Wei-tsung Kao - Academia.edu (original) (raw)

Papers by Wei-tsung Kao

Research paper thumbnail of TACAS2019 Artifact Evaluation for paper "J-ReCoVer: Java Reducer Commutativity Verifier

This archive is for the artifact evaluation for the paper "J-ReCoVer: Java Reducer Commutati... more This archive is for the artifact evaluation for the paper "J-ReCoVer: Java Reducer Commutativity Verifier". In this paper, we propose effective heuristics for reducer commutativity analysis and implement them as a tool named J-ReCoVer (Java Reducer Commutativity Verifier). To our knowledge, J-ReCoVer is the first tool specialized in checking reducer commutativity.<br><br>The archive contains the scripts and the benchmarks used to replicate the results showed in the paper, more specifically, to redo the experiments and produce the three result tables and the diagram that we showed in the evaluation chapter of the paper. There are step-by-step instructions included in the README file for how to run the scripts.

Research paper thumbnail of The benefit of clinical psychologists in prevention from the suicide in one hospital in Taiwan, Republic of China

life science journal, 2015

Research paper thumbnail of The benefit of clinical psychologists in prevention from the suicide in one hospital in Taiwan, Republic of China

life science journal, 2015

PURPOSE: To investigate the suicide intention by using the 5-items Brief Symptoms Rating Scales (... more PURPOSE: To investigate the suicide intention by using the 5-items Brief Symptoms Rating Scales (BSRS) to screen high risk of patients with suicidal intention and compare the benefits from clinical psychologists. METHODS: All 80 patients with BSRS score > 10 were advised to be hospitalized and were divided into 4 groups. In group 1 (n =17), patients with score>15 received the care from psychologists. There were 23 patients in group 2 (n=23; total BSRS>15) without psychologist's help. 16 subjects with score ≦ 15 (>10) in group 3 did not seek help from psychologists. In group 4 (n=24), they had lower score (≦ 15, but >10) with counseling from psychologists. We compared the score change in the 4 groups. RESULTS: In group 1, the mean score was 17.7 ± 0.2 before admission, and the mean score after therapy was 9.9 ± 0.8 (P0.05). In group 3, the score showed 16.9 ± 0.4 before admission and it revealed 13.9 ± 0.8 (P>0.05) after medical care. There are 13.3 ± 1.2 and 7....

Research paper thumbnail of Poster #S229 DOSING AND CLINICAL STRATEGIES OF PALIPERIDONE ER IN ACUTELY EXACERBATED SCHIZOPHRENIA: AN EXPERT CONSENSUS IN TAIWAN

Schizophrenia Research, 2014

Research paper thumbnail of Development and Implementation of Societal Influences Survey Questionnaire (SISQ) for Peoples during COVID-19 Pandemic: A Validity and Reliability Analysis

International Journal of Environmental Research and Public Health, 2020

The emergence of Coronavirus disease 2019 (COVID-19) had rapidly spread since FEB/MAR 2020. Polic... more The emergence of Coronavirus disease 2019 (COVID-19) had rapidly spread since FEB/MAR 2020. Policy to prevent transmission of COVDI-19 resulted in multi-dimensional impact on social interaction. We aimed to develop a beneficial survey tool with favorable quality and availability, the Societal Influences Survey Questionnaire (SISQ), to evaluate social influences on people during this pandemic. The SISQ was developed with 15 items and 4-point Likert scales consisting of five factors. These include social distance, social anxiety, social desirability, social information, and social adaptation. Construct validity and reliability were performed to verify the SISQ. A total of 1912 Taiwanese were recruited. The results demonstrated that the SISQ has acceptable reliability, with Cronbach’s alphas ranging between 0.57 and 0.76. The SISQ accounted for 58.86% and satisfied the requirement of Kaiser–Mayer–Olkinvalues (0.78) and significant Bartlett’s Test of sphericity. Moreover, the confirmato...

Research paper thumbnail of Predictors for Depression, Sleep Disturbance, and Subjective Pain among Inpatients with Depressive Disorders during the COVID-19 Pandemic: A Cross-Sectional Study

International Journal of Environmental Research and Public Health, 2021

The coronavirus disease 2019 (COVID-19) pandemic can have a negative impact on patients with mood... more The coronavirus disease 2019 (COVID-19) pandemic can have a negative impact on patients with mood disorders. The aim of this study is to explore the societal influence of COVID-19 and associated impacts on levels of depression, sleep disturbance, and subjective pain among patients with mood disorders. This cross-sectional study recruited inpatients with depression and bipolar disorder. Levels of depression, sleep disturbance, subjective pain, and related demographic variables were collected through self-reported questionnaires. Potential factors associated with levels of depression, sleep disturbance, and subjective pain were identified using univariate linear regression and further entered into a stepwise multivariate linear regression model to identify the independent predictors. A total of 119 participants were included in the analysis, of whom 50.42% had bipolar disorder and 49.58% had unipolar depression. Multivariate analysis showed that a higher level of depression was associ...

