Hasil untuk "machine learning"

Simona Vojtekova, Katarina Kramarova, Veronika Labosova et al.

Despite the rapid expansion of AI and machine-learning techniques in bankruptcy prediction, classical statistical methods such as discriminant analysis and logistic regression remain relevant because of their transparency and interpretability. These characteristics are crucial for stakeholders who require understandable decision-making tools, especially in NACE Rev. 2 Section I—Accommodation and Food Service Activities, a sector characterized by high operating leverage, vulnerability to economic shocks, and strong macroeconomic importance. The study aims to evaluate and compare the predictive performance of discriminant analysis and logistic regression for bankruptcy prediction and to identify key predictors that can serve as managerial early-warning signals for companies in crisis across B9 countries. The sample of 4395 companies was used. The classification ability of all models is assessed using multiple performance metrics, including overall accuracy, sensitivity, specificity, precision, the F1-score, the F2-score, the Matthews correlation coefficient, and the area under the receiver operating characteristic curve. The results show that both approaches achieve consistently high predictive performance, with all major metrics exceeding 0.92 on the test sample of prosperous and non-prosperous enterprises. Six significant bankruptcy predictors are identified for each method, with three common indicators: financial leverage, total liabilities to assets, and return on costs. The comparative analysis results in a methodological “draw,” confirming comparable predictive power. These findings reaffirm the relevance of classical prediction models and identify key financial indicators that can be used as practical early-warning signals by managers in the sector.

Hongbing Zhou, Rumao Tao, Xi Feng et al.

Machine learning has already shown promising potential in tiled-aperture coherent beam combining (CBC) to achieve versatile advanced applications. By sampling the spatially separated laser array before the combiner and detuning the optical path delays, deep learning techniques are incorporated into filled-aperture CBC to achieve single-step phase control. The neural network is trained with far-field diffractive patterns at the defocus plane to establish one-to-one phase-intensity mapping, and the phase prediction accuracy is significantly enhanced thanks to the strategies of sin-cos loss function and two-layer output of the phase vector that are adopted to resolve the phase discontinuity issue. The results indicate that the trained network can predict phases with improved accuracy, and phase-locking of nine-channel filled-aperture CBC has been numerically demonstrated in a single step with a residual phase of λ/70. To the best of our knowledge, this is the first time that machine learning has been made feasible in filled-aperture CBC laser systems.

Jan Corazza, Ivan Gavran, Daniel Neider

Reward machines are an established tool for dealing with reinforcement learning problems in which rewards are sparse and depend on complex sequences of actions. However, existing algorithms for learning reward machines assume an overly idealized setting where rewards have to be free of noise. To overcome this practical limitation, we introduce a novel type of reward machines, called stochastic reward machines, and an algorithm for learning them. Our algorithm, based on constraint solving, learns minimal stochastic reward machines from the explorations of a reinforcement learning agent. This algorithm can easily be paired with existing reinforcement learning algorithms for reward machines and guarantees to converge to an optimal policy in the limit. We demonstrate the effectiveness of our algorithm in two case studies and show that it outperforms both existing methods and a naive approach for handling noisy reward functions.

Nico Meyer, Christian Ufrecht, George Yammine et al.

Benchmarking and establishing proper statistical validation metrics for reinforcement learning (RL) remain ongoing challenges, where no consensus has been established yet. The emergence of quantum computing and its potential applications in quantum reinforcement learning (QRL) further complicate benchmarking efforts. To enable valid performance comparisons and to streamline current research in this area, we propose a novel benchmarking methodology, which is based on a statistical estimator for sample complexity and a definition of statistical outperformance. Furthermore, considering QRL, our methodology casts doubt on some previous claims regarding its superiority. We conducted experiments on a novel benchmarking environment with flexible levels of complexity. While we still identify possible advantages, our findings are more nuanced overall. We discuss the potential limitations of these results and explore their implications for empirical research on quantum advantage in QRL.

