심박변이도를 이용한 적응적 뉴로 퍼지 감정예측 모형에 관한 연구

박성수,이건창 한국디지털정책학회 2019 디지털융복합연구 Vol.17 No.1

An accurate prediction of emotion is a very important issue for the sake of patient-centered medical device development and emotion-related psychology fields. Although there have been many studies on emotion prediction, no studies have applied the heart rate variability and neuro-fuzzy approach to emotion prediction. We propose ANFEP(Adaptive Neuro Fuzzy System for Emotion Prediction) HRV. The ANFEP bases its core functions on an ANFIS(Adaptive Neuro-Fuzzy Inference System) which integrates neural networks with fuzzy systems as a vehicle for training predictive models. To prove the proposed model, 50 participants were invited to join the experiment and Heart rate variability was obtained and used to input the ANFEP model. The ANFEP model with STDRR and RMSSD as inputs and two membership functions per input variable showed the best results. The result out of applying the ANFEP to the HRV metrics proved to be significantly robust when compared with benchmarking methods like linear regression, support vector regression, neural network, and random forest. The results show that reliable prediction of emotion is possible with less input and it is necessary to develop a more accurate and reliable emotion recognition system. 감정을 정확히 예측하는 것은 환자중심의 의료디바이스 개발 및 감성관련 산업에서 매우 중요한 이슈이다. 감정 예측에 관한 많은 연구 중 감정 예측에 심박 변동성과 뉴로-퍼지 접근법을 적용한 연구는 없다. 본 연구는 HRV를 이용한 ANFEP(Adaptive Neuro Fuzzy system for Emotion Prediction)을 제안한다. ANFEP의 핵심 기능은 인공 신경망과 퍼지 시스템을 통합해 예측 모델을 학습하는 ANFIS(Adaptive Neuro-Fuzzy Inference System)에 기반한다. 제안 모형의 검증을 위해 50명의 실험자를 대상으로 청각자극으로 감정을 유발하고, 심박변이도를 구하여 ANFEP 모형에 입력하였다. STDRR과 RMSSD를 입력으로 하고 입력변수 당 2개의 소속함수로 하는 ANFEP모형이 가장 좋은 결과를 나타났다. 제안한 감정예측 모형을 선형회귀 분석, 서포트 벡터 회귀, 인공신경망, 랜덤 포레스트와 비교한 결과 본 제안모형이 가장 우수한 성능을 보였다. 연구 결과는 보다 적은 입력으로 신뢰성 높은 감정인식이 가능함을 입증했고, 이를 활용해 보다 정확하고 신뢰성 높은 감정인식 시스템 개발에 대한 연구가 필요하다.

심박변이도 기반 감정예측 인공신경망을 이용한 감정예측 추론과정 메커니즘에 관한 연구

박성수(Sung Soo Park),이건창(Kun Chang Lee) 한국컴퓨터정보학회 2017 韓國컴퓨터情報學會論文誌 Vol.22 No.7

Correct prediction of emotion is essential for developing advanced health devices. For this purpose, neural network has been successfully used. However, interpretation of how a certain emotion is predicted through the emotion prediction neural network is very tough. When interpreting mechanism about how emotion is predicted by using the emotion prediction neural network can be developed, such mechanism can be effectively embedded into highly advanced health-care devices. In this sense, this study proposes a novel approach to interpreting how the emotion prediction neural network yields emotion. Our proposed mechanism is based on HRV (heart rate variability) measurements, which is based on calculating physiological data out of ECG (electrocardiogram) measurements. Experiment dataset with 23 qualified participants were used to obtain the seven HRV measurement such as Mean RR, SDNN, RMSSD, VLF, LF, HF, LF/HF. Then emotion prediction neural network was modelled by using the HRV dataset. By applying the proposed mechanism, a set of explicit mathematical functions could be derived, which are clearly and explicitly interpretable. The proposed mechanism was compared with conventional neural network to show validity.

