications of autonomous vehicles. As a result of checking through extensive experiments, the performance improved by up to 76.6% in the actual commercial board. [ABSTRACT FROM AUTHOR]

Purpose: We evaluated results of arthroscopically assisted volar and dorsal scapholunate (SL) ligament reconstruction using flexor carpi radialis (FCR). Methods: We reviewed 31 cases who had been operated from August 2015 to June 2018. Two cases were excluded and 4 cases lost contact. Twenty-five wrists were included. The average follow-up duration was 25.3 months. Diagnostic arthroscopy was performed to evaluate European Wrist Arthroscopy Society (EWAS) stage of SL instability and the reparability of SL ligament. The SL reconstruction was performed for the cases which showed EWAS stage IIIC, IV having gap over 3 mm and irreparable SL ligament. The interval between the diagnosis and operation was 5.9 days in average. We modified the arthroscopic Corella technique by repairing the graft tendon to FCR. Results: All SL joints were stabilized to EWAS stage I arthroscopically after reconstruction. Disabilities of Arm, Shoulder and Hand (DASH) score changed from 32.7 to 9.7. Grip power changed from 74.1% to 93.3%. Modified Mayo wrist score was 83.8. Preoperative SL angle was 53.8˚, postoperative 49.6˚ and the last was 51.2˚. The improvement between preoperative and last follow-up SL angle was statistically significant (p<0.05). The SL angle from operation to last follow-up has increased by 1.67˚ (39.8%, p<0.01). Twelve cases showed SL gap over 2 mm in clenched fist view before operation, but no case at follow-up. Conclusion: Volar and dorsal SL reconstruction using FCR should provide more stability and strength than only dorsal SL reconstruction. Arthroscopic technique has the merit that it causes minimal damage to extrinsic carpal ligaments and posterior interosseous nerve and can provide rapid recovery. [ABSTRACT FROM AUTHOR]

DNN is expanding its use in embedded systems such as robots and autonomous vehicles. For high recognition accuracy, computational complexity is greatly increased, and multiple DNNs are running aperiodically. Therefore, the ability processing multiple DNNs in embedded environments is a crucial issue. Accordingly, multicore based platforms are being released. However, most DNN models are operated in a batch process, and when multiple DNNs are operated in multicore together, the execution time deviation between each DNN may be large and the end-to-end execution time of the whole DNNs could be long depending on how they are allocated to the cores. In this paper, we solve these problems by providing a framework that decompose each DNN into individual layers and then distribute to multicores through a global queue. As a result of the experiment, the total DNN execution time was reduced by 31%, and when operating multiple identical DNNs, the deviation in execution time was reduced by up to 95.1%. [ABSTRACT FROM AUTHOR]

Objectives This study investigated associations between symptom severity and neurocognitive functions in patients with major depressive disorder using comprehensive neuropsychological tests. Methods Ten patients with low depression (LD) and 22 patients with high depression (HD) participated. Symptom severity was determined by both the Beck Depression Inventory and the Hamilton Depression Rating Scale. Additionally, anxiety was measured by the Korean version of the State-Trait Anxiety Inventory (STAI-KYZ). Ten subtests of the Korean-Wechsler Adult Intelligence scale (K-WAIS), Stroop test, the word/design fluency tests were administered to assess cognition. Results The LD and HD groups did not differ in any of cognitive measures but anxiety level. Namely, the HD group obtained significantly higher scores on the state (U = 29.50, p ＜ 0.01) and trait (U = 28.50, p ＜ 0.001) anxiety scales than the LD group. A significant interaction effect between trait anxiety and depression severity was observed on the picture arrangement subtest of the K-WAIS [F (1, 27) = 5.09, p ＜ 0.05]. Conclusions Cognitive deficits observed in patients with major depressive disorder may be related to other factors rather than current symptom severity. Trait anxiety possibly moderates the effect of current symptom severity on social judgment in these patients. [ABSTRACT FROM AUTHOR]

This study investigated the cortical generators of P300 in college students with schizotypal traits by using an auditory oddball paradigm, event-related potentials (ERPs) and standardized low resolution brain electromagnetic tomography (sLORETA) model. We also investigated the relationship between the current density of P300 and the clinical symptoms of schizophrenia. Based on the scores of Schizotypal Personality Questionnaire(SPQ), schizotypal trait (n=37) and control (n=42) groups were selected. For the measurement of P300, an auditory oddball paradigm, in which frequent standard tones (1000Hz) and rare target tones (1500Hz) were presented randomly, was used. Participants were required to count the number of the target tones during the task and report this at the end of the experiment. The two groups did not differ significantly in the accuracy of the oddball task. The schizotypal trait group showed significantly smaller P300 amplitudes than control group. In terms of source localization, both groups showed the P300 current density over bilateral frontal, parietal, temporal and occipital lobes. However, the schizotypal trait group showed significantly reduced activations in the left superior temporal gyrus and the right middle temporal gyrus, but increased activations in both left inferior frontal gyrus and right superior frontal gyrus compared to the control group. Furthermore, a negative correlation between the current density of the right superior frontal gyrus and SPQ disorganization score was found in the schizotypal trait group. These findings indicate that the individuals with schizotypal traits have dysfunctions of frontal and temporal areas, which are known to be the source of P300, as observed in patients with schizophrenia. In addition, the present results indicate that the disorganization score, rather than total score, of the SPQ is useful in predicting the risk of future schizophrenia. [ABSTRACT FROM AUTHOR]