Research paper thumbnail of Cognitive function and alcohol use disorder: Path analysis for a cross-sectional study in Taiwan

Taiwanese Journal of Psychiatry, 2021

Objective: Alcohol has cognitive impacts on patients with alcohol use disorder (AUD). In this stu... more Objective: Alcohol has cognitive impacts on patients with alcohol use disorder (AUD). In this study, we intended to study cognitive impairments in patients with AUD and their potential interrelationships. Methods: We enrolled 60 patients with AUD or alcohol intoxication in Taiwan. The severity of alcohol use was assessed using a copy for severity of alcohol dependence questionnaire (SADQ). Cognitive function was evaluated using Stroop color and word test, continuous performance test-identical pairs, trail making test, visual alternating and divided attention subscales of computerized everyday attention test, visual elevator subscale of test of everyday attention, Benton judgment of line orientation test, spatial perception subscale of visual object perception test, visual motor organization subscale of Loewenstein occupational therapy, thinking operations subscale of Loewenstein occupational therapy cognitive assessment, digit symbol coding subscale of Wechsler adult intelligence scale-third edition, as well as symbol digit modalities test. Moreover, we used a structural equation modeling (SEM) to link age, duration of alcohol use, SADQ, and cognitive impairments. Results: Patients with AUD had significantly impairments of “attention” (p < 0.01), “visual motor coordination” (p < 0.001), and “executive function” (p < 0.01). SEM analysis showed that the higher level of attention, visual motor coordination, and executive functional impairments were significantly linked with old age (p < 0.01), long duration of alcohol use (p < 0.01), and higher score of SADQ (p < 0.01). Furthermore, we found that the three cognitive impairments were positively associated with each other. Conclusion: Old age, long duration of alcohol use, and severe alcohol use were the predictors of cognitive impairments and early detection. The early screening of predictive factors and timely interventions should be considered to improve cognitive function in patients with AUD.

Research paper thumbnail of Membership Inference Attacks Against Self-supervised Speech Models

Interspeech 2022

Recently, adapting the idea of self-supervised learning (SSL) on continuous speech has started ga... more Recently, adapting the idea of self-supervised learning (SSL) on continuous speech has started gaining attention. SSL models pre-trained on a huge amount of unlabeled audio can generate general-purpose representations that benefit a wide variety of speech processing tasks. Despite their ubiquitous deployment, however, the potential privacy risks of these models have not been well investigated. In this paper, we present the first privacy analysis on several SSL speech models using Membership Inference Attacks (MIA) under black-box access. The experiment results show that these pre-trained models are vulnerable to MIA and prone to membership information leakage with high adversarial advantage scores in both utterance-level and speaker-level. Furthermore, we also conduct several ablation studies to understand the factors that contribute to the success of MIA.

Research paper thumbnail of DDOS: A MOS Prediction Framework utilizing Domain Adaptive Pre-training and Distribution of Opinion Scores

Interspeech 2022

Mean opinion score (MOS) is a typical subjective evaluation metric for speech synthesis systems. ... more Mean opinion score (MOS) is a typical subjective evaluation metric for speech synthesis systems. Since collecting MOS is time-consuming, it would be desirable if there are accurate MOS prediction models for automatic evaluation. In this work, we propose DDOS, a novel MOS prediction model. DDOS utilizes domain-adaptive pre-training to further pre-train selfsupervised learning models on synthetic speech. And a proposed module is added to model the opinion score distribution of each utterance. With the proposed components, DDOS outperforms previous works on BVCC dataset. And the zeroshot transfer result on BC2019 dataset is significantly improved. DDOS also wins second place in Interspeech 2022 VoiceMOS challenge in terms of system-level score.

Research paper thumbnail of On the Efficiency of Integrating Self-supervised Learning and Meta-learning for User-defined Few-shot Keyword Spotting

Cornell University - arXiv, Apr 1, 2022

User-defined keyword spotting is a task to detect new spoken terms defined by users. This can be ... more User-defined keyword spotting is a task to detect new spoken terms defined by users. This can be viewed as a few-shot learning problem since it is unreasonable for users to define their desired keywords by providing many examples. To solve this problem, previous works try to incorporate self-supervised learning models or apply meta-learning algorithms. But it is unclear whether self-supervised learning and meta-learning are complementary and which combination of the two types of approaches is most effective for few-shot keyword discovery. In this work, we systematically study these questions by utilizing various self-supervised learning models and combining them with a wide variety of meta-learning algorithms. Our result shows that HuBERT combined with Matching network achieves the best result and is robust to the changes of few-shot examples.

Research paper thumbnail of BERT's output layer recognizes all hidden layers? Some Intriguing Phenomena and a simple way to boost BERT

Cornell University - arXiv, Jan 25, 2020

Although Bidirectional Encoder Representations from Transformers (BERT) have achieved tremendous ... more Although Bidirectional Encoder Representations from Transformers (BERT) have achieved tremendous success in many natural language processing (NLP) tasks, it remains a black box. A variety of previous works have tried to lift the veil of BERT and understand each layer's functionality. In this paper, we found that surprisingly the output layer of BERT can reconstruct the input sentence by directly taking each layer of BERT as input, even though the output layer has never seen the input other than the final hidden layer. This fact remains true across a wide variety of BERT-based models, even when some layers are duplicated. Based on this observation, we propose a quite simple method to boost the performance of BERT. By duplicating some layers in the BERT-based models to make it deeper (no extra training required in this step), they obtain better performance in the downstream tasks after fine-tuning.

