In this paper, incremental abbreviation detection in clinical texts is considered for the practical context where new clinical texts are added, processed, and exploited over time. We propose a parameter-free semi-supe...
详细信息
ISBN:
(纸本)9781728107882
In this paper, incremental abbreviation detection in clinical texts is considered for the practical context where new clinical texts are added, processed, and exploited over time. We propose a parameter-free semi-supervised method, named nonThreshold-ST, based on Self-training and C4.5. It inherits the prediction capability and simplicity of Self-training, while exploiting additional instances that are the most confidently predicted for enhancing the training dataset in the parameter-free configuration manner. The experimental results show that it outperforms its base supervised learning method on different English clinical text sets. Moreover, its Recall, Precision, F-measure, and Accuracy are among the highest values as compared to those of some other semi-supervised learning methods.
In the process of deepening the data processing of State Grid Power, a machinelearning algorithm is proposed to solve the problem that the data efficiency of State Grid Power cannot be effectively improved. Using mec...
详细信息
Scene classification is a popular and important question in computer vision and has been developed in different areas. Applying computer vision to artworks has become a popular topic in recent years. However, the trad...
详细信息
The proceedings contain 88 papers. The topics discussed include: information processing system for detection impurity in technical oil based on laser;hybrid machinelearning system for solving fraud detection tasks;sp...
ISBN:
(纸本)9781728132143
The proceedings contain 88 papers. The topics discussed include: information processing system for detection impurity in technical oil based on laser;hybrid machinelearning system for solving fraud detection tasks;speech signal structuring method for biometric personality identification;kln: a deep neural network architecture for keypoint localization;training neural network over encrypted data;forecasting nonlinear nonstationary processes in machinelearning task;the algorithmic classification trees;development of an adaptive module of the distance education system based on a hybrid neuro-fuzzy network;new approaches in the learning of complex-valued neural networks;trainable neural networks modeling for a forecasting of start-up product development;deep neo-fuzzy neural network and its accelerated learning;and two approaches to machinelearning classification of time series based on recurrence plots.
Nowadays, with the growing population of elderly people, the number of elderly without caregivers at home has also increased. It is clear that an elderly living alone at home is at higher risk of severe damage, due to...
详细信息
ISBN:
(纸本)9781509064540
Nowadays, with the growing population of elderly people, the number of elderly without caregivers at home has also increased. It is clear that an elderly living alone at home is at higher risk of severe damage, due to potential delays in notifying caregivers and providing care at healthcare facilities. This especially becomes critical in case of high-risk incidents such as stroke or heart attack. To address this issue, an increasing number of methods have been proposed that employ various fall detection algorithms for elderly people. In this paper, we propose a new algorithm to detect falls, using a multi-level fuzzy min-max neural network. The proposed algorithm is compared with three other machine-learning algorithms (MLP, KNN, SVM). The main focus of this paper is on the effect of dimensionality reduction with using the Principal Component Analysis (PCA) method inside the proposed algorithm. The evaluations show that the multi-level fuzzy min-max neural network provides a high level of accuracy with a small number of dimensions. This is in contrast to the other algorithms, where accuracy is further lowered after applying dimensionality reduction. The performance evaluation of this algorithm on a public dataset obtained using accelerometer sensor data with using three dimensions indicates an accuracy of 97.29% for the sensitivity metric and 98.70% for the specifity metric.
The low error rate of Naive Bayes (NB) classifier has been described as surprising. It is known that class conditional independence of the features is sufficient but not a necessary condition for optimality of NB. Thi...
详细信息
ISBN:
(纸本)9781424401956
The low error rate of Naive Bayes (NB) classifier has been described as surprising. It is known that class conditional independence of the features is sufficient but not a necessary condition for optimality of NB. This study is about the difference between the estimated error and the true error of NB taking into account feature dependencies. Analytical results are derived for two binary features. Illustration examples are also provided.
The project aims to perform patternrecognition of thumb and index linger gestures from the Electromyography (EMG) recordings acquired by a recently introduced External Wearable device. On the basis of the selected ti...