ON THE EFFECT OF EMOTIONAL UNCERTAINTY ON PREDICTED UTILITY AND FORECASTING ERROR: THE UNCERTAINTY-PREDICTION ASYMMETRY (UPA) HYPOTHESIS

Athanasios Polyportis,Flora Kokkinaki 글로벌지식마케팅경영학회 2018 Global Marketing Conference Vol.2018 No.07

The present research examines the Uncertainty-Prediction Asymmetry (UPA) hypothesis, that low certainty incidental emotions, compared to their high certainty counterparts, lead to utility overprediction and to lower forecasting error. Introduction Cognitive appraisals of emotion have been included in the state-of-the-art theory of emotion and decision-making (Lerner & Keltner, 2000; Lerner, Li, Valdesolo, & Kassam, 2015). For instance, Tiedens & Linton (2001) discuss how happiness involves appraisals of high certainty, and sadness involves appraisals of low certainty. In terms of forecasting, systematic processing is generally considered to lead to less forecasting error compared to heuristic processing. Tiedens & Linton (2001) argue that, if accuracy is the ultimate goal the individual needs to rely on more thoughtful processes. Seeking a state of certainty is more cognitively engaging and requires more cognitive resources. But how do people predict future utilities in the first place? Theoretical background Kahneman & Thaler (2006) analyze forecasting as a two-step procedure, encompassing a current prediction as well as a future event. Breaking down the present and future situation allows researchers to assess accuracy and detect how errors occur. Kahneman & Snell (1992) report that people tend to underpredict future utilities. Typically, the experienced utility is higher (i.e. more liked or less disliked) compared to the earlier prediction. In the present paper we argue that emotional uncertainty leads to utility overprediction and thus reduces forecasting error. This hypothesis is in line with the Appraisal-Tendency Framework (ATF-overview in Lerner et al., 2015). According to the ATF, an emotion may trigger a cognitive predisposition to assess future events in line with the central appraisal dimensions that triggered that emotion. Such appraisals provide a perceptual schema for interpreting subsequent situations. In the context of the present research, the certainty-uncertainty cognitive appraisal is hypothesized to trigger a predisposition that affects the utility prediction mechanism and leads to utility overprediction. This hypothesis is also in line with the uncertainty intensification hypothesis (Bar-Anan, Wilson, & Gilbert, 2009), according to which the uncertainty of experienced emotions makes unpleasant events more unpleasant and pleasant events more pleasant. The present research examines an Uncertainty-Prediction Asymmetry (UPA) hypothesis. In three experimental studies we test the hypotheses that low certainty incidental emotions, compared to their high certainty counterparts, lead to utility overprediction (H1) and to lower forecasting error (H2). Emotional certainty, as an appraisal dimension of emotions, is expected to create a prediction asymmetry through its effect on both predicted utility and forecasting error. The mediating role of heuristic processing in the relationship between emotional certainty and forecasting error is also investigated. Experiment 1 The first experiment examines the hypothesis that low emotional certainty leads to utility overprediction (H1). Eighty postgraduate students were randomly assigned to a high emotional certainty (disgust) vs. a low emotional certainty (fear) condition. Emotion induction involved exposure to pretested video clips (see Han et al., 2012). Following this manipulation, the experimental utility (a small candy bar) was distributed and participants were encouraged to consume it (see Kahneman & Snell, 1992). They were then asked to report on 13-point scales how much they liked the utility and to predict how much they would like it in the future consumption occasion (a week later). The results revealed a significant difference in predicted utility between the high (M = 2.22, SD = 1.33) and low (M =3.65, SD = 1.37) emotional certainty conditions (F = 4.43, p = 0.04, partial eta squared = 0.10). Experiment 2 The second experiment includes a “future event”, that is measures of the utility that was originally predicted, in order to also estimate forecasting error. The experiment therefore tests if (a) the main effect of emotional uncertainty on predicted utility is confirmed (H1) and (b) there is a significant main effect of emotional uncertainty on forecasting error (H2). In addition, this experiment examines whether these effects are independent of the valence appraisal dimension of emotions. Given that Experiment 1 involved two negatively valenced emotions, emotional valence (positive vs. negative) was included in the experimental design. Seventy three postgraduate students participated in a five-consecutive-days experiment. During the first day, participants were randomly assigned to a fear (negative valence, low certainty), disgust (negative valence, high certainty), hope (positive valence, low certainty) or happiness (positive valence, high certainty) condition. Specifically, participants were asked to report an experience in which they had felt this particular emotion through an Autobiographical Emotional Memory Task (AEMT) (as in Smith & Ellsworth, 1985). Following this experimental manipulation, the experimental utility (a small chocolate bar) was distributed and they were again encouraged to consume. Subsequently, they were asked to rate how much they liked and how much they would like the utility on the fifth day. Depth of