Research paper thumbnail of Is BERT a Cross-Disciplinary Knowledge Learner? A Surprising Finding of Pre-trained Models’ Transferability

Findings of the Association for Computational Linguistics: EMNLP 2021, 2021

This paper investigates whether the power of the models pre-trained on text data, such as BERT, c... more This paper investigates whether the power of the models pre-trained on text data, such as BERT, can be transferred to general token sequence classification applications. To verify pre-trained models' transferability, we test the pre-trained models on text classification tasks with meanings of tokens mismatches, and realworld non-text token sequence classification data, including amino acid, DNA, and music. We find that even on non-text data, the models pre-trained on text converge faster, perform better than the randomly initialized models, and only slightly worse than the models using task-specific knowledge. We also find that the representations of the text and non-text pretrained models share non-trivial similarities.

Research paper thumbnail of TP 53 Codon 72 Polymorphisms in Schizophrenia ( R 1 ) Running Title : TP 53 Pro 72 Arg in schizophrenia

Background: TP53 is a recently be suspected to be involves in neural apoptotic events. Also, TP53... more Background: TP53 is a recently be suspected to be involves in neural apoptotic events. Also, TP53 codon72 and long-form variants of DRD4 uVNTR polymorphism are reported to be conferred susceptibility to schizophrenia, respectively. Methods: Nine hundred and thirty-four schizophrenia patients and 433 healthy individuals were recruited, genotyped for TP53 codon72 and DRD4 uVNTR polymorphism by using PCR, PCR-RFLP and direct sequencing. Results: No significant difference was found in both genotype and allele frequency of TP53 codon72 polymorphism between schizophrenia patients and controls. However, long-form alleles (嗢5-repeat) of DRD4 uVNTR polymorphism were more frequent in schizophrenia patients (P = 0.001), which may a risk factor to be associated with enhanced vulnerability to schizophrenia (OR = 3.189, 95% C.I. = 1.535-6.622). In logistic regression, long-from variants did predict schizophrenia after controlling age and gender (P = 0.036, OR = 2.319), but not the genotype CC or ...

Research paper thumbnail of Priapism after Long-acting Risperidone Injection

Objective: Priapism is a persistent, painful erection that occurs without sexual stimulation. It ... more Objective: Priapism is a persistent, painful erection that occurs without sexual stimulation. It has been associated with the use of many antipsychotic agents, including clozapine, olanzapine and risperidone but has not been reported as a side effect of long-acting risperidone injection. Here we report a case of this adverse event in a patient treated with long-acting risperidone. Case report: A 27-yearold man with a 10-year history of paranoid-type schizophrenia developed priapism two weeks after a long-acting risperidone injection. After treatment, he denied any erectile abnormalities and had had no subsequent episodes of priapism during 6 months of follow up. Conclusion: Priapism rarely results from treatment with antipsychotic agents. Clinicians, however, must be aware of the possibility of this complication when prescribing these drugs.

Research paper thumbnail of Utilizing Self-Supervised Representations for MOS Prediction

Interspeech 2021, 2021

Speech quality assessment has been a critical issue in speech processing for decades. Existing au... more Speech quality assessment has been a critical issue in speech processing for decades. Existing automatic evaluations usually require clean references or parallel ground truth data, which is infeasible when the amount of data soars. Subjective tests, on the other hand, do not need any additional clean or parallel data and correlates better to human perception. However, such a test is expensive and time-consuming because crowd work is necessary. It thus becomes highly desired to develop an automatic evaluation approach that correlates well with human perception while not requiring ground truth data. In this paper, we use selfsupervised pre-trained models for MOS prediction. We show their representations can distinguish between clean and noisy audios. Then, we fine-tune these pre-trained models followed by simple linear layers in an end-to-end manner. The experiment results showed that our framework outperforms the two previous state-of-the-art models by a significant improvement on Voice Conversion Challenge 2018 and achieves comparable or superior performance on Voice Conversion Challenge 2016. We also conducted an ablation study to further investigate how each module benefits the task. The experiment results are implemented and reproducible with publicly available toolkits 1 .

