• Informatization Policy
• /
• v.31 no.2
• /
• pp.3-38
• /
• 2024

Hallucination is a significant barrier to the utilization of large-scale language models or multimodal models. In this study, we collected 654 computer science papers with "hallucination" in the abstract from arXiv from December 2022 to January 2024 following the advent of Chat GPT and conducted frequency analysis, knowledge network analysis, and literature review to explore the latest trends in hallucination research. The results showed that research in the fields of "Computation and Language," "Artificial Intelligence," "Computer Vision and Pattern Recognition," and "Machine Learning" were active. We then analyzed the research trends in the four major fields by focusing on the main authors and dividing them into data, hallucination detection, and hallucination mitigation. The main research trends included hallucination mitigation through supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF), inference enhancement via "chain of thought" (CoT), and growing interest in hallucination mitigation within the domain of multimodal AI. This study provides insights into the latest developments in hallucination research through a technology-oriented literature review. This study is expected to help subsequent research in both engineering and humanities and social sciences fields by understanding the latest trends in hallucination research.

• Geophysics and Geophysical Exploration
• /
• v.13 no.1
• /
• pp.40-50
• /
• 2010

The building process of any geological map involves linking sparse lithological outcrop information with equally sparse geometrical measurements, all in a single entity which is the preferred interpretation of the field geologist. The actual veracity of this interpretative map is partially dependent upon the frequency and distribution of geological outcrops compounded by the complexity of the local geology. Geophysics is commonly used as a tool to augment the distribution of data points, however it normally does not have sufficient geometrical constraints due to: a) all geophysical inversion models being inherently non-unique; and b) the lack of knowledge of the physical property contrasts associated with specific lithologies. This contribution proposes the combined use of geophysical edge detection routines and 'three point' solutions from topographic data as a possible approach to obtaining geological contact geometry information (strike and dip), which can be used in the construction of a preliminary geological model. This derived geological information should first be assessed for its compatibility with the scale of the problem, and any directly observed geological data. Once verified it can be used to help constrain the preferred geological map interpretation being developed by the field geologist. The method models the contacts as planar surfaces. Therefore, it must be ensured that this assumption fits the scale and geometry of the problem. Two examples are shown from folded sequences at the Bathurst Mining Camp, New Brunswick, Canada.

• Korean Journal of Remote Sensing
• /
• v.34 no.5
• /
• pp.787-799
• /
• 2018

Accurate and effective mapping is critical step to monitor the spatial distribution and change of flood inundated area in large scale flood event. In this study, we try to suggest methods to use low spatial resolution satellite optical imagery for flood mapping, which has high temporal resolution to cover wide geographical area several times per a day. We selected the Sebou watershed flood in Morocco that was occurred in early 2010, in which several hundred $km^2$ area of the Gharb lowland plain was inundated. MODIS daily surface reflectance product was used to detect the flooded area. The study area showed several distinct spectral patterns within the flooded area, which included pure turbid water and turbid water with vegetation. The flooded area was extracted by thresholding on selected band reflectance and water-related spectral indices. Accuracy of these flooding detection methods were assessed by the reference map obtained from Landsat-5 TM image and qualitative interpretation of the flood map derived. Over 90% of accuracies were obtained for three methods except for the NDWI threshold. Two spectral bands of SWIR and red were essential to detect the flooded area and the simple thresholding on these bands was effective to detect the flooded area. NIR band did not play important role to detect the flooded area while it was useful to separate the water-vegetation mixed flooded classes from the purely water surface.

• Korean Journal of Remote Sensing
• /
• v.36 no.5_1
• /
• pp.749-762
• /
• 2020

Precipitation is closely related to various hydrometeorological phenomena, such as runoff and evapotranspiration. In Korean Peninsula, observing rainfall intensity using weather radar and rain gauge network is dominating due to their accurate, intuitive and precise detecting power. However,since these methods are not suitable at ungauged regions, rainfall detection using satellite is required. Satellite-based rainfall data has coarse spatial resolution (10 km, 25 km), and has a limited range of usage due to its reliability of data. The aim of this study is to obtain finer scale precipitation. Especially, to make the applicability of satellite higher at ungauged regions, 10 km satellite-based rainfall data was downscaled to 1 km data using MODerate Resolution Imaging Spectroradiometer (MODIS) based cloud property. Downscaled precipitation was verified in urban region, which has complex topographical and environmental characteristics. Correlation coefficient was similar in summer (+0), decreased in spring (-0.08) and autumn (-0.01), and increased in winter (+0.04) season compared to Global Precipitation Measurement (GPM) based precipitation. Downscaling without calibration using in situ data could be useful in areas where rain gauge system is not sufficient or ground observations are rarely available.