Karthika Balasubramani, Uma Maheswari Natarajan

Traffic go with the flow forecasting is essential in urban planning and management, optimizing transportation structures and resource allocation. However, accurately predicting visitors glide is tough because of its inherent complexity, nonlinearity, and diverse uncertain factors. The trouble declaration underscores the issue in as it should be forecasting site visitors flow, mainly in urban environments characterized through dynamic and complex site visitor’s styles. In the existing paintings there are numerous traditional devices getting to know models used for visitors flow prediction, however those conventional strategies show off barriers in reaching excessive prediction accuracy. Therefore, the proposed work targets to put into effect hybrid optimization techniques for correct prediction in shipping machine. Here fuzzy wavelet neural community (FWNN) is used to address complicated nonlinear structures with uncertain conditions and hybrid optimization method called hybrid firefly and particle swarm optimization (HFO-PSO) which combines the exploration and exploitation talents of firefly and this fusion allows the version to capture intricate visitor’s styles efficiently and optimize the prediction technique, improving accuracy and efficiency. Moreover, the prediction performance of the proposed model is established and compared by means of the usage of distinct measures.

Baejah, Nur Ahmadi, Rahmat Mulyawan et al.

Indoor localization systems are critical in various domains, particularly healthcare, where real-time monitoring of elderly and dementia patients is essential. Current systems face significant challenges in achieving both high accuracy and real-time performance in indoor environments. To address this issue, this study proposes an accurate and real-time indoor localization system that integrates Bluetooth Low Energy (BLE) beacons, wearable device, and advanced machine learning algorithm to enhance room-level localization accuracy. We explored and optimized six machine learning models, including XGBoost, LightGBM, Random Forest, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Multilayer Perceptron (MLP). A Bayesian optimization framework, Optuna, was used to optimize the hyperparameters of machine learning models. Received Signal Strength Indicator (RSSI) data from 15 participants across 10 rooms were collected and processed for performance evaluation and comparison. Based on the experimental results, XGBoost emerged as the highest performing model, with an average accuracy, precision, recall, and F1-score of 0.91. The complete system demonstrates real-time capability, with an end-to-end execution time of 1,346.27 ms. This highlights the system’s potential for practical, accurate, and real-time indoor localization.

Xiaoyi Zhang, Muhammad Usman, Ateeq ur Rehman Irshad et al.

While socioeconomic gradients in regional health inequalities are firmly established, the synergistic interactions between socioeconomic deprivation and climate vulnerability within convenient proximity and neighbourhood locations with health disparities remain poorly explored and thus require deep understanding within a regional context. Furthermore, disregarding the importance of spatial spillover effects and nonlinear effects of covariates on childhood stunting are inevitable in dealing with an enduring issue of regional health inequalities. The present study aims to investigate the spatial inequalities in childhood stunting at the district level in Pakistan and validate the importance of spatial lag in predicting childhood stunting. Furthermore, it examines the presence of any nonlinear relationships among the selected independent features with childhood stunting. The study utilized data related to socioeconomic features from MICS 2017–2018 and climatic data from Integrated Contextual Analysis. A multi-model approach was employed to address the research questions, which included Ordinary Least Squares Regression (OLS), various Spatial Models, Machine Learning Algorithms and Explainable Artificial Intelligence methods. Firstly, OLS was used to analyse and test the linear relationships among selected variables. Secondly, Spatial Durbin Error Model (SDEM) was used to detect and capture the impact of spatial spillover on childhood stunting. Third, XGBoost and Random Forest machine learning algorithms were employed to examine and validate the importance of the spatial lag component. Finally, EXAI methods such as SHapley were utilized to identify potential nonlinear relationships. The study found a clear pattern of spatial clustering and geographical disparities in childhood stunting, with multidimensional poverty, high climate vulnerability and early marriage worsening childhood stunting. In contrast, low climate vulnerability, high exposure to mass media and high women’s literacy were found to reduce childhood stunting. The use of machine learning algorithms, specifically XGBoost and Random Forest, highlighted the significant role played by the average value in the neighbourhood in predicting childhood stunting in nearby districts, confirming that the spatial spillover effect is not bounded by geographical boundaries. Furthermore, EXAI methods such as partial dependency plot reveal the existence of a nonlinear relationship between multidimensional poverty and childhood stunting. The study’s findings provide valuable insights into the spatial distribution of childhood stunting in Pakistan, emphasizing the importance of considering spatial effects in predicting childhood stunting. Individual and household-level factors such as exposure to mass media and women’s literacy have shown positive implications for childhood stunting. It further provides a justification for the usage of EXAI methods to draw better insights and propose customised intervention policies accordingly.