Research paper thumbnail of A positive relationship between Apo ε2 allele and high-density lipoprotein cholesterol

Nutrition Research, 2006

This study explored the relationship between ApoE genotypes and the fasting serum levels of total... more This study explored the relationship between ApoE genotypes and the fasting serum levels of total cholesterol, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol, very-low-density lipoprotein cholesterol, and triacylglycerol in an elderly Chinese population. A total of 500 subjects aged 65 to 74 years were randomly selected in southern Taiwan from April to June 2001. Two hundred fifty-six participants, who agreed to have additional venous blood withdrawal for genetic study, were regarded as the participant group, and the other 244 participants, who disagreed to have additional venous blood withdrawal, were included in the control group. In the participant group, the most prevalent ApoE allele was the E3 allele (89.65%), followed by the E4 allele (7.81%), and the E2 allele (2.54%). After holding sex, age, education, and income factors as constant, only the E2 allele associated with plasma HDL-C was statistically significant. Elderly women tended to have higher plasma total cholesterol and HDL-C than elderly men. The Apo E2 allele had an increasing effect on plasma HDL-C relative to the non-Apo E2 allele. Such a relationship is particularly important in investigating the roles of genetic and environmental factors in cardiovascular disease.

Research paper thumbnail of No Allele Variation of the MAOA Gene Promoter in Male Chinese Subjects with Attention Deficit Hyperactivity Disorder

Neuropsychobiology, 2006

The aim of the current study was to examine any difference in the repeat sequence of the monoamin... more The aim of the current study was to examine any difference in the repeat sequence of the monoamine oxidase A (MAOA) gene promoter between males with attention deficit hyperactivity disorder (ADHD) and randomly selected subjects in a community. The role of the MAOA gene promoter in ADHD cases was also investigated. The total of 244 participants included 57 male ADHD subjects as the case group and 187 males, also in southern Taiwan, as the community controls. There was no significantly different distribution in the repeat sequence of the MAOA gene promoter (χ2 = 3.895, d.f. = 3, p = 0.273), and no significantly different distribution of ‘long-form’ and ‘short-form’ alleles between the ADHD group and the male community group was noted (χ2 = 2.484, d.f. = 1, p = 0.115). Some aspects of clinical response are mentioned in the discussion of this study which are worth exploring further in the future.

Research paper thumbnail of Protective effect of the apo e2 allele in major depressive disorder in Taiwanese

Acta Psychiatrica Scandinavica, 2006

Lung F-W. Protective effect of the apo e2 allele in major depressive disorder in Taiwanese.

Research paper thumbnail of The Societal Influences and Quality of Life Among Healthcare Team Members During the COVID-19 Pandemic

Frontiers in Psychiatry, 2021

Background: The coronavirus infection disease 2019 (COVID-19) pandemic is likely to put healthcar... more Background: The coronavirus infection disease 2019 (COVID-19) pandemic is likely to put healthcare professionals across the world in an unprecedented situation.Methods: A total of 683 healthcare workers were recruited in this study. Short form-12 items (SF-12), Societal Influences Survey Questionnaire (SISQ), and Disaster-Related Psychological Screening Test (DRPST) were used to survey participants. Multiple linear regression and structural equation model (SEM) were used to explore the possible factors to the societal influences and quality of life.Results: After multiple linear regression analysis, female, older, more education years, married, regular intake, and posttraumatic stress disorder (PTSD) frequency had positive association with SISQ. To physical component summary (PCS) of SF-12, chronic illness, sleep score, PTSD frequency, and social distance had negative association, and exercise habits had positive association. A mental component summary (MCS) value of SF-12, age, par...

Research paper thumbnail of Association of DRD4 uVNTR and TP53 codon 72 polymorphisms with schizophrenia: a case-control study

BMC Medical Genetics, 2009

Background: The tumour supressor gene TP53 is thought to be involved in neural apoptosis. The pol... more Background: The tumour supressor gene TP53 is thought to be involved in neural apoptosis. The polymorphism at codon 72 in TP53 and the long form variants of the upstream variable number of tandem repeats (uVNTR) polymorphism in the dopamine D4 receptor (DRD4) gene are reported to confer susceptibility to schizophrenia. Methods: We recruited 934 patients with schizophrenia and 433 healthy individuals, and genotyped the locus of the TP53 codon 72 and DRD4 uVNTR polymorphisms by combining the polymerase chain reaction-restriction fragment length polymorphism method (PCR-RFLP) with direct sequencing. Results: No significant differences were found in the frequency of the genotype of the TP53 codon72 polymorphism between patients with schizophrenia and their controls. However, the long form alleles (≥ 5 repeats) of the DRD4 uVNTR polymorphism were more frequent in patients with schizophrenia than in controls (p = 0.001). Hence, this class of alleles might be a risk factor for enhanced vulnerability to schizophrenia (odds ratio = 3.189, 95% confidence interval = 1.535-6.622). In the logistic regression analysis, the long form variants of the DRD4 polymorphism did predict schizophrenia after the contributions of the age and gender of the subjects were included (p = 0.036, OR = 2.319), but the CC and GG genotypes of the codon 72 polymorphism of TP53 did not. Conclusions: The long form variants of the uVNTR polymorphism in DRD4 were associated with schizophrenia, in a manner that was independent of the TP53 codon 72 polymorphism. In addition, given that the genetic effect of the TP53 codon 72 polymorphism on the risk of developing schizophrenia was very small, this polymorphism is unlikely to be associated with schizophrenia. The roles that other single nucleotide polymorphisms (SNPs) in the TP53 gene or in other apoptosis-related genes play in the synaptic dysfunction involved in the pathogenesis of schizophrenia should be investigated.