• Korean Journal of Biological Psychiatry
• /
• v.23 no.2
• /
• pp.63-68
• /
• 2016

Objectives Haenyeo are Korean professional women breath-hold divers in Jeju island. The aim of this study was to investigate the clinical characteristics of depressed Haenyeo group, compared to non-Haenyeo depressed group. Methods This study included 75 Haenyeo and 340 non-Haenyeo with depressive disorders recruited from the Dementia Early Detection Program in Jeju island. Structural diagnostic interviews were performed using the Korean version of Mini International Neuropsychiatric Interview. All patients completed the questionnaires, including the Subjective Memory Complaints Questionnaire (SMCQ), the Patient Health Questionnaire-15 (PHQ-15), and the Blessed dementia scale. Depression was evaluated by the Korean version of short form the Geriatric Depression Scale (K-SGDS) and cognition was assessed by the Korean version of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) assessment packet. Results Although the mean scores of the K-SGDS were similar between Haenyeo and non-Haenyeo depressed groups, the Haenyeo group showed a higher mean score on the PSQ-15 (p < 0.001, ANCOVA adjusting for age, the K-SGDS and education). The Haenyeo group showed poorer performance on the Korean Version of Frontal Assessment Batter (p < 0.001), the Mini-Mental State Examination in the Korean version of the CERAD Assessment Packet (p < 0.018), the word fluency test (p < 0.001), and the word list memory test (p = 0.012) in ANCOVA adjusting for age and education. The mean SMCQ score was higher in the Haenyeo depressed group than in the non-Haenyeo depressed group. Conclusions The Haenyeo depressed group shows cognitive dysfunction, especially frontal lobe dysfunction, compared to the non-Haenyeo depressed group, indicating the Haenyeo depressed group may have more severe frontolimbic dysfunction due to chronic exposure to hypoxia. The Haenyeo depressed group suffers more somatic symptoms than the non-Haenyeo depressed group.

• Journal of the Korean Academy of Child and Adolescent Psychiatry
• /
• v.4 no.1
• /
• pp.114-123
• /
• 1993

The authors applied Toddler Temperament Scale(TTS) to ADHD patient group, other disease group and normal control group to elucidate temperamental characteristics of ADHD patients in their toddler stage. 1) ADHD group showed significant difference from the control group in 'activity', 'rhythmicity', 'approach', 'adaptability', 'intensity', 'persistence' and 'threshold' categories. 2) ADHD group and other disease showed significant difference from the control group in 'adaptability' and 'persistence' categories. 3) ADHD group was significantly different from other disease group or control group in 'activity' and 'approach' categories. 4) In total scores of TTS, ADHD group and other diease group showed significant difference from control group. 5) Classification ratio of TTS for ADHD by discriminant analysis was 84.3%. Above results suggest that the ADHD patients may reveal characteristic traits in their toddler stage Further research should be focused on the development and refinement of assessment tools for the early detection and prevention for ADHD.

• Journal of Broadcast Engineering
• /
• v.17 no.5
• /
• pp.838-850
• /
• 2012

In this paper, we propose a novel algorithm that generates multi-view images by using various image features obtained from the given stereoscopic images. In the proposed algorithm, we first create an intensity gradient saliency map from the given stereo images. And then we calculate a block-based optical flow that represents the relative movement(disparity) of each block with certain size between left and right images. And we also obtain the disparities of feature points that are extracted by SIFT(scale-invariant We then create a disparity saliency map by combining these extracted disparity features. Disparity saliency map is refined through the occlusion detection and removal of false disparities. Thirdly, we extract straight line segments in order to minimize the distortion of straight lines during the image warping. Finally, we generate multi-view images by grid mesh-based image warping algorithm. Extracted image features are used as constraints during grid mesh-based image warping. The experimental results show that the proposed algorithm performs better than the conventional DIBR algorithm in terms of visual quality.