Yuan Li, Bolun Zhang, Dengke Li et al.

This study aimed to investigate the molecular mechanisms of periodontitis and identify key immune-related biomarkers using machine learning and Mendelian randomization (MR). Differentially expressed gene (DEG) analysis was performed on periodontitis datasets GSE16134 and GSE10334 from the Gene Expression Omnibus (GEO) database, followed by weighted gene co-expression network analysis (WGCNA) to identify relevant gene modules. Various machine learning algorithms were utilized to construct predictive models, highlighting core genes, while MR assessed the causal relationships between these genes and periodontitis. Additionally, immune infiltration analysis and single-cell sequencing were employed to explore the roles of key genes in immunity and their expression across different cell types. The integration of machine learning, MR, and single-cell sequencing represents a novel approach that significantly enhances our understanding of the immune dynamics and gene interactions in periodontitis. The study identified 682 significant DEGs, with WGCNA revealing seven gene modules associated with periodontitis and 471 core candidate genes. Among the 113 machine learning algorithms tested, XGBoost was the most effective in identifying periodontitis samples, leading to the selection of 19 core genes. MR confirmed significant causal relationships between CD93, CD69, and CXCL6 and periodontitis. Further analysis showed that these genes were correlated with various immune cells and exhibited specific expression patterns in periodontitis tissues. The findings suggest that CD93, CD69, and CXCL6 are closely related to the progression of periodontitis, with MR confirming their causal links to the disease. These genes have potential applications in the diagnosis and treatment of periodontitis, offering new insights into the disease’s molecular mechanisms and providing valuable resources for precision medicine approaches in periodontitis management. Limitations of this study include the demographic and sample size constraints of the datasets, which may impact the generalizability of the findings. Future research is needed to validate these biomarkers in larger, diverse cohorts and to investigate their functional roles in the pathogenesis of periodontitis.

Nathan Doumèche, Francis Bach, Gérard Biau et al.

Physics-informed machine learning typically integrates physical priors into the learning process by minimizing a loss function that includes both a data-driven term and a partial differential equation (PDE) regularization. Building on the formulation of the problem as a kernel regression task, we use Fourier methods to approximate the associated kernel, and propose a tractable estimator that minimizes the physics-informed risk function. We refer to this approach as physics-informed kernel learning (PIKL). This framework provides theoretical guarantees, enabling the quantification of the physical prior's impact on convergence speed. We demonstrate the numerical performance of the PIKL estimator through simulations, both in the context of hybrid modeling and in solving PDEs. In particular, we show that PIKL can outperform physics-informed neural networks in terms of both accuracy and computation time. Additionally, we identify cases where PIKL surpasses traditional PDE solvers, particularly in scenarios with noisy boundary conditions.

Oresti Banos, Zhoe Comas-González, Javier Medina et al.