Research paper thumbnail of TACAS2019 Artifact Evaluation for paper "J-ReCoVer: Java Reducer Commutativity Verifier

This archive is for the artifact evaluation for the paper "J-ReCoVer: Java Reducer Commutati... more This archive is for the artifact evaluation for the paper "J-ReCoVer: Java Reducer Commutativity Verifier". In this paper, we propose effective heuristics for reducer commutativity analysis and implement them as a tool named J-ReCoVer (Java Reducer Commutativity Verifier). To our knowledge, J-ReCoVer is the first tool specialized in checking reducer commutativity.<br><br>The archive contains the scripts and the benchmarks used to replicate the results showed in the paper, more specifically, to redo the experiments and produce the three result tables and the diagram that we showed in the evaluation chapter of the paper. There are step-by-step instructions included in the README file for how to run the scripts.

Research paper thumbnail of The benefit of clinical psychologists in prevention from the suicide in one hospital in Taiwan, Republic of China

life science journal, 2015

Research paper thumbnail of The benefit of clinical psychologists in prevention from the suicide in one hospital in Taiwan, Republic of China

life science journal, 2015

PURPOSE: To investigate the suicide intention by using the 5-items Brief Symptoms Rating Scales (... more PURPOSE: To investigate the suicide intention by using the 5-items Brief Symptoms Rating Scales (BSRS) to screen high risk of patients with suicidal intention and compare the benefits from clinical psychologists. METHODS: All 80 patients with BSRS score > 10 were advised to be hospitalized and were divided into 4 groups. In group 1 (n =17), patients with score>15 received the care from psychologists. There were 23 patients in group 2 (n=23; total BSRS>15) without psychologist's help. 16 subjects with score ≦ 15 (>10) in group 3 did not seek help from psychologists. In group 4 (n=24), they had lower score (≦ 15, but >10) with counseling from psychologists. We compared the score change in the 4 groups. RESULTS: In group 1, the mean score was 17.7 ± 0.2 before admission, and the mean score after therapy was 9.9 ± 0.8 (P0.05). In group 3, the score showed 16.9 ± 0.4 before admission and it revealed 13.9 ± 0.8 (P>0.05) after medical care. There are 13.3 ± 1.2 and 7....

Research paper thumbnail of Poster #S229 DOSING AND CLINICAL STRATEGIES OF PALIPERIDONE ER IN ACUTELY EXACERBATED SCHIZOPHRENIA: AN EXPERT CONSENSUS IN TAIWAN

Schizophrenia Research, 2014

Research paper thumbnail of Development and Implementation of Societal Influences Survey Questionnaire (SISQ) for Peoples during COVID-19 Pandemic: A Validity and Reliability Analysis

International Journal of Environmental Research and Public Health, 2020

The emergence of Coronavirus disease 2019 (COVID-19) had rapidly spread since FEB/MAR 2020. Polic... more The emergence of Coronavirus disease 2019 (COVID-19) had rapidly spread since FEB/MAR 2020. Policy to prevent transmission of COVDI-19 resulted in multi-dimensional impact on social interaction. We aimed to develop a beneficial survey tool with favorable quality and availability, the Societal Influences Survey Questionnaire (SISQ), to evaluate social influences on people during this pandemic. The SISQ was developed with 15 items and 4-point Likert scales consisting of five factors. These include social distance, social anxiety, social desirability, social information, and social adaptation. Construct validity and reliability were performed to verify the SISQ. A total of 1912 Taiwanese were recruited. The results demonstrated that the SISQ has acceptable reliability, with Cronbach’s alphas ranging between 0.57 and 0.76. The SISQ accounted for 58.86% and satisfied the requirement of Kaiser–Mayer–Olkinvalues (0.78) and significant Bartlett’s Test of sphericity. Moreover, the confirmato...

Research paper thumbnail of Predictors for Depression, Sleep Disturbance, and Subjective Pain among Inpatients with Depressive Disorders during the COVID-19 Pandemic: A Cross-Sectional Study

International Journal of Environmental Research and Public Health, 2021

The coronavirus disease 2019 (COVID-19) pandemic can have a negative impact on patients with mood... more The coronavirus disease 2019 (COVID-19) pandemic can have a negative impact on patients with mood disorders. The aim of this study is to explore the societal influence of COVID-19 and associated impacts on levels of depression, sleep disturbance, and subjective pain among patients with mood disorders. This cross-sectional study recruited inpatients with depression and bipolar disorder. Levels of depression, sleep disturbance, subjective pain, and related demographic variables were collected through self-reported questionnaires. Potential factors associated with levels of depression, sleep disturbance, and subjective pain were identified using univariate linear regression and further entered into a stepwise multivariate linear regression model to identify the independent predictors. A total of 119 participants were included in the analysis, of whom 50.42% had bipolar disorder and 49.58% had unipolar depression. Multivariate analysis showed that a higher level of depression was associ...

Research paper thumbnail of Cognitive function and alcohol use disorder: Path analysis for a cross-sectional study in Taiwan

Taiwanese Journal of Psychiatry, 2021

Objective: Alcohol has cognitive impacts on patients with alcohol use disorder (AUD). In this stu... more Objective: Alcohol has cognitive impacts on patients with alcohol use disorder (AUD). In this study, we intended to study cognitive impairments in patients with AUD and their potential interrelationships. Methods: We enrolled 60 patients with AUD or alcohol intoxication in Taiwan. The severity of alcohol use was assessed using a copy for severity of alcohol dependence questionnaire (SADQ). Cognitive function was evaluated using Stroop color and word test, continuous performance test-identical pairs, trail making test, visual alternating and divided attention subscales of computerized everyday attention test, visual elevator subscale of test of everyday attention, Benton judgment of line orientation test, spatial perception subscale of visual object perception test, visual motor organization subscale of Loewenstein occupational therapy, thinking operations subscale of Loewenstein occupational therapy cognitive assessment, digit symbol coding subscale of Wechsler adult intelligence scale-third edition, as well as symbol digit modalities test. Moreover, we used a structural equation modeling (SEM) to link age, duration of alcohol use, SADQ, and cognitive impairments. Results: Patients with AUD had significantly impairments of “attention” (p < 0.01), “visual motor coordination” (p < 0.001), and “executive function” (p < 0.01). SEM analysis showed that the higher level of attention, visual motor coordination, and executive functional impairments were significantly linked with old age (p < 0.01), long duration of alcohol use (p < 0.01), and higher score of SADQ (p < 0.01). Furthermore, we found that the three cognitive impairments were positively associated with each other. Conclusion: Old age, long duration of alcohol use, and severe alcohol use were the predictors of cognitive impairments and early detection. The early screening of predictive factors and timely interventions should be considered to improve cognitive function in patients with AUD.

Research paper thumbnail of Membership Inference Attacks Against Self-supervised Speech Models

Interspeech 2022

Recently, adapting the idea of self-supervised learning (SSL) on continuous speech has started ga... more Recently, adapting the idea of self-supervised learning (SSL) on continuous speech has started gaining attention. SSL models pre-trained on a huge amount of unlabeled audio can generate general-purpose representations that benefit a wide variety of speech processing tasks. Despite their ubiquitous deployment, however, the potential privacy risks of these models have not been well investigated. In this paper, we present the first privacy analysis on several SSL speech models using Membership Inference Attacks (MIA) under black-box access. The experiment results show that these pre-trained models are vulnerable to MIA and prone to membership information leakage with high adversarial advantage scores in both utterance-level and speaker-level. Furthermore, we also conduct several ablation studies to understand the factors that contribute to the success of MIA.

Research paper thumbnail of DDOS: A MOS Prediction Framework utilizing Domain Adaptive Pre-training and Distribution of Opinion Scores

Interspeech 2022

Mean opinion score (MOS) is a typical subjective evaluation metric for speech synthesis systems. ... more Mean opinion score (MOS) is a typical subjective evaluation metric for speech synthesis systems. Since collecting MOS is time-consuming, it would be desirable if there are accurate MOS prediction models for automatic evaluation. In this work, we propose DDOS, a novel MOS prediction model. DDOS utilizes domain-adaptive pre-training to further pre-train selfsupervised learning models on synthetic speech. And a proposed module is added to model the opinion score distribution of each utterance. With the proposed components, DDOS outperforms previous works on BVCC dataset. And the zeroshot transfer result on BC2019 dataset is significantly improved. DDOS also wins second place in Interspeech 2022 VoiceMOS challenge in terms of system-level score.

Research paper thumbnail of On the Efficiency of Integrating Self-supervised Learning and Meta-learning for User-defined Few-shot Keyword Spotting

Cornell University - arXiv, Apr 1, 2022

User-defined keyword spotting is a task to detect new spoken terms defined by users. This can be ... more User-defined keyword spotting is a task to detect new spoken terms defined by users. This can be viewed as a few-shot learning problem since it is unreasonable for users to define their desired keywords by providing many examples. To solve this problem, previous works try to incorporate self-supervised learning models or apply meta-learning algorithms. But it is unclear whether self-supervised learning and meta-learning are complementary and which combination of the two types of approaches is most effective for few-shot keyword discovery. In this work, we systematically study these questions by utilizing various self-supervised learning models and combining them with a wide variety of meta-learning algorithms. Our result shows that HuBERT combined with Matching network achieves the best result and is robust to the changes of few-shot examples.

Research paper thumbnail of BERT's output layer recognizes all hidden layers? Some Intriguing Phenomena and a simple way to boost BERT

Cornell University - arXiv, Jan 25, 2020

Although Bidirectional Encoder Representations from Transformers (BERT) have achieved tremendous ... more Although Bidirectional Encoder Representations from Transformers (BERT) have achieved tremendous success in many natural language processing (NLP) tasks, it remains a black box. A variety of previous works have tried to lift the veil of BERT and understand each layer's functionality. In this paper, we found that surprisingly the output layer of BERT can reconstruct the input sentence by directly taking each layer of BERT as input, even though the output layer has never seen the input other than the final hidden layer. This fact remains true across a wide variety of BERT-based models, even when some layers are duplicated. Based on this observation, we propose a quite simple method to boost the performance of BERT. By duplicating some layers in the BERT-based models to make it deeper (no extra training required in this step), they obtain better performance in the downstream tasks after fine-tuning.

Research paper thumbnail of Is BERT a Cross-Disciplinary Knowledge Learner? A Surprising Finding of Pre-trained Models’ Transferability

Findings of the Association for Computational Linguistics: EMNLP 2021, 2021

This paper investigates whether the power of the models pre-trained on text data, such as BERT, c... more This paper investigates whether the power of the models pre-trained on text data, such as BERT, can be transferred to general token sequence classification applications. To verify pre-trained models' transferability, we test the pre-trained models on text classification tasks with meanings of tokens mismatches, and realworld non-text token sequence classification data, including amino acid, DNA, and music. We find that even on non-text data, the models pre-trained on text converge faster, perform better than the randomly initialized models, and only slightly worse than the models using task-specific knowledge. We also find that the representations of the text and non-text pretrained models share non-trivial similarities.

Research paper thumbnail of TP 53 Codon 72 Polymorphisms in Schizophrenia ( R 1 ) Running Title : TP 53 Pro 72 Arg in schizophrenia

Background: TP53 is a recently be suspected to be involves in neural apoptotic events. Also, TP53... more Background: TP53 is a recently be suspected to be involves in neural apoptotic events. Also, TP53 codon72 and long-form variants of DRD4 uVNTR polymorphism are reported to be conferred susceptibility to schizophrenia, respectively. Methods: Nine hundred and thirty-four schizophrenia patients and 433 healthy individuals were recruited, genotyped for TP53 codon72 and DRD4 uVNTR polymorphism by using PCR, PCR-RFLP and direct sequencing. Results: No significant difference was found in both genotype and allele frequency of TP53 codon72 polymorphism between schizophrenia patients and controls. However, long-form alleles (嗢5-repeat) of DRD4 uVNTR polymorphism were more frequent in schizophrenia patients (P = 0.001), which may a risk factor to be associated with enhanced vulnerability to schizophrenia (OR = 3.189, 95% C.I. = 1.535-6.622). In logistic regression, long-from variants did predict schizophrenia after controlling age and gender (P = 0.036, OR = 2.319), but not the genotype CC or ...

Research paper thumbnail of Priapism after Long-acting Risperidone Injection

Objective: Priapism is a persistent, painful erection that occurs without sexual stimulation. It ... more Objective: Priapism is a persistent, painful erection that occurs without sexual stimulation. It has been associated with the use of many antipsychotic agents, including clozapine, olanzapine and risperidone but has not been reported as a side effect of long-acting risperidone injection. Here we report a case of this adverse event in a patient treated with long-acting risperidone. Case report: A 27-yearold man with a 10-year history of paranoid-type schizophrenia developed priapism two weeks after a long-acting risperidone injection. After treatment, he denied any erectile abnormalities and had had no subsequent episodes of priapism during 6 months of follow up. Conclusion: Priapism rarely results from treatment with antipsychotic agents. Clinicians, however, must be aware of the possibility of this complication when prescribing these drugs.

Research paper thumbnail of Utilizing Self-Supervised Representations for MOS Prediction

Interspeech 2021, 2021

Speech quality assessment has been a critical issue in speech processing for decades. Existing au... more Speech quality assessment has been a critical issue in speech processing for decades. Existing automatic evaluations usually require clean references or parallel ground truth data, which is infeasible when the amount of data soars. Subjective tests, on the other hand, do not need any additional clean or parallel data and correlates better to human perception. However, such a test is expensive and time-consuming because crowd work is necessary. It thus becomes highly desired to develop an automatic evaluation approach that correlates well with human perception while not requiring ground truth data. In this paper, we use selfsupervised pre-trained models for MOS prediction. We show their representations can distinguish between clean and noisy audios. Then, we fine-tune these pre-trained models followed by simple linear layers in an end-to-end manner. The experiment results showed that our framework outperforms the two previous state-of-the-art models by a significant improvement on Voice Conversion Challenge 2018 and achieves comparable or superior performance on Voice Conversion Challenge 2016. We also conducted an ablation study to further investigate how each module benefits the task. The experiment results are implemented and reproducible with publicly available toolkits 1 .

Research paper thumbnail of A positive relationship between Apo ε2 allele and high-density lipoprotein cholesterol

Nutrition Research, 2006

This study explored the relationship between ApoE genotypes and the fasting serum levels of total... more This study explored the relationship between ApoE genotypes and the fasting serum levels of total cholesterol, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol, very-low-density lipoprotein cholesterol, and triacylglycerol in an elderly Chinese population. A total of 500 subjects aged 65 to 74 years were randomly selected in southern Taiwan from April to June 2001. Two hundred fifty-six participants, who agreed to have additional venous blood withdrawal for genetic study, were regarded as the participant group, and the other 244 participants, who disagreed to have additional venous blood withdrawal, were included in the control group. In the participant group, the most prevalent ApoE allele was the E3 allele (89.65%), followed by the E4 allele (7.81%), and the E2 allele (2.54%). After holding sex, age, education, and income factors as constant, only the E2 allele associated with plasma HDL-C was statistically significant. Elderly women tended to have higher plasma total cholesterol and HDL-C than elderly men. The Apo E2 allele had an increasing effect on plasma HDL-C relative to the non-Apo E2 allele. Such a relationship is particularly important in investigating the roles of genetic and environmental factors in cardiovascular disease.