processing was assessed with four items (α=0.77), adjusted from Griffin et al. (2002). Specifically, these items measured the heuristic processing performed during the prediction process. Participants were contacted again on each of the remaining four days and were asked to consume the utility and to complete a short questionnaire (comprising ratings of the consumption experience and of the predicted utility on the fifth day). The results reported here involve only the data obtained on the first and final day of the experiment, and the forecasting error was estimated as the difference between the experienced utility of the last day and the predicted utility of the first day. In line with hypothesis H1, emotional certainty had a significant main effect on predicted utility (F = 6.18, p = 0.002, partial eta squared = 0.08). Specifically, predicted utility in the low emotional certainty condition was higher (M = 2.69, SD = 1.09), compared to that of the high certainty condition (M = 0.78, SD = 1.66). There was no significant interaction effect between certainty and valence. These findings provide further support for our H1 and indicate that emotional certainty influences utility prediction irrespective of the valence of incidental emotions. Moreover, a significant main effect of certainty on forecasting error was observed (F = 4.16, p = 0.045, partial eta squared = 0.06). Forecasting error was lower in the low certainty condition (M = 0.59, SD = 1.28) compared to the high certainty condition (M = 2.19, SD = 1.48). There was no significant interaction effect. Moreover, a mediation analysis revealed that heuristic processing mediated the effect of certainty on forecasting error (p**<0.05). Experiment 3 The previous two experiments indicate that the effects of incidental emotional states on predicted utility and forecasting error may be due to the certainty-appraisal dimension of these emotional states. A possible criticism and an inherent limitation of Experiments 1 and 2 might lie on the possibility that these effects are not independent of the other appraisal dimensions. This is related to a key methodological issue. In Experiments 1 and 2, the induced emotions were different in terms of certainty or uncertainty, but these emotions might have differed in other ways and across other appraisal dimensions as well. To eliminate this possibility and to strengthen our argument, we employ here a manipulation of the certainty appraisal of the same emotion. We therefore compare predicted utility and forecasting error in the same emotional state under conditions of low and high certainty. In Experiments 1 and 2 the emotions induced are strong representatives of each side of the certainty appraisal dimension. However, emotions located in the middle of this dimension provide an interesting opportunity since they might allow us to compare their effects when they are associated with lower or higher levels of certainty. In this experiment we have chosen to focus on the emotional state of sadness. Sadness was selected because it is near the middle of the certainty-uncertainty dimension (Smith & Ellsworth, 1985). Similar manipulations of sadness have been reported in the literature (Tiedens & Linton, 2001). Sixty postgraduate students were randomly assigned to a low vs. high certainty sadness condition. High certainty participants were asked to recall and report an experience or event in which they had felt high certainty sadness (i.e. during which they understood what was happening and could predict what was going to happen next), through an Autobiographical Emotional Memory Task (AEMT) as in Experiment 2. Similarly, low certainty participants were asked to recall and report an event or experience that had generated low certainty sadness. Following the experimental manipulation, the experimental utility (a small chocolate bar) was served. Participants were again encouraged to consume some of it and were asked to complete 13-point ratings of how much they liked it and how much they would like it in the future occasion (a week later). Eight items (α=0.81), adapted from Griffin et al. (2002), measured the heuristic processing performed during the prediction process. Participants also completed ten items adjusted from PANAS questionnaire (Watson et al., 1988). A week later, participants consumed the utility and completed a short questionnaire. The results revealed a significant main effect of certainty on the predicted utility (F = 4.00, p = 0.05, partial eta squared = 0.06). Predicted utility in the low certainty sadness condition was higher (M = 4.21, SD = 1.55) compared to that of the high certainty condition (M = 3.35, SD = 1.78). A significant main effect of certainty on forecasting error was also observed (F = 5.04, p = 0.03, partial eta squared = 0.10). Forecasting error in the low certainty condition (M = -0.10, SD = 1.65) was lower compared to that of the high certainty condition (M = 1.02, SD = 1.81). A mediation analysis revealed that heuristic processing again mediated the effect of certainty on forecasting error (p**<0.05). Conclusion The contribution of this research is mostly highlighted by the counter-intuitive findings that lower certainty emotions lead to judgment with higher accuracy, as well as to an overprediction of utilities, related to their certainty counterparts. Therefore, the current findings provide support for the proposed Uncertainty-Prediction dual Asymmetry (UPA) hypothesis. Future research needs to corroborate these findings, to clarify the mechanisms underlying the observed asymmetry and to identify boundary conditions.