Background: Human Emotion Recognition (HER) has been a popular field of study in the past years. Despite the great progresses made so far, relatively little attention has been paid to the use of HER in autism. People with autism are known to face problems with daily social communication and the prototypical interpretation of emotional responses, which are most frequently exerted via facial expressions. This poses significant practical challenges to the application of regular HER systems, which are normally developed for and by neurotypical people. Objective: This study reviews the literature on the use of HER systems in autism, particularly with respect to sensing technologies and machine learning methods, as to identify existing barriers and possible future directions. Methods: We conducted a systematic review of articles published between January 2011 and June 2023 according to the 2020 PRISMA guidelines. Manuscripts were identified through searching Web of Science and Scopus databases. Manuscripts were included when related to emotion recognition, used sensors and machine learning techniques, and involved children with autism, young, or adults. Results: The search yielded 346 articles. A total of 65 publications met the eligibility criteria and were included in the review. Conclusions: Studies predominantly used facial expression techniques as the emotion recognition method. Consequently, video cameras were the most widely used devices across studies, although a growing trend in the use of physiological sensors was observed lately. Happiness, sadness, anger, fear, disgust, and surprise were most frequently addressed. Classical supervised machine learning techniques were primarily used at the expense of unsupervised approaches or more recent deep learning models.

Peter Mostowsky, Vincent Dutordoir, Iskander Azangulov et al.

Kernels are a fundamental technical primitive in machine learning. In recent years, kernel-based methods such as Gaussian processes are becoming increasingly important in applications where quantifying uncertainty is of key interest. In settings that involve structured data defined on graphs, meshes, manifolds, or other related spaces, defining kernels with good uncertainty-quantification behavior, and computing their value numerically, is less straightforward than in the Euclidean setting. To address this difficulty, we present GeometricKernels, a Python software package which implements the geometric analogs of classical Euclidean squared exponential - also known as heat - and Matérn kernels, which are widely-used in settings where uncertainty is of key interest. As a byproduct, we obtain the ability to compute Fourier-feature-type expansions, which are widely used in their own right, on a wide set of geometric spaces. Our implementation supports automatic differentiation in every major current framework simultaneously via a backend-agnostic design. In this companion paper to the package and its documentation, we outline the capabilities of the package and present an illustrated example of its interface. We also include a brief overview of the theory the package is built upon and provide some historic context in the appendix.

Kundan Meshram

David Guile

Yang Wang, Wenli Gao, Shuo Yang et al.

Highlights A humanoid intelligent display platform (HIDP) is created using stretchable and resilient ionotronic materials, and can be applicable in extreme environments and complex mechanical stimulations. HIDP links sound amplitude and brightness through machine learning for audiovisual interaction. HIDP identifies and displays animal species and corresponding frequencies in real-time.

Lorenzo Dall'Amico, Enrico Maria Belliardo

Learning distributed representations, or embeddings, that encode the relational similarity patterns among objects is a relevant task in machine learning. A popular method to learn the embedding matrices $X, Y$ is optimizing a loss function of the term ${\rm SoftMax}(XY^T)$. The complexity required to calculate this term, however, runs quadratically with the problem size, making it a computationally heavy solution. In this article, we propose a linear-time heuristic approximation to compute the normalization constants of ${\rm SoftMax}(XY^T)$ for embedding vectors with bounded norms. We show on some pre-trained embedding datasets that the proposed estimation method achieves higher or comparable accuracy with competing methods. From this result, we design an efficient and task-agnostic algorithm that learns the embeddings by optimizing the cross entropy between the softmax and a set of probability distributions given as inputs. The proposed algorithm is interpretable and easily adapted to arbitrary embedding problems. We consider a few use cases and observe similar or higher performances and a lower computational time than similar ``2Vec'' algorithms.

Rogier van der Sluijs, Nandita Bhaskhar, Daniel Rubin et al.

