354 research outputs found
Reviews: Derek Bickerton, Bastard Tongues. A Trailblazing Linguist Finds Clues to Our Common Humanity in the World’s Lowliest Languages
No full textBASTARD TONGUES: A Trailblazing Linguist Finds Clues to Our Common Humanity in the World’s Lowliest Languages. Author: Derek Bickerton (270 pp. Hill & Wang. New York - 2008. $ 26.) Review by Leonardo Caffo
Editorial: Explainable artificial intelligence models and methods in finance and healthcare
No full textThis article is a foreword to a special issue on "Explainable artificial intelligence models and methods in finance and healthcare" and introduces the main articles of the collection. The core topic of this special issue is explainability and trusting algorithmic output
Explainable Artificial Intelligence models and methods in Finance and Healthcare
No full textIn the last years, the fast and growing data availability has allowed providing highly predictive responses for the advanced research fields. Sophisticated machine learning models and techniques were then developed together with artificial intelligence-based systems. Increasing attention is given to Artificial Intelligence (AI), especially due to its algorithms, which give rise to robust predictions. Nevertheless, AI systems have a black box nature resulting in automated decision-making. This can classify a user into a category associated with the prediction of the individual behavior without specifying the underlying rationale. Some concerns about the adequacy of the AI models and methods in regulatory scenarios arise, primarily due to the possible biases generated by the Machine Learning algorithms. This leads organizations to claim high credibility and interpretability to provide effective operational control. The lack of transparency and explainability is, therefore, a critical point for policymakers and regulators aimed at avoiding wrong actions with adverse consequences on society. This issue is more evident in the financial and banking sectors, where the use cases of AI extend to the contexts of risk management, predictive analytics, and fraud detection, as well as in the healthcare field, where the focus is on both the funding management process of the healthcare services and the improvement of the diagnostic precision. We can resort to AI-based systems to predict the financial, default, funding loss, and diagnostic-related risks. However, AI-based systems require that the main criteria, which support the predictions, are known in order to assess the related severity and foster the appropriate measures to reduce the risks in case of shocks in the financial systems, changes in market conditions, or monitoring of the healthcare policies. For the purpose of explaining and interpreting machine learning models, eXplainable Artificial Intelligence (XAI) represents a fundamental field for understanding the steps and methods driving the decision process. In line with the policy requirements of transparency, this Research Topic aims to include original papers proposing the development of innovative XAI methodologies for global or local explanations in the research area of: • the financial and banking sectors - mainly focused on credit scoring, which involves lending algorithms, price discovery (representing the basis of financial robot advisory algorithms), and cyber risk management (greatly critical due to the increasingly online connections); • the healthcare field mainly focused on the evaluation of the funding and management policies
Combining hidden Markov models for comparing the dynamics of multiple sleep electroencephalograms
No full textLangrock R, Swihart BJ, Caffo BS, Punjabi NM, Crainiceanu CM. Combining hidden Markov models for comparing the dynamics of multiple sleep electroencephalograms. Statistics in Medicine. 2013;32(19):3342-3356
Exact Hypothesis Tests for Log-linear Models with exactLoglinTest
No full textThis manuscript overviews exact testing of goodness of fit for log-linear models using the R package exactLoglinTest. This package evaluates model fit for Poisson log-linear models by conditioning on minimal sufficient statistics to remove nuisance parameters. A Monte Carlo algorithm is proposed to estimate P values from the resulting conditional distribution. In particular, this package implements a sequentially rounded normal approximation and importance sampling to approximate probabilities from the conditional distribution. Usually, this results in a high percentage of valid samples. However, in instances where this is not the case, a Metropolis Hastings algorithm can be implemented that makes more localized jumps within the reference set. The manuscript details how some conditional tests for binomial logit models can also be viewed as conditional Poisson log-linear models and hence can be performed via exactLoglinTest. A diverse battery of examples is considered to highlight use, features and extensions of the software. Notably, potential extensions to evaluating disclosure risk are also considered
Leveraging Meta-analytic Topic Maps to Identify and Remove Residual Motion Artifacts from fMRI Data
No full textHead motion during functional magnetic resonance imaging (fMRI) can seriously affect the integrity of the blood-oxygen-level-dependent (BOLD) signal and lead to biased or distorted inferences about neural responses to tasks or resting state. Most studies treat motion purely as a technical artifact to be removed, but overlook the possibility that sudden or voluntary head movements may evoke neural responses that can produce systematic variance in the BOLD signal, potentially leading to false inferences if not properly accounted for. Here, we leverage large-scale meta-analytic brain maps to (1) model the temporal and spatial characteristics of the head-motion-evoked BOLD response during resting state in a sample of healthy young adults, and (2) evaluate the effects of regressing out this modeled evoked response, compared to using standard motion regression, on resting-state data and subject-level contrasts for a working memory (WM) task. Analysis revealed that head movements produced a systematic increase in BOLD activation around 3.6 seconds post-movement, and this activation was particularly prevalent within voxels associated with motor-related meta-analytic topics. Because of this, we introduce a new method of motion regression, which involves convolving subject-level thresholded framewise displacements (FDs) with the topic-modeled BOLD response to motion, and then including this as a regressor in the subject-level general linear model (GLM). We found that this method of motion regression, which we call FD-regression, was more effective at reducing BOLD activations due to head motion in resting-state data than standard motion regression. However, FD-regression had little to no effects on subject-level WM contrasts. That is, average correlations between subject-level WM contrasts and the WM/cognitive load meta-analysis topic map did not change significantly after applying FD-regression. We conclude that while FD-regression may be an effective method for reducing BOLD signal associated with head motion in resting-state data, further analysis is needed to evaluate its effects on data quality for task-based fMRI
Leveraging Meta-analytic Topic Maps to Identify and Remove Residual Motion Artifacts from fMRI Data
No full textHead motion during functional magnetic resonance imaging (fMRI) can seriously affect the integrity of the blood-oxygen-level-dependent (BOLD) signal and lead to biased or distorted inferences about neural responses to tasks or resting state. Most studies treat motion purely as a technical artifact to be removed, but overlook the possibility that sudden or voluntary head movements may evoke neural responses that can produce systematic variance in the BOLD signal, potentially leading to false inferences if not properly accounted for. Here, we leverage large-scale meta-analytic brain maps to (1) model the temporal and spatial characteristics of the head-motion-evoked BOLD response during resting state in a sample of healthy young adults, and (2) evaluate the effects of regressing out this modeled evoked response, compared to using standard motion regression, on resting-state data and subject-level contrasts for a working memory (WM) task. Analysis revealed that head movements produced a systematic increase in BOLD activation around 3.6 seconds post-movement, and this activation was particularly prevalent within voxels associated with motor-related meta-analytic topics. Because of this, we introduce a new method of motion regression, which involves convolving subject-level thresholded framewise displacements (FDs) with the topic-modeled BOLD response to motion, and then including this as a regressor in the subject-level general linear model (GLM). We found that this method of motion regression, which we call FD-regression, was more effective at reducing BOLD activations due to head motion in resting-state data than standard motion regression. However, FD-regression had little to no effects on subject-level WM contrasts. That is, average correlations between subject-level WM contrasts and the WM/cognitive load meta-analysis topic map did not change significantly after applying FD-regression. We conclude that while FD-regression may be an effective method for reducing BOLD signal associated with head motion in resting-state data, further analysis is needed to evaluate its effects on data quality for task-based fMRI
Resolution and Reliability in Functional Connectivity Analysis
No full textIncreasingly, researchers are interested in the functional connectivity between different brain regions of resting-state imaging. However, quantifying the reliability of measures of both brain function and structure is difficult, while reliability is essential to accurately detect differences between subjects or groups and predict outcomes. Here, we seek to evaluate the reliability and prediction performance of functional connectivity networks obtained using independent components analysis (ICA) with varying number of nodes. Reliability is measured by intra-class correlation coefficient (ICC), image intra-class correlation coefficient (I2C2) and Kolmogorov–Smirnov (KS) test. In particular, we evaluated how the number of components influence the results. We found the local (ICC) and global (I2C2) reliability have different trends with the change in the resolution of the ICA-based pacellation, and higher reliability was achieved on sessions that took place on the same day compared to different days for both ICC and I2C2. Individual patterns of functional connectivity do exist. First, KS test shown greater similarity between different sessions than different subjects. Second, fingerprinting and prediction of sex using connectivity matrix yield high prediction accuracy (over 90%). In addition, a moderate resolution of the parcellation seems to be an optimal choice. It provides more information about functional connectivity, improves the performance of prediction and avoids unnecessary computational cost, which only contribute little improvement to prediction accuracy
FACE TOUCH DETECTION TO REDUCE DISEASE TRANSMISSION
No full textIn the last several years, multiple research teams have investigated the face-touch detection problem for the purpose of developing technology that can detect and subsequently prevent face-touching. IMU-based datasets have been cultivated under controlled settings of trial participants performing different face-touches. This thesis seeks to use one well-outlined dataset in particular, to build a “pre-face touch detection model” and a “completed face-touch detection model”. These two Human Activity Recognition (HAR) models are designed to be situated in a free-living study of participants “in the wild”. The pre-face touch detection model is the focus of this thesis, as we primarily seek to build a face-touch prevention tool that potentially can reduce disease transmission for its user. The need for researchers to collect free-living face-touch trials such that a pre-face touch detection model can be built that is generalizable to users “in the wild” is what motivated the development of the completed face-touch detection model. The main contribution of this thesis to the sub-problem of face-touch detection within the HAR space is our investigation of utilizing Discrete Wavelet Transform (DWT) on IMU sensor signals in order to extract localized time-frequency features relevant to the tasks of pre-face touch detection and completed face-touch detection. Additionally, we identify constraints on Precision and Specificity metrics in order to minimize the invasiveness of the pre-face touch detection model, and we maximize Recall score with these constraints considered
FACE TOUCH DETECTION TO REDUCE DISEASE TRANSMISSION
No full textIn the last several years, multiple research teams have investigated the face-touch detection problem for the purpose of developing technology that can detect and subsequently prevent face-touching. IMU-based datasets have been cultivated under controlled settings of trial participants performing different face-touches. This thesis seeks to use one well-outlined dataset in particular, to build a “pre-face touch detection model” and a “completed face-touch detection model”. These two Human Activity Recognition (HAR) models are designed to be situated in a free-living study of participants “in the wild”. The pre-face touch detection model is the focus of this thesis, as we primarily seek to build a face-touch prevention tool that potentially can reduce disease transmission for its user. The need for researchers to collect free-living face-touch trials such that a pre-face touch detection model can be built that is generalizable to users “in the wild” is what motivated the development of the completed face-touch detection model. The main contribution of this thesis to the sub-problem of face-touch detection within the HAR space is our investigation of utilizing Discrete Wavelet Transform (DWT) on IMU sensor signals in order to extract localized time-frequency features relevant to the tasks of pre-face touch detection and completed face-touch detection. Additionally, we identify constraints on Precision and Specificity metrics in order to minimize the invasiveness of the pre-face touch detection model, and we maximize Recall score with these constraints considered
- …