기계학습을 이용한 회화 감성 예측 모델에 관한 분석 연구

이태민 차세대컨버전스정보서비스학회 2021 디지털예술공학멀티미디어논문지 Vol.8 No.3

Techniques for predicting emotions in images have been studied a lot. As machine learning and deep learning technologies developed, more studies were conducted. Among the images, artworks in particular are very related to emotions. In general, artists often put their emotions into their works. Emotions are controlled by artistic features such as symmetry and composition, which combine physical elements such as color and texture. In this study, these features are extracted and analyzed from paintings. Features that are expected to affect emotions in paintings are extracted and used to predict emotions. Various machine learning models are built by extracted physical features such as color, line, texture, etc. and artistic features such as symmetry and color combination from a given painting. Through the built machine learning models, this paper analyze which machine learning models are suitable for the most relevant characteristics and emotional extraction in conversation-emotional predictions. Finally, we verify the legitimacy and accuracy of machine learning models by comparing them with predictive models based on deep learning. 이미지에서 감성을 예측하는 기술들은 많이 연구되어 지고 있다. 기계학습 및 딥러닝 기술들이 발전함에 따라서, 더 많은 연구들이 진행되었다. 이미지중에서도 특히 예술작품들은 감성과의 연관이 매우 크다. 일반적으로 예술가들이 자신의 감성을 작품에 넣는 경우가 많기 때문이다. 이런 감성들은 색상, 질감 등의 물리적 요소들이 결합된 대칭성, 구도 등의 예술적 요소들로 제어가 된다. 본 연구에서는 이런 특징들을 회화로부터 추출 및 분석한다. 회화에서 감성에 영향을 미칠 것으로 예상되는 특징들을 추출하여 이를 감성 예측에 활용한다. 주어진 회화로부터 색상, 선, 질감등의 물리적 특징과, 대칭성, 색상조합 등과 같은 예술적 특징을 추출하여, 다양한 기계학습 모델을 제작한다. 제작된 기계학습 모델들을 통해 회화-감성 예측에서 가장 관련이 깊은 특징들 및 감성 추출에 어울리는 기계학습 모델이 무엇인지 분석한다. 최종적으로 딥러닝 기반의 예측 모델과의 비교를 통해 기계학습 모델의 정당성 및 정확도에 대해 검증한다.

공조방식에 의한 예상 온열감 반응(PMV) 변화에 따른 심리/생리적 감성반응의 변화

김보성 ( Bo Seong Kim ),민윤기 ( Yoon Ki Min ),민병찬 ( Byung Chan Min ),김진호 ( Jin Ho Kim ) 한국감성과학회 2011 감성과학 Vol.14 No.4

This study examined changes of both psychological and physiological emotional responses according to change of the PMV (predicted mean vote) in the heating and the cooling air conditions. For this purpose, the changes of PMV were induced by the heating and cooling operations of the HVAC (heating, ventilation, and air conditioning) systems. In addition, positive/negative and arousal/relaxation were measured as the participant`s psychological emotional responses, and HR (heart rate) was measured as the participant`s physiological emotional responses. As a result, in same range of the PMV, both psychological and physiological emotional-responses were changed by air conditioning. It is suggested that occupant`s emotional responses would depend on the operational conditions of heating and cooling in indoor thermal environments, and both psychological and physiological emotional response should be considered when occupants try to match the indoor thermal environments to their thermal expectations.