Image augmentations are quintessential for effective visual representation learning across self-supervised learning techniques. While augmentation strategies for natural imaging have been studied extensively, medical images are vastly different from their natural counterparts. Thus, it is unknown whether common augmentation strategies employed in Siamese representation learning generalize to medical images and to what extent. To address this challenge, in this study, we systematically assess the effect of various augmentations on the quality and robustness of the learned representations. We train and evaluate Siamese Networks for abnormality detection on chest X-Rays across three large datasets (MIMIC-CXR, CheXpert and VinDR-CXR). We investigate the efficacy of the learned representations through experiments involving linear probing, fine-tuning, zero-shot transfer, and data efficiency. Finally, we identify a set of augmentations that yield robust representations that generalize well to both out-of-distribution data and diseases, while outperforming supervised baselines using just zero-shot transfer and linear probes by up to 20%. Our code is available at https://github.com/StanfordMIMI/siaug.

Mrinmoy Roy, Srabonti Das, Anica Tasnim Protity

Obesity, the leading cause of many non-communicable diseases, occurs mainly for eating more than our body requirements and lack of proper activity. So, being healthy requires heathy diet plans, especially for patients with comorbidities. But it is difficult to figure out the exact quantity of each nutrient because nutrients requirement varies based on physical and disease conditions. In our study we proposed a novel machine learning based system to predict the amount of nutrients one individual requires for being healthy. We applied different machine learning algorithms: linear regression, support vector machine (SVM), decision tree, random forest, XGBoost, LightGBM on fluid and 3 other major micronutrients: carbohydrate, protein, fat consumption prediction. We achieved high accuracy with low root mean square error (RMSE) by using linear regression in fluid prediction, random forest in carbohydrate prediction and LightGBM in protein and fat prediction. We believe our diet recommender system, OBESEYE, is the only of its kind which recommends diet with the consideration of comorbidities and physical conditions and promote encouragement to get rid of obesity.

Benoit Dherin

Using backward error analysis, we compute implicit training biases in multitask and continual learning settings for neural networks trained with stochastic gradient descent. In particular, we derive modified losses that are implicitly minimized during training. They have three terms: the original loss, accounting for convergence, an implicit flatness regularization term proportional to the learning rate, and a last term, the conflict term, which can theoretically be detrimental to both convergence and implicit regularization. In multitask, the conflict term is a well-known quantity, measuring the gradient alignment between the tasks, while in continual learning the conflict term is a new quantity in deep learning optimization, although a basic tool in differential geometry: The Lie bracket between the task gradients.

Jean Mouchotte, Matthieu LeBerre, Théo Cojean et al.

Alma Ramić, Ana Matošević, Barbara Debanić et al.

A series of 46 <i>Cinchona</i> alkaloid derivatives that differ in positions of fluorine atom(s) in the molecule were synthesized and tested as human acetylcholinesterase (AChE) and butyrylcholinesterase (BChE) inhibitors. All tested compounds reversibly inhibited AChE and BChE in the nanomolar to micromolar range; for AChE, the determined enzyme-inhibitor dissociation constants (<i>K</i>_i) ranged from 3.9–80 µM, and 0.075–19 µM for BChE. The most potent AChE inhibitor was <i>N</i>-(<i>para</i>-fluorobenzyl)cinchoninium bromide, while <i>N</i>-(<i>meta</i>-fluorobenzyl)cinchonidinium bromide was the most potent BChE inhibitor with <i>K</i>_i constant in the nanomolar range. Generally, compounds were non-selective or BChE selective cholinesterase inhibitors, where <i>N</i>-(<i>meta</i>-fluorobenzyl)cinchonidinium bromide was the most selective showing 533 times higher preference for BChE. In silico study revealed that twenty-six compounds should be able to cross the blood-brain barrier by passive transport. An extensive machine learning procedure was utilized for the creation of multivariate linear regression models of AChE and BChE inhibition. The best possible models with predicted R² (CD-derivatives) of 0.9932 and R²(CN-derivatives) of 0.9879 were calculated and cross-validated. From these data, a smart guided search for new potential leads can be performed. These results pointed out that quaternary <i>Cinchona</i> alkaloids are the promising structural base for further development as selective BChE inhibitors which can be used in the central nervous system.