Introducing SPINE: A Holistic Approach to Synthetic Pulmonary Imaging Evaluation Through End-to-End Data and Model Management.

In the evolving field of medical imaging and machine learning (ML), this paper introduces a novel framework for evaluating synthetic pulmonary imaging aiming to assess synthetic data quality and applicability. Our study concentrates on synthetic X-ray chest images, crucial for diagnosing respiratory diseases. We employ SPINE (Synthetic Pulmonary Imaging Evaluation) framework, a threefold synthetic images evaluation method including expert domain assessment, statistical data analysis and adversarial evaluation. In order to replicate and validate our methodology, we followed an End-to-End data and model management process which begins with a dataset of Normal and Pneumonia chest X-rays, generating synthetic images using Generative Adversarial Networks (GANs) and training a baseline classifier, essential in the adversarial evaluation axis, testing synthetic images against real data assessing their predictive value. The critical outcome of our approach is the post-market analysis of synthetic images. This innovative method evaluates synthetic images using clinical, statistical, and scientific criteria independently from traditional generation performance metrics. This independent evaluation provides deep insights into the clinical and research effectiveness of the synthetic data. By ensuring these images mirror real data's statistical properties and maintain clinical accuracy, our framework establishes a new standard for the ethical and reliable use of synthetic data in medical imaging and research.

Multi3Generation: Multitask, Multilingual, and Multimodal Language Generation.

The article emphasizes the critical importance of language generation today, particularly focusing on three key aspects: Multitasking, Multilinguality, and Multimodality, which are pivotal for the Natural Language Generation community. It delves into the activities conducted within the Multi3Generation COST Action (CA18231) and discusses current trends and future perspectives in language generation.

The Multi3Generation COST Action is a collaborative project that brings together researchers from various fields, all centered around Natural Language Generation. Natural Language Generation involves using computers to generate human-like language for tasks such as translation, summarization, question-answering, and dialogue interaction, among others. The Action addresses common challenges including efficient information representation, advanced machine learning techniques, managing uncertainty in human-Natural Language Generation interactions, and using structured knowledge from diverse sources like databases, images, and videos. Its overarching goal is to make NLG beneficial to society and widely accessible by fostering collaboration between industry and academic experts. Structured into five working groups, the Action focuses on specific aspects of Natural Language Generation, such as understanding and generating different types of information, developing efficient machine learning algorithms, enhancing dialogue and conversational language generation using knowledge bases, and fostering industry collaboration and end-user engagement. With over 133 scientists from 34 countries involved, spanning disciplines from computer science to linguistics, the project promotes diversity and inclusivity, with 60% male and 40% female participants. Relevant businesses like Unbabel and JabberBrain and other AI stakeholders like the Center for Responsible AI contribute to the Action, aiming to have a broader European impact. The Multi3Generation Action prioritizes three main areas: Multitasking, Multilinguality, and Multimodality, aiming to enhance language generation in these domains to support underrepresented languages and meet diverse user needs. The article provides insights into the initiatives and planned activities of Multi3Generation, offering valuable information for those interested in NLG and shedding light on future perspectives in this field.

Diagnosis of Alzheimer's disease and Mild Cognitive Impairment using EEG and Recurrent Neural Networks.

Alzheimer's disease (AD) is the main cause of dementia and Mild cognitive impairment (MCI) is a prodromal stage of AD whose early detection is considered crucial as it can contribute in slowing the progression of AD. In our study we attempted to classify a subject into AD, MCI, or Healthy Control (HC) groups with the use of electroencephalogram (EEG) data. Due to the time-series nature of EEG we exper-imented with the powerful recurrent neural network (RNN) classifiers and more specifically with models including basic or bidirectional Long Short-Term Memory (LSTM) modules. The EEG signals from 17 channels were preprocessed using a 0.1-32 Hz band-pass filter and then segmented into 2-second epochs during which, the subject had closed eyes. Finally, on each segment Fast Fourier Transform (FFT) was applied. To evaluate our models we studied four different classification problems: problem 1: separating subject into three classes (HC, MCI, AD) and problems 2-4: pairwise classifications AD vs. MCI, AD vs. HC and MCI vs. HC. For each problem we employed two different cross-validation approaches ( a) by segment and (b) by patient. In the first one, segments from a subject EEG may exist in both training and validations set, while in the second one, all the EEG segments of a subject can only exist in either the training or the validation set. In the AD-MCI-HC classification we achieved an accuracy of 99% by segment cross-validation, which was an improvement to earlier studies that utilized recurrent neural network models. In the pairwise classification problems we achieved over 90% accuracy by segment and over 80% by subject.

CLAss-Specific Subspace Kernel Representations and Adaptive Margin Slack Minimization for Large Scale Classification.

In kernel-based classification models, given limited computational power and storage capacity, operations over the full kernel matrix becomes prohibitive. In this paper, we propose a new supervised learning framework using kernel models for sequential data processing. The framework is based on two components that both aim at enhancing the classification capability with a subset selection scheme. The first part is a subspace projection technique in the reproducing kernel Hilbert space using a CLAss-specific Subspace Kernel representation for kernel approximation. In the second part, we propose a novel structural risk minimization algorithm called the adaptive margin slack minimization to iteratively improve the classification accuracy by an adaptive data selection. We motivate each part separately, and then integrate them into learning frameworks for large scale data. We propose two such frameworks: the memory efficient sequential processing for sequential data processing and the parallelized sequential processing for distributed computing with sequential data acquisition. We test our methods on several benchmark data sets and compared with the state-of-the-art techniques to verify the validity of the proposed techniques.