Research paper thumbnail of No Allele Variation of the MAOA Gene Promoter in Male Chinese Subjects with Attention Deficit Hyperactivity Disorder

Neuropsychobiology, 2006

The aim of the current study was to examine any difference in the repeat sequence of the monoamin... more The aim of the current study was to examine any difference in the repeat sequence of the monoamine oxidase A (MAOA) gene promoter between males with attention deficit hyperactivity disorder (ADHD) and randomly selected subjects in a community. The role of the MAOA gene promoter in ADHD cases was also investigated. The total of 244 participants included 57 male ADHD subjects as the case group and 187 males, also in southern Taiwan, as the community controls. There was no significantly different distribution in the repeat sequence of the MAOA gene promoter (χ2 = 3.895, d.f. = 3, p = 0.273), and no significantly different distribution of ‘long-form’ and ‘short-form’ alleles between the ADHD group and the male community group was noted (χ2 = 2.484, d.f. = 1, p = 0.115). Some aspects of clinical response are mentioned in the discussion of this study which are worth exploring further in the future.

Research paper thumbnail of Protective effect of the apo e2 allele in major depressive disorder in Taiwanese

Acta Psychiatrica Scandinavica, 2006

Lung F-W. Protective effect of the apo e2 allele in major depressive disorder in Taiwanese.

Research paper thumbnail of The Societal Influences and Quality of Life Among Healthcare Team Members During the COVID-19 Pandemic

Frontiers in Psychiatry, 2021

Background: The coronavirus infection disease 2019 (COVID-19) pandemic is likely to put healthcar... more Background: The coronavirus infection disease 2019 (COVID-19) pandemic is likely to put healthcare professionals across the world in an unprecedented situation.Methods: A total of 683 healthcare workers were recruited in this study. Short form-12 items (SF-12), Societal Influences Survey Questionnaire (SISQ), and Disaster-Related Psychological Screening Test (DRPST) were used to survey participants. Multiple linear regression and structural equation model (SEM) were used to explore the possible factors to the societal influences and quality of life.Results: After multiple linear regression analysis, female, older, more education years, married, regular intake, and posttraumatic stress disorder (PTSD) frequency had positive association with SISQ. To physical component summary (PCS) of SF-12, chronic illness, sleep score, PTSD frequency, and social distance had negative association, and exercise habits had positive association. A mental component summary (MCS) value of SF-12, age, par...

Research paper thumbnail of Association of DRD4 uVNTR and TP53 codon 72 polymorphisms with schizophrenia: a case-control study

BMC Medical Genetics, 2009

Background: The tumour supressor gene TP53 is thought to be involved in neural apoptosis. The pol... more Background: The tumour supressor gene TP53 is thought to be involved in neural apoptosis. The polymorphism at codon 72 in TP53 and the long form variants of the upstream variable number of tandem repeats (uVNTR) polymorphism in the dopamine D4 receptor (DRD4) gene are reported to confer susceptibility to schizophrenia. Methods: We recruited 934 patients with schizophrenia and 433 healthy individuals, and genotyped the locus of the TP53 codon 72 and DRD4 uVNTR polymorphisms by combining the polymerase chain reaction-restriction fragment length polymorphism method (PCR-RFLP) with direct sequencing. Results: No significant differences were found in the frequency of the genotype of the TP53 codon72 polymorphism between patients with schizophrenia and their controls. However, the long form alleles (≥ 5 repeats) of the DRD4 uVNTR polymorphism were more frequent in patients with schizophrenia than in controls (p = 0.001). Hence, this class of alleles might be a risk factor for enhanced vulnerability to schizophrenia (odds ratio = 3.189, 95% confidence interval = 1.535-6.622). In the logistic regression analysis, the long form variants of the DRD4 polymorphism did predict schizophrenia after the contributions of the age and gender of the subjects were included (p = 0.036, OR = 2.319), but the CC and GG genotypes of the codon 72 polymorphism of TP53 did not. Conclusions: The long form variants of the uVNTR polymorphism in DRD4 were associated with schizophrenia, in a manner that was independent of the TP53 codon 72 polymorphism. In addition, given that the genetic effect of the TP53 codon 72 polymorphism on the risk of developing schizophrenia was very small, this polymorphism is unlikely to be associated with schizophrenia. The roles that other single nucleotide polymorphisms (SNPs) in the TP53 gene or in other apoptosis-related genes play in the synaptic dysfunction involved in the pathogenesis of schizophrenia should be investigated.