详细信息
ISBN:
(纸本)9781538670668
The project aims to perform patternrecognition of thumb and index linger gestures from the Electromyography (EMG) recordings acquired by a recently introduced External Wearable device. On the basis of the selected time domain features as reviewed based on classification performance, machinelearning techniques, such as K-nearest neighbour (KNN), Support Vector machine (SVM), Discriminant Analysis etc. are compared to choose a suitable model for recognition of same and different finger movements. The recognition model obtained for a set of six hand-finger gestures shows an accuracy of 80 - 86% in KNN model for two Different movements of Thumb and index linger and about 82 - 88% in SVM model for two same movements of index finger and thumb using single myo armband. The trained model obtained from single myo armband was also tested with data from double myo armbands. As a result, the accuracy obtained was in a range of 66-82% for various gestures. The post-analysis results are promising and competent evidence for available literature and for developing user-friendly medical devices. The purpose of analyzing the following gestures using Myo armband is to implement a suitable model for creating an intuitive human-machine interface like robotic Hand exoskeleton for rehabilitation purposes.
In order to promote the utilization of lifelog videos, an effective retrieval framework of the emotional scenes, which are considered to be important scenes, is proposed in this paper. The proposed method is based on ...
详细信息
ISBN:
(纸本)9781479941735
In order to promote the utilization of lifelog videos, an effective retrieval framework of the emotional scenes, which are considered to be important scenes, is proposed in this paper. The proposed method is based on facial expression recognition since the emotional scenes can be detected by taking the facial expressions into consideration. Most of conventional facial expression recognition methods require a large amount of training data to construct a recognition model. Adopting such methods for large-scale video databases is unrealistic because preparing sufficient training data requires considerable human efforts. We introduce an unsupervised machinelearning framework to solve this issue by making it possible to construct a facial expression recognition model without any training data set. The proposed method is evaluated through an emotional scene detection experiment. A prototype of the emotional scene retrieval system based on the proposed emotional scene detection method is implemented.
Wireless Sensor Networks (WSNs) have revolutionized data collection, especially in Human Activity recognition (HAR). Multisensor datasets are crucial for a comprehensive understanding of human behavior, enabling more ...
详细信息
ISBN:
(纸本)9798350372977;9798350372984
Wireless Sensor Networks (WSNs) have revolutionized data collection, especially in Human Activity recognition (HAR). Multisensor datasets are crucial for a comprehensive understanding of human behavior, enabling more advanced classification techniques. This study explores the essential role of machinelearning in categorizing activities, especially given the abundance of available multi-sensor data from WSN. The research utilizes information fusion as a pivotal mechanism to boost the accuracy of activity classifications. Employing Support Vector machine (SVM) and Decision Tree (DT) algorithms, the project utilizes advanced data fusion techniques, specifically Kalman Filter (KF) and Covariance Intersection (CI), to optimize information extraction from the provided data. The study encompasses six experiments, including applying SVM and DT on raw data, SVM and DT on data fused by CI, and SVM and DT on data fused by KF. The results of these experiments reveal a significant improvement in the accuracy of SVM and DT classification when incorporating CI and KF. This emphasizes the effectiveness of information fusion techniques in refining the outcomes of human activity recognition systems, showcasing their vital role in enhancing the reliability and precision of activity classifications. This research not only contributes to the field of HAR but also establishes a foundation for further advancements in real-world applications where precise activity classification holds utmost importance.
Recent prolific advances in artificial intelligence through the incorporation of domain knowledge have constituted a new paradigm for AI and datamining communities. For example, the human feedback-based language gene...
详细信息
ISBN:
(纸本)9798400701030
Recent prolific advances in artificial intelligence through the incorporation of domain knowledge have constituted a new paradigm for AI and datamining communities. For example, the human feedback-based language generation in ChatGPT (a large language model (LLM)), the use of Protein Bank in DeepMind's AlphaFold, and the use of 23 rules of safety in DeepMind's Sparrow have demonstrated the success of teaming human knowledge and AI. In addition, the knowledge retrieval-guided language modeling methods have strengthened the association between knowledge and AI. However, translating research methods and resources into practice presents a new challenge for the machinelearning and data/knowledge mining communities. For example, in DARPA's Explainable AI seminar, the need for explainable contextual adaptation is seen as the 3rd phase of AI, facilitating the interplay between data and knowledge for explainability, safety, and, eventually, trust. However, policymakers and practitioners assert serious usability and privacy concerns that constrain adoption, notably in high-consequence domains, such as cybersecurity, healthcare, and other social good domains. In addition, limitations in output quality, measurement, and interactive ability, including both the provision of explanations and the acceptance of user preferences, result in low adoption rates in such domains. This workshop aims to accelerate our pace towards creating innovative methods for integrating knowledge into contemporary AI and data science methods and develop metrics for assessing performance in various applications.
暂无评论