Identifying the most appropriate therapeutic course for oncology patients is a complex task, even for clinicians. To address this difficulty, many healthcare systems worldwide have established mandatory Multidisciplinary Team Meetings, in which specialists from various medical fields jointly review each case to reach a consensus treatment recommendation. Despite their importance, these meetings face significant challenges: the growing number of cases to review, limited time for examining patient records, occasional unavailability of experts, and varying levels of familiarity with the latest national guidelines. With the impressive progress in Natural Language Processing (NLP), developing tools to retrieve the best therapeutic options from official treatment guidelines for each patient can help enlighten medical experts to support their clinical decisions during these meetings. We propose a lightweight information retrieval approach based on a fine-tuned BERT-based cross-encoder that fully benefits from both domain expert knowledge and the structure of the reference documents considered. Through experiments on a real-world colorectal cancer treatment dataset created in collaboration with our local hospital, we show that our lightweight model reaches a F2-score of 95% on retrieving the proper fragments of official guidelines and outperforms by far Large Language Models (LLMs) with some of the standard prompt strategies.

Anomaly detection, or more generally outlier detection, is one of the most popular and challenging topics in theoretical and applied machine learning. The main challenge is that in general we have access to very few labeled data or no labels at all. In this paper, we present a new semi-supervised anomaly detection method called AnoRand by combining a deep learning architecture with random synthetic label generation. The proposed architecture has two building blocks: (1) a noise detection (ND) block composed of feed forward perceptron and (2) an autoencoder (AE) block. The main idea of this new architecture is to learn one class (e.g. the majority class in case of anomaly detection) as well as possible by taking advantage of the ability of auto encoders to represent data in a latent space and the ability of Feed Forward Perceptron (FFP) to learn one class when the data is highly imbalanced. First, we create synthetic anomalies by randomly disturbing (add noise) a few samples (e.g., 2%) from the training set. Second, we use the normal and synthetic samples as input to our model. We compared the performance of the proposed method to 17 state-of-the-art unsupervised anomaly detection method on synthetic datasets and 57 real-world datasets. Our results show that this new method generally outperforms most of the state-of-the-art methods and has the best performance (AUC ROC and AUC PR) on the vast majority of reference datasets.

Large Language Models have undoubtedly revolutionized the Natural Language Processing field, the current trend being to promote one-model-for-all tasks (sentiment analysis, translation, etc.). However, the statistical mechanisms at work in the larger language models struggle to exploit the relevant information when it is very sparse, when it is a weak signal. This is the case, for example, for the classification of long domain-specific documents, when the relevance relies on a single relevant word or on very few relevant words from technical jargon. In the medical domain, it is essential to determine whether a given report contains critical information about a patient's condition. This critical information is often based on one or few specific isolated terms. In this paper, we propose a hierarchical model which exploits a short list of potential target terms to retrieve candidate sentences and represent them into the contextualized embedding of the target term(s) they contain. A pooling of the term(s) embedding(s) entails the document representation to be classified. We evaluate our model on one public medical document benchmark in English and on one private French medical dataset. We show that our narrower hierarchical model is better than larger language models for retrieving relevant long documents in a domain-specific context.

Anomaly detection or more generally outliers detection is one of the most popular and challenging subject in theoretical and applied machine learning. The main challenge is that in general we have access to very few labeled data or no labels at all. In this paper, we present a new semi-supervised anomaly detection method called \textbf{AnoRand} by combining a deep learning architecture with random synthetic label generation. The proposed architecture has two building blocks: (1) a noise detection (ND) block composed of feed forward ferceptron and (2) an autoencoder (AE) block. The main idea of this new architecture is to learn one class (e.g. the majority class in case of anomaly detection) as well as possible by taking advantage of the ability of auto encoders to represent data in a latent space and the ability of Feed Forward Perceptron (FFP) to learn one class when the data is highly imbalanced. First, we create synthetic anomalies by randomly disturbing (add noise) few samples (e.g. 2\%) from the training set. Second, we use the normal and the synthetic samples as input to our model. We compared the performance of the proposed method to 17 state-of-the-art unsupervised anomaly detection method on synthetic datasets and 57 real-world datasets. Our results show that this new method generally outperforms most of the state-of-the-art methods and has the best performance (AUC ROC and AUC PR) on the vast majority of reference datasets. We also tested our method in a supervised way by using the actual labels to train the model. The results show that it has very good performance compared to most of state-of-the-art supervised algorithms.