• The Journal of Korean Obstetrics and Gynecology
• /
• v.23 no.2
• /
• pp.106-115
• /
• 2010

Purpose: The purposes of this study were in understanding maternal and neonatal risk factors for postpartum depression using Edinburgh Postnatal Depression Scale(EPDS). Methods: Among 788 women, who had delivery include cesarean section in the department of obstetrics and gynecology at OO medical center from May 28th 2008 to October 6st 2009, 72 women filled out EPDS questionnaire sheets. Additional aspects included for the analysis are maternal factors including age, number of children, parity, delivery method, and hemoglobin; and neonatal factors such as weight, sex, gestational age, apgar score, and neonatal intensive care unit admission. Comparison was performed between the women with EPDS score equal or less than 8 and the women with EPDS score equal to or higher than 9 using statistical methods of student t-test for linear variables and chi-square test for non-linear variables. SPSS version 13.0 for windows was used for analysis. Results: Thirty women(41.7%) were included in the postpartum depression risk group (EPDS score ${\geqq}9$). Statistically significant difference(P<0.05) was found in gestational ages of the risk group($36.57{\pm}29.6$ weeks) and the non-risk group ($38.10{\pm}1.97$ weeks). Identified statistically significant risk factors(P<0.05) include cesarean section (OR=3.304 [1.121-9.744]), low birth weight infant(OR =6.500 [1.606-26.314]), preterm delivery(OR=2.857[1.071-7.621]), low apgar score (1minute) after delivery (OR=14.909 [1.750-127.025]). There was no statistically significant difference in maternal age, number of children, parity, hemoglobin, neonatal sex, apgar score (5minutes), NICU admission. Conclusions: Through the results showed, gestational age, delivery method, neonatal weight, apgar score(1minute) were identified as risk factors for postpartum depression. To prevent or minimize postpartum depression, oriental medical intervention is recommended for pregnant women through early detection.

• Journal of Korean Society of Environmental Engineers
• /
• v.28 no.7
• /
• pp.697-703
• /
• 2006

Endocrine disruptors were measured with GC/MS in effluents discharged from sewage treatment processes in pilot scale for the purpose of water reuse. From that analysis, we compared the removal rate of them by treatment processes. Nonylphenol was mainly detected in effluents and high concentration from 0.36 to 0.94 ${\mu}g/L$. $17{\beta}$-estradiol(E2) and $17{\alpha}$-ethynylestradiol(EE2) were detected as below the limit of detection in effluent. Endocrine disruptors were removed effectively in the range from 50 to 100% by treatment process. EC50 value($9.0{\times}10^{-3}$ M) of $17{\beta}$-estradiol(E2) by dose response curve of E-screen assay has higher than that of bisphenol A($2.736{\times}10^{-5}M$) and p-octylphenol($9.760{\times}10^{-6}$ M). These results showed that alkylphenols have lower relative estrogen potency than other estrogens such as $17{\beta}$-estradiol(E2). Calculated estrogenic activity(ng-EEQ/L) was 2 times higher than measured total estrogenic activity which estimated by E-screen assay. Moreover estrogenic activity of effluent by treatment process showed very low as below 1 ng-EEQ/L.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.19 no.2
• /
• pp.89-103
• /
• 2020

As we face an aging society, the demand for personal mobility for disabled and aged people is increasing. In fact, as of 2017, the number of electric wheelchair in the country continues to increase to 90,000. However, people with disabilities and seniors are more likely to have accidents while driving, because their judgment and coordination are inferior to normal people. One of the causes of the accident is the interference of personal vehicle steering control due to unbalanced road surface conditions. In this paper, we introduce a encoder type semantic segmentation algorithm that can recognize road conditions at high speed to prevent such accidents. To this end, more than 1,500 training data and 150 test data including road surface damage were newly secured. With the data, we proposed a deep neural network composed of encoder stages, unlike the Auto-encoding type consisting of encoder and decoder stages. Compared to the conventional method, this deep neural network has a 4.45% increase in mean accuracy, a 59.2% decrease in parameters, and an 11.9% increase in computation speed. It is expected that safe personal transportation will be come soon by utilizing such high speed algorithm.