영화 시나리오와 영화촬영기법을 이용한 감정 예측 시스템

김진수 한국융합학회 2018 한국융합학회논문지 Vol.9 No.12

Recently, we are trying to predict the emotion from various information and to convey the emotion information that the supervisor wants to inform the audience. In addition, audiences intend to understand the flow of emotions through various information of non-dialogue parts, such as cinematography, scene background, background sound and so on. In this paper, we propose to extract emotions by mixing not only the context of scripts but also the cinematography information such as color, background sound, composition, arrangement and so on. In other words, we propose an emotional prediction system that learns and distinguishes various emotional expression techniques into dialogue and non-dialogue regions, contributes to the completeness of the movie, and quickly applies them to new changes. The precision of the proposed system is improved by about 5.1% and 0.4%, and the recall is improved by about 4.3% and 1.6%, respectively, when compared with the modified n-gram and morphological analysis. 최근에 다양한 정보로부터 감정을 예측하여 청중에게 감독이 알리고자 하는 정보를 빠르게 전달하고자 한다. 또한, 청중은 감독의 의도를 대화 내용에 나타나는 대사뿐만 아니라, 영상내의 다양한 정보인 촬영 기법, 장면의 배경, 배경 음악 등을 통해 비대사 구간에서도 감정의 흐름을 이해하려고 한다. 본 논문에서는 대사와 같은 문맥의 상황뿐만 아니라, 촬영 영상에 담아낸 색상, 음향, 구도, 배치 등에 의해 표현된 정보를 혼합하여 감정을 추출하고자 한다. 즉, 다양한 감정 표현 기법을 대사 구간, 비대사 구간으로 나누어 학습하고 판별하여 영상의 완성도에 기여하고 새로운 변화에 빠르게 적용할 수 있는 감정 예측 시스템을 제안한다. 본 논문에서 제안한 감정 예측시스템이 변형된 n-gram 방식과 형태소 분석을 적용한 사례와 비교했을 때, 정확도는 약 5.1%, 0.4% 향상되었고, 재현율은 약 4.3%, 1.6% 향상되었다.

사용자 감정 예측을 통한 상황인지 추천시스템의 개선

안현철(Hyunchul Ahn) 한국데이타베이스학회 2014 Journal of information technology applications & m Vol.21 No.4

This study proposes a novel context-aware recommender system, which is designed to recommend the items according to the customer’s responses to the previously recommended item. In specific, our proposed system predicts the user’s emotional state from his or her responses (such as facial expressions and movements) to the previous recommended item, and then it recommends the items that are similar to the previous one when his or her emotional state is estimated as positive. If the customer’s emotional state on the previously recommended item is regarded as negative, the system recommends the items that have characteristics opposite to the previous item. Our proposed system consists of two sub modules-(1) emotion prediction module, and (2) responsive recommendation module. Emotion prediction module contains the emotion prediction model that predicts a customer’s arousal level-a physiological and psychological state of being awake or reactive to stimuli-using the customer’s reaction data including facial expressions and body movements, which can be measured using Microsoft’s Kinect Sensor. Responsive recommendation module generates a recommendation list by using the results from the first module-emotion prediction module. If a customer shows a high level of arousal on the previously recommended item, the module recommends the items that are most similar to the previous item. Otherwise, it recommends the items that are most dissimilar to the previous one. In order to validate the performance and usefulness of the proposed recommender system, we conducted empirical validation. In total, 30 undergraduate students participated in the experiment. We used 100 trailers of Korean movies that had been released from 2009 to 2012 as the items for recommendation. For the experiment, we manually constructed Korean movie trailer DB which contains the fields such as release date, genre, director, writer, and actors. In order to check if the recommendation using customers’ responses outperforms the recommendation using their demographic information, we compared them. The performance of the recommendation was measured using two metrics-satisfaction and arousal levels. Experimental results showed that the recommendation using customers’ responses (i.e. our proposed system) outperformed the recommendation using their demographic information with statistical significance.

