Predictive Analytics for Blood Supply and Demand: Review

Despite the efforts of the World Health Organization, blood transfusions and delivery are still the crucial challenges in blood supply chain management, especially when there is a high demand and not enough blood inventory. Consequently, reducing uncertainty in blood demand, waste, and shortages has become a primary goal. In this paper, we propose a smart platform-oriented approach that will create a robust blood demand and supply chain able to achieve the goals of reducing uncertainty in blood demand by forecasting blood collection/demand, and reducing blood wastage and shortage by balancing blood collection and distribution based on an effective blood inventory management. We use machine learning and time series forecasting models to develop an AI/ML decision support system. It is an effective tool with three main modules that directly and indirectly impact all phases of the blood supply chain: (i) the blood demand forecasting module is designed to forecast blood demand; (ii) blood donor classification helps predict daily unbooked donors thereby enhancing the ability to control the volume of blood collected based on the results of blood demand forecasting; and (iii) scheduling blood donation appointments according to the expected number and type of blood donations, thus improving the quantity of blood by reducing the number of canceled appointments, and indirectly improving the quality and quantity of blood supply by decreasing the number of unqualified donors, thereby reducing the amount of invalid blood after and before preparation. As a result of the system’s improvements, blood shortages and waste can be reduced. The proposed solution provides robust and accurate predictions and identifies important clinical predictors for blood demand forecasting. Compared with the past year’s historical data, our integrated proposed system increased collected blood volume by 11%, decreased inventory wastage by 20%, and had a low incidence of shortages.

Even though active learning forms an important pillar of machine learning, deep learning tools are not prevalent within it. Deep learning poses several difficulties when used in an active learning setting. First, active learning (AL) methods generally rely on being able to learn and update models from small amounts of data. Recent advances in deep learning, on the other hand, are notorious for their dependence on large amounts of data. Second, many AL acquisition functions rely on model uncertainty, yet deep learning methods rarely represent such model uncertainty. In this paper we combine recent advances in Bayesian deep learning into the active learning framework in a practical way. We develop an active learning framework for high dimensional data, a task which has been extremely challenging so far, with very sparse existing literature. Taking advantage of specialised models such as Bayesian convolutional neural networks, we demonstrate our active learning techniques with image data, obtaining a significant improvement on existing active learning approaches. We demonstrate this on both the MNIST dataset, as well as for skin cancer diagnosis from lesion images (ISIC2016 task).

Climate change substantially threatens public health, including the blood supply chain, which is crucial for medical treatments such as surgeries, trauma care, and chronic disease management. Extreme weather events, vector-borne disease shifts, and temperature fluctuations can disrupt blood collection, testing, transport, and storage, threatening both the safety and sufficiency of blood products. Although studies have highlighted some connections between climate change, transfusion-transmissible infections, and blood safety, there remains a lack of comprehensive understanding of the climate effects on each supply chain stage. In this Personal View, we address the potential climate-driven challenges across the blood supply chain, from donor health to blood component stability, emphasising the importance of proactive measures. To protect the availability and safety of blood supplies in an evolving climate, further research and adaptive strategies are needed to build a resilient blood supply system that can withstand emerging climate-related disruptions.

Statistical Foundations of Data Science gives a thorough introduction to commonly used statistical models, contemporary statistical machine learning techniques and algorithms, along with their mathematical insights and statistical theories. It aims to serve as a graduate-level textbook and a research monograph on high-dimensional statistics, sparsity and covariance learning, machine learning, and statistical inference. It includes ample exercises that involve both theoretical studies as well as empirical applications. The book begins with an introduction to the stylized features of big data and their impacts on statistical analysis. It then introduces multiple linear regression and expands the techniques of model building via nonparametric regression and kernel tricks. It provides a comprehensive account on sparsity explorations and model selections for multiple regression, generalized linear models, quantile regