• 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.

• Korean Journal of Biological Psychiatry
• /
• v.17 no.3
• /
• pp.127-135
• /
• 2010

Objectives : This study was conducted to compare between hallucination group and delusion group in patient with schizophrenia, using Brain $^{99m}Tc$-ECD PECT. Methods : Among 16 patients with less than 3 schizophrenic episodes, 8 patients whose initial symptom was hallucination were assigned to the hallucination group, and other 8 patients with initial sumptom of delusion were assinged to the hallucination group. All of the patients clinically evaluated using the PANSS and BPRS. Both groups of patients and 8 healthy subjects underwent $^{99m}Tc$-ECD PECT. Results : Score of thinking disturbance subscale of BPRS were significantly lower in the hallucination group than the delusion group. In SPECT analysis, the hallucination group showed significantly increased perfusion in some areas of the right temporal lobe, bilateral limbic lobes and left parietal lobe compared to delusion group. Both group had a reduced rCBF in some areas of the frontal lobe. Conclusion : The hallucniation group, compared with the delusion group, showed significantly increased regional cerebral blood flow in some regions. Therefore, this data suggests that different neural substrates may affect the process of auditory hallucination and delusion.

• Sleep Medicine and Psychophysiology
• /
• v.8 no.2
• /
• pp.98-106
• /
• 2001

Auditory hallucinations are cardinal feature of psychosis. But the mechanism of hallucinated speech is unknown. The hypothesis that these hallucinations arise from pathologically altered brain monitoring system underlying speech perception is influential. With the help of rapidly developing neuroimaging study technologies, many researchers have been finding new organic deficits in the hallucinated schizophrenic patient's brain. In this article, we reviewed the general appearance of hallucination, a computer simulation model of hallucination and several neuroimaging study findings on hallucinating schizophrenic patients. In conclusion, we presented the presumptive mechanism of hallucination based on the anatomical dysfunction of schizophrenia.

• Anxiety and mood
• /
• v.9 no.2
• /
• pp.113-120
• /
• 2013

Objectives : The aim of this study was to investigate clinical meaning of hallucination among patients with posttraumatic stress disorder (PTSD). Methods : PTSD patients with hallucination (n=50) were selected and matched with the PTSD patients without hallucination (n=50) by age and sex. The patients were evaluated in two different aspects, pathologic aspects which can develop after traumatic event, and positive/adaptive aspects which are concerned in positive psychology. In addition to these, temperament and character features between two groups were also evaluated using the Temperament and Character Inventory (TCI). Results : Except for the total score of DES (Dissociative Experience Scale) and the score of re-experience subscale of CAPS (Clinician Administered PTSD Scale), clinical characteristics did not show any significant difference between the two groups. Among the comparison of seven higher dimensions in TCI profiles, only Self-transcendence showed significant difference between the two groups. Conclusion : The two groups did not show significant differences in pathologic aspects except for factors that may contribute to development of hallucination. Furthermore, authors could not find any differences in positive/adaptive aspects.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.461-464
• /
• 2023

최근 GPT-3 와 LLaMa 같은 생성형 거대 언어모델을 활용한 서비스가 공개되었고, 실제로 많은 사람들이 사용하고 있다. 해당 모델들은 사용자들의 다양한 질문에 대해 유창한 답변을 한다는 이유로 주목받고 있다. 하지만 LLMs 의 답변에는 종종 Inconsistent content 와 non-factual statement 가 존재하며, 이는 사용자들로 하여금 잘못된 정보의 전파 등의 문제를 야기할 수 있다. 이에 논문에서는 동일한 질문에 대한 LLM 의 답변 샘플과 외부 지식을 활용한 Hallucination Detection 방법을 제안한다. 제안한 방법은 동일한 질문에 대한 LLM 의 답변들을 이용해 일관성 점수(Consistency score)를 계산한다. 거기에 외부 지식을 이용한 사실검증을 통해 사실성 점수(Factuality score)를 계산한다. 계산된 일관성 점수와 사실성 점수를 활용하여 문장 수준의 Hallucination Detection 을 가능하게 했다. 실험에는 GPT-3 를 이용하여 WikiBio dataset 에 있는 인물에 대한 passage 를 생성한 데이터셋을 사용하였으며, 우리는 해당 방법을 통해 문장 수준에서의 Hallucination Detection 성능이 baseline 보다 AUC-PR scores 에서 향상됨을 보였다.

• 대한상한금궤의학회지
• /
• v.6 no.1
• /
• pp.79-88
• /
• 2014

Objective : We report efficacy of Daeseunggi-tang on auditory hallucination in schizophrenia. From this, we can expect therapeutic possibility of herbal medicine chosen by 'A disease pattern identification diagnostic system based on Shanghanlun provisions' in schizophrenia. Methods : Daeseunggi-tang written in YangMyeong-JiWiByeong is applied to 31-year-old patient who diagnosed as schizophrenia and hospitalized for 3months taking antipsychotics. The patient was treated for a year. Results : Scores in Auditory Hallucination appended to PANSS and K-BDI were highly decreased. The patient takes no antipsychotics anymore for now. Conclusions : Auditory hallucination in Schizophrenia which needs antipsychotics can be managed by herbal medicine chosen by 'A disease pattern identification diagnostic system based on Shanghanlun provisions'.

• Anxiety and mood
• /
• v.12 no.1
• /
• pp.7-12
• /
• 2016

Objective : This study was conducted to evaluate the relationships between selective attention bias for fear stimuli and hallucination in patients with schizophrenia Method : A total of 66 patients with schizophrenia admitted to psychiatry clinics were included in the study. Selective attention bias was measured by the dot-probe task. Patient symptoms were measured using the Positive and Negative Symptom Scale, Psychotic Symptom Rating Scale (PSYRATS), Korean version of the Scale to Assess Unawareness of Mental Disorder, and Clinical Global Impression-Severity scale. Results : Selective attention bias was correlated with the hallucination subscale of PSYTATS (r=0.268, p=0.029). No correlation was found between selective attention bias and other clinical measures. There was no significant difference, but a statistical trend was found (p=0.092) in hallucination severities between the biased and non-biased groups. Conclusion : The results suggest that selective attention bias for fear stimuli is associated with auditory hallucination. This preliminary study suggests the possibility of correlation between auditory hallucination in the psychotic domain and anxiety of the affective component.

• Korean Journal of Biological Psychiatry
• /
• v.22 no.4
• /
• pp.163-172
• /
• 2015

Objectives This study was aimed to examine the multidimensional relationship between auditory verbal hallucinations (AVHs) and Positive and Negative Syndrome Scale (PANSS) factors of psychopathology in the patients with schizophrenia. And we explored the differences between assessments to hallucination by the clinicians and patients. Methods 82 patients with schizophrenia who were assessed by the Hamilton Program for Schizophrenia Voices Questionnaire (HPSVQ), Psychotic Symptom Rating Scale-Auditory Hallucination (PSYRATS-AHS), and the PANSS were recruited. Hwang's five-factor model of PANSS, items and total scores of hallucination scales, Kim's and Haddock's factor models of hallucination were applied to examine the correlations between psychopathology and AVHs. AVH-positive patients was 50 in PANSS-HPSVQ group and 24 in PANSS-PSYRATS-AHS. These two groups were separately analyzed. Results Among the five factors of the PANSS, negative and depression/anxiety factors were correlated with the total scores of HPSVQ and PSYRATS-AHS, and positive and autistic preoccupation factors were correlated only with the total score of PSYRATS-AHS. The activation factor was correlated with none of the total scores of HPSVQ/PSYRATS-AHS. These correlation patterns of a total score of HPSVQ/PSYRATS-AHS were same in the emotional factor of HPSVQ and physical factor of PSYRATS-AHS respectively. In the items which showed significant correlations, correlation coefficients of PANSS-PSYRATS-AHS group ranged between 0.406-0.755 and those of PANSS-HPSVQ ranged between 0.283-0.420. Conclusions This study suggested that the psychopathological domains of schizophrenia were differentially correlated with AVHs and the assessment of AVHs by clinicians and patients showed substantial differences which should be integrated into the therapeutic interventions.

• Proceedings of the KIEE Conference
• /
• 2008.10b
• /
• pp.292-293
• /
• 2008

In this paper, we propose a face hallucination method based on example-learning. The traditional approach based on example-learning requires alignment of face images. In the proposed method, facial images are segmented into patches and the weights are computed to represent input low resolution facial images into weighted sum of low resolution example images. High resolution facial images are hallucinated by combining the weight vectors with the corresponding high resolution patches in the training set. Experimental results show that the proposed method produces more reliable results of face hallucination than the ones by the traditional approach based on example-learning.

• 대한상한금궤의학회지
• /
• v.10 no.1
• /
• pp.69-80
• /
• 2018

Objective : Patients complaining of dizziness and olfactory hallucination who were diagnosed and treated with the Shanghanlun disease pattern identification diagnostic system were analyzed to report cases. Methods : This is a case report on a patient who has been suffering from Dizziness andolfactory hallucination. According to Diagnostic system based on Shanghanlun provisions, Gyeji-tang was provided. The severity of Dizziness was evaluated by NRS. Results : When analyzing the patient's history and symptoms, it can be found in article No. 42 in Shanghanlun. In the first case, NRS changed from 8 to 2. In the second case, NRS change 8 to 1. Conclusions : Gyeji-tang can be effective in dizziness and olfactory hallucination with the Shanghanlun disease pattern identification diagnostic system.