유아 정서지식의 기초학습능력에 대한 단기종단적 예측력에 관한 연구

이옥경 ( Ok Kyoung Lee ),이진희 ( Jin Hee Lee ) 한국아동교육학회 2015 아동교육 Vol.24 No.3

본 연구는 유아의 정서지식이 기초학습능력을 예측할 수 있는지 알아보고자 단기종단연구로 실시되었다. 이를 위해 만 4, 5세 유아 159명을 대상으로 1차시기에는 정서지식 검사와 기초학습능력 검사로 수학능력검사와 어휘력검사를 실시하였다. 그리고 1년 후 1차시기에 검사한 정서지식이 1년후의 기초학습능력을 예측할 수 있는지 알아보기 위해 기초학습능력 검사로 수학능력검사와 어휘력 검사를 실시하였다. 수집된 자료는 기술통계분석, 단기종단적 예측력 모형 적합도, 구조모형을 통해 정서지식이 기초학습능력에 미치는 영향이 시간의 흐름에 따라 어떻게 변화하는지를 알아보았다. 그 결과 유아의 정서지식은 기초학습능력을 단기종단적으로 예측할 수 있는 것으로 나타났다. 즉 본 연구의 결과는 유아의 정서지식이 기초학습능력에 지속적인 영향을 줌으로써 차후의 학습능력을 예측할 수 있음을 제안한다. 이는 유아의 정서지식이 학습에 영향을 줄 수 있고 그 영향이 지속적이고 안정적일 수 있음을 의미한다고 할 수 있다. 따라서 유아교육기관이나 부모들이 이 시기의 유아와 상호작용할 때 자신과 타인의 정서를 이해하고 적절하게 대처할 수 있는 능력을 지원해주는 것이 중요하다는 것을 시사한다. This study aimed at exploring the short-term longitudinal predictive power of children`s emotion knowledge on their basic academic abilities. Tests were conducted on 159 children aged 4 and 5 to measure their emotion knowledge, and mathematical ability and vocabulary as basic academic abilities in the first round of the study. In the second round of the study, 1 year later, the same children were tested again for mathematical ability and vocabulary level, in order to find out if the emotion knowledge tested 1 year before can predict their basic academic abilities the next year. The data collected were analyzed through descriptive statistics, short-term longitudinal predictive models fit and structural equation modeling, to figure out how the influence of emotion knowledge on basic academic abilities may change as time goes by. The results showed that children`s emotion knowledge can predict their basic academic abilities longitudinally in the short term. It is indicated that children`s emotion knowledge may influence their learning consistently and stably to predict their future academic achievement. The results imply that when early childhood educators and parents interact with young children, they need to support the children to better understand their own and others` emotions and respond effectively.

Deep neural network predicts emotional responses of the human brain from functional magnetic resonance imaging

Kim, Hyun-Chul,Bandettini, Peter A.,Lee, Jong-Hwan Elsevier 2019 NeuroImage Vol.186 No.-

<P><B>Abstract</B></P> <P>An artificial neural network with multiple hidden layers (known as a deep neural network, or DNN) was employed as a predictive model (DNN<SUB>p</SUB>) for the first time to predict emotional responses using whole-brain functional magnetic resonance imaging (fMRI) data from individual subjects. During fMRI data acquisition, 10 healthy participants listened to 80 International Affective Digital Sound stimuli and rated their own emotions generated by each sound stimulus in terms of the arousal, dominance, and valence dimensions. The whole-brain spatial patterns from a general linear model (i.e., beta-valued maps) for each sound stimulus and the emotional response ratings were used as the input and output for the DNN<SUB>P</SUB>, respectively. Based on a nested five-fold cross-validation scheme, the paired input and output data were divided into training (three-fold), validation (one-fold), and test (one-fold) data. The DNN<SUB>P</SUB> was trained and optimized using the training and validation data and was tested using the test data. The Pearson's correlation coefficients between the rated and predicted emotional responses from our DNN<SUB>P</SUB> model with weight sparsity optimization (mean ± standard error 0.52 ± 0.02 for arousal, 0.51 ± 0.03 for dominance, and 0.51 ± 0.03 for valence, with an input denoising level of 0.3 and a mini-batch size of 1) were significantly greater than those of DNN models with conventional regularization schemes including elastic net regularization (0.15 ± 0.05, 0.15 ± 0.06, and 0.21 ± 0.04 for arousal, dominance, and valence, respectively), those of shallow models including logistic regression (0.11 ± 0.04, 0.10 ± 0.05, and 0.17 ± 0.04 for arousal, dominance, and valence, respectively; average of logistic regression and sparse logistic regression), and those of support vector machine-based predictive models (SVM<SUB>p</SUB>s; 0.12 ± 0.06, 0.06 ± 0.06, and 0.10 ± 0.06 for arousal, dominance, and valence, respectively; average of linear and non-linear SVM<SUB>p</SUB>s). This difference was confirmed to be significant with a Bonferroni-corrected <I>p</I>-value of less than 0.001 from a one-way analysis of variance (ANOVA) and subsequent paired <I>t</I>-test. The weights of the trained DNN<SUB>P</SUB>s were interpreted and input patterns that maximized or minimized the output of the DNN<SUB>P</SUB>s (i.e., the emotional responses) were estimated. Based on a binary classification of each emotion category (e.g., high arousal vs. low arousal), the error rates for the DNN<SUB>P</SUB> (31.2% ± 1.3% for arousal, 29.0% ± 1.7% for dominance, and 28.6% ± 3.0% for valence) were significantly lower than those for the linear SVM<SUB>P</SUB> (44.7% ± 2.0%, 50.7% ± 1.7%, and 47.4% ± 1.9% for arousal, dominance, and valence, respectively) and the non-linear SVM<SUB>P</SUB> (48.8% ± 2.3%, 52.2% ± 1.9%, and 46.4% ± 1.3% for arousal, dominance, and valence, respectively), as confirmed by the Bonferroni-corrected <I>p</I> < 0.001 from the one-way ANOVA. Our study demonstrates that the DNN<SUB>p</SUB> model is able to reveal neuronal circuitry associated with human emotional processing – including structures in the limbic and paralimbic areas, which include the amygdala, prefrontal areas, anterior cingulate cortex, insula, and caudate. Our DNN<SUB>p</SUB> model was also able to use activation patterns in these structures to predict and classify emotional responses to stimuli.</P> <P><B>Highlights</B></P> <P> <UL> <LI> Deep neural network (DNN) was trained to predict the human emotion measured from fMRI. </LI> <LI> The prediction performance of the DNN was superior to that of the support vector machine. </LI> <LI> Weight representation and input pattern estimation were introduced to interpret the trained DNN. </LI> <LI> Brain re

연구논문 : 생물분류탐구과정에서 호르몬 변화를 이용한 부정감성예측모델 개발

박진선 ( Jin Sun Park ),이일선 ( Il Sun Lee ),이준기 ( Jun Ki Lee ),권용주 ( Yong Ju Kwon ) 경북대학교 과학교육연구소 2010 科學敎育硏究誌 Vol.34 No.2

이 연구의 목적은 생물분류탐구과정에서 나타나는 부정감성을 호르몬 변화로 예측할 수 있는 부정감성예측모델을 개발하는 것이다. 이를 위해 통합적인 과학 탐구가 가능하도록 깃털 분류 탐구 활동을 개발하였다. 연구대상은 호르몬 변화 측정에 문제가 없는 서울, 안산, 청주 소재 중학교 2학년 47명(남 18, 여 29)으로 하였다. 피험자들은 개인별로 깃털 분류 탐구 활동을 수행하였다. 깃털 분류 탐구 활동 전과 후에 형용사 이모티콘 척도법을 이용하여 부정감성 검사를 하였고, 타액 시료를 채취하여 코르티솔 호르몬 변화를 분석하였다. 연구결과 부정감성 변화량과 타액 코르티솔 변화량 사이에서 유의미한 정적 상관관계(R=0.39, P<0.001)가 나타났으며, 회귀분석을 이용하여 생물분류탐구에서 나타나는 타액 코르티솔 변화량을 이용한 부정감성 예측모델을 개발하였다. The purpose of this study was to develope the negative-emotion prediction model by hormonal changes during the scientific inquiry. For this study, biological classification task was developed that are suitable for comprehensive scientific inquiry. Forty-seven 2nd grade secondary school students (boy 18, girl 29) were participated in this study. The students are healthy for measure hormonal changes. The students performed the feathers classification task individually. Before and after the task, the strength of negative emotion was measured using adjective emotion check lists and they extracted their saliva sample for salivary hormone analysis. The results of this study, student`s change of negative emotion during the feathers classification process was significant positive correlation(R=0.39, P<0.001) with student`s salivary cortisol concentration. According to this results, we developed the negative emotion prediction model by salivary cortisol changes.