Machine learning models are increasingly being used in predictive maintenance. However, due to the complexity of vibration and audio signals used in fault diagnosis, some pre-processing is required before feeding them into the machine learning algorithm. Fast Fourier Transform (FFT) and the Hilbert transform (HT) envelope spectrum are mostly used in the literature for pre-processing. However, these frequency domain transforms are not very effective when applied to rotating systems (e.g. bearings) fault detection. In this paper we propose to use signature coefficients to feed machine learning models for fault detection. Our experimental results show that this method outperforms most state-of-the-art methods on fault diagnosis data sets. Moreover, the results show that this method is particularly well suited for high dimensional time series. The results also show that compared to Fast Fourier Transform (FFT), the signature method requires fewer data points to detect failure. This means that in a situation where the two methods have similar performances, the signature method detects failure faster than FFT.

In this paper, we address the challenge of efficient local training on heterogeneous data in Federated Learning, where devices have limited computation and communication resources. The proposed algorithm generates personalized masks for groups of nodes by applying structured or unstructured pruning methods. We find a trade-off between global generalization and per-device personalization by optimizing the personalized masks to local data distributions through clustering methods and by aggregating the parameters in a unique model. Our results demonstrate effectiveness of the method for training personalized and sparsified models on non-IID distributions. The maximal accuracy achieved by our Clustered Adaptive Pruning method outperforms the Federated Average algorithm, while using a small percentage of the original set of coefficients per dataset. In particular, for MNIST, CIFAR10, and FMNIST datasets, our method uses approximately two-thirds, one-third, and less than one-third of the total number of parameters required for Federated Average training, respectively.

Sentiment analysis or opinion mining is used to understand the community’s opinions on a particular product. This is a system of selection and classification of opinions on sentences or documents. At a more detailed level, aspect-based sentiment analysis makes an effort to extract and categorize sentiments on aspects of entities in opinion text. In this paper, we propose a novel supervised learning approach using deep learning techniques for a multitasking aspect-based opinion mining system that supports four main subtasks: extract opinion target, classify aspect, classify entity (category) and estimate opinion polarity (positive, neutral, negative) on each extracted aspect of the entity. We have used a part-of-speech (POS) layer to define the words’ morphological features integrated with GloVe word embedding in the previous layer and fed to the convolutional neural network_bidirectional long-short term memory (CNN_BiLSTM) stacked construction to improve the model’s accuracy in the opinion classification process and related tasks. Our multitasking aspect-based sentiment analysis experiments on the dataset of SemEval 2016 showed that our proposed models have obtained and categorized core tasks mentioned above simultaneously and attained considerably better accurateness than the advanced researches.

This study aims to use artificial neural network based classifiers to predict fraud, particularly that related to health insurance. Medicare fraud results in considerable losses for governments and insurance companies and results in higher premiums from clients. Medicare fraud costs around 13 billion euros in Europe and between 21 billion and 71 billion US dollars per year in the United States. To detect medicare frauds, we propose a multiple inputs deep neural network based classifier with an autoencoder component. This architecture makes it possible to take into account many sources of data without mixing them and makes the classification task easier for the final model. We use the data sets from the Centers for Medicaid and Medicare Services (CMS) of the US federal government and four benchmarks fraud detection data sets. Our results show that although baseline artificial neural network give good performances, they are outperformed by our multiple inputs neural networks. We have shown that using an autoencoder to embed the provider behavior gives better results and makes the classifiers more robust to class imbalance.

Advanced computing resources are requested by different interest groups, with the acceleration of digital transformation generated both by the pandemic and by the explosive increase in available information. Data centers have increased significantly over the past years to answer this demand but are now facing energy efficiency issues and performance requirements. Renewable energy is a promising solution by supplying local energy, but this power generation is reduced, intermittent, and requires changing how computation and data storage is managed. Edge/Fog computing is a promising solution to latency concerns proximity to data centers; however, this solution often extends existing clouds, imposing no specific principles to reduce energy and greenhouse emissions. This paper discusses challenges of a scalable architecture to support sustainable advanced information technology requirements inspired by the circular economy characteristics. The architecture has to keep rethinking computation, optimizing deployment, and regenerating energy.

Sentiment analysis aids in obtaining the opinion of the users towards a particular product, service or policy. Focusing on classifying the sentiment that corresponds to each aspect of the entity in the document will help to identify the sentiment more clearly. This is also the mission of aspect-based sentiment analysis (ABSA). The vast majority of prior studies in ABSA have implemented single-task execution models on single-domain datasets. This is inconvenient when it is necessary to perform the full range of tasks in ABSA and on domain-independent datasets. In this paper, we offer to operate the advanced arrangement of deep learning techniques for multidomain and multitask approach in ABSA. The main tasks in ABSA: aspect extraction, category identification, sentiment classification and domain classification are all finished by an integration framework of Convolutional Neural Network (CNN), Bidirectional Independent Long Short Term Memory (BiIndyLSTM) and Attention mechanism. In addition, we use a POS tag layer combined with GloVe in word embedding layer to get the morphological attributes of each token word from review sentences. Through the experimenting process in the Laptop_Restaurant_Hotel multidomain dataset, we found that our proposed model has achieved high precision in multitasking ABSA. With this approach, we hope our proposed model will lay the foundation for ensuring flexibility and multiutility compared to previous opinion analysis models.

Medicare fraud results in considerable losses for governments and insurance companies and results in higher premiums from clients. Medicare fraud costs around 13 billion euros in Europe and between 21 billion and 71 billion US dollars per year in the United States. This study aims to use artificial neural network based classifiers to predict medicare fraud. The main difficulty using machine learning techniques in fraud detection or more generally anomaly detection is that the data sets are highly imbalanced. To detect medicare frauds, we propose a multiple inputs deep neural network based classifier with a Long-short Term Memory (LSTM) autoencoder component. This architecture makes it possible to take into account many sources of data without mixing them and makes the classification task easier for the final model. The latent features extracted from the LSTM autoencoder have a strong discriminating power and separate the providers into homogeneous clusters. We use the data sets from the Centers for Medicaid and Medicare Services (CMS) of the US federal government. The CMS provides publicly available data that brings together all of the cost price requests sent by American hospitals to medicare companies. Our results show that although baseline artificial neural network give good performances, they are outperformed by our multiple inputs neural networks. We have shown that using a LSTM autoencoder to embed the provider behavior gives better results and makes the classifiers more robust to class imbalance.

Sentiment analysis or opinion mining used to capture the community’s attitude who have experienced the specific service/product. Sentiment analysis usually concentrates to classify the opinion of whole document or sentence. However, in most comments, users often express their opinions on different aspects of the mentioned entity rather than express general sentiments on entire document. In this case, using aspect-based sentiment analysis (ABSA) is a solution. ABSA emphases on extracting and synthesizing sentiments on particular aspects of entities in opinion text. The previous studies have difficulty working with aspect extraction and sentiment polarity classification in multiple domains of review. We offer an innovative deep learning approach with the integrated construction of bidirectional Long Short Term Memory (BiLSTM) and Convolutional Neural Network (CNN) for multidomain ABSA in this article. Our system finished the following tasks: domain classification, aspect extraction and opinion determination of aspect in the document. Besides applying GloVe word embedding for input sentences from mixed Laptop_Restaurant domain of the SemEval 2016 dataset, we also use the additional layer of POS to pick out the word morphological attributes before feeding to the CNN_BiLSTM architecture to enhance the flexibility and precision of our suggested model. Through experiment, we found that our proposed model has performed the above mentioned tasks of domain classification, aspect and sentiment extraction concurrently on a mixed domain dataset and achieved the positive results compared to previous models that were performed only on separated domain dataset.

Data leakage can lead to severe issues for a company, including financial loss, damage of goodwill, reputation, lawsuits and loss of future sales. To prevent these problems, a company can use other mechanisms on top of traditional Access Control. These mechanisms include for instance Data Leak Prevention or Information Rights Management and can be referred as Transmission Control. However, such solutions can lack usability and can be intrusive for end-users employees. To have a better understanding of the perception and usage of such mechanisms within business infrastructures, we have conducted in this article an online survey on 150 employees. These employees come from different companies of different sizes and sectors of activity. The results show that whatever the size of the company or its sector of activity, security mechanisms such as access control and transmission control can be considered as quite intrusive and blocking for employees. Moreover, our survey also shows interesting results regarding more acceptable and user-friendly anti-data leakage mechanisms that could be used within companies.

Determining an optimal generalization model with deep neural networks for a medical task is an expensive process that generally requires large amounts of data and computing power. Furthermore, the complexity of the programming expressiveness increases to scale deep learning workflows over new heterogeneous system architectures for training each model and efficiently configure the computing resources. We introduce DiagnoseNET, an automatic framework designed for scaling deep learning models over heterogeneous systems applied to medical diagnosis. DiagnoseNET is designed as a modular framework to enable the deep learning workflow management and allows the expressiveness of neural networks written in TensorFlow, while the DiagnoseNET runtime abstracts the data locality, micro batching and the distributed orchestration to scale the neural network model from a GPU workstation to multi-nodes. The main approach is composed through a set of gradient computation modes to adapt the neural network according to the memory capacity, the workers’ number, the coordination method and the communication protocol (GRPC or MPI) for achieving a balance between accuracy and energy consumption. The experiments carried out allow to evaluate the computational performance in terms of accuracy, convergence time and worker scalability to determine an optimal neural architecture over a mini-cluster of Jetson TX2 nodes. These experiments were performed using two medical cases of study, the former dataset is composed by clinical descriptors collected during the first week of hospitalization of patients in the Provence-Alpes-Côe d’Azur region; the second dataset uses a short ECG records between 30 and 60 s, obtained as part of the PhysioNet 2017 Challenge.

Sentiment analysis or opinion mining used to understand the community\u0027s opinions on a particular product. Sentiment analysis involves building the opinion collection and classification system. Aspect-based sentiment analysis focuses on the ability to extract and summarize opinions on specific aspects of entities within sentiment document. In this paper, we propose a novel supervised learning approach using deep learning techniques for multitask aspect-based opinion mining system that support four main subtasks: extract opinion target, classify aspect-entity (category), and estimate opinion polarity (positive, neutral, negative) on each extracted aspect of entity. Using extra POS layer to identify morphological features of words combines with stacking architecture of BiLSTM and CNN with word embeddings achieved by training GloVe on Restaurant domain reviews of the SemEval 2016 benchmark dataset in our proposed method is aimed at increasing the accuracy of the model. Experimental results showed that our multitask aspect-based sentiment analysis model has extracted and classified main above subtasks concurrently and achieved significantly better accuracy than the state-of-the-art methods.

The research of Human Action Recognition (HAR) has made a lot of progress in recent years, and the research based on RGB images is the most extensive. However, there are two main shortcomings: the recognition accuracy is insufficient, and the time consumption of the algorithm is too large. In order to improve these issues our project attempts to optimize the algorithm based on the random forest algorithm by extracting the features of the human body 3D, trying to obtain more accurate human behavior recognition results, and can calculate the prediction results at a lower time cost. In this study, we used the 3D spatial coordinate data of multiple Kinect sensors to overcome these problems and make full use of each data feature. Then, we use the data obtained from multiple Kinects to get more accurate recognition results through post processing.

Sentiment analysis involves building the opinion collection and classification system. Aspect-based sentiment analysis focuses on the ability to extract and summarize opinions on specific aspects of entities within sentiment document. In this paper, we propose a novel supervised learning approach using deep learning techniques for multitask aspect-based opinion mining system that support four main subtasks: extract opinion target, classify aspect-entity (category), and estimate opinion polarity (positive, neutral, negative) on each extracted aspect of entity. Using extra POS layer to identify morphological features of words combines with stacking architecture of BiLSTM and CNN with word embeddings achieved by training GloVe on Restaurant domain reviews of the SemEval 2016 benchmark dataset in our proposed method is aimed at increasing the accuracy of the model. Experimental results showed that our multitask aspect-based sentiment analysis model has extracted and classified main above subtasks concurrently and achieved significantly better accuracy than the state-of-the-art methods.

The value of data-driven healthcare is the possibility to detect new patterns for inpatient care, treatment, prevention, and comprehension of disease or to predict the duration of hospitalization, its cost or whether death is likely to occur during the hospital stay. Modeling precise patients phenotype representation from clinical data is challenging over its high-dimensionality, noisy and missing data to be processed into a new low-dimensionality space. Likewise, processing unsupervised learning models into a growing clinical data raises many issues, in terms of algorithmic complexity, such as time to model convergence and memory capacity. This paper presents DiagnoseNET framework to automate patient phenotype extractions and apply them to predict different medical targets. It provides three high-level features: a full-workflow orchestration into stage pipelining for mining clinical data and using unsupervised feature representations to initialize supervised models; a data resource management for training parallel and distributed deep neural networks. As a case of study, we have used a clinical dataset from admission and hospital services to build a general purpose inpatient phenotype representation to be used in different medical targets, the first target is to classify the main purpose of inpatient care. The research focuses on managing the data according to its dimensions, the model complexity, the workers number selected and the memory capacity, for training unsupervised staked denoising auto-encoders over a Mini-Cluster Jetson TX2. Therefore, mapping tasks that fit over computational resources is a key factor to minimize the number of epochs necessary to model converge, reducing the execution time and maximizing the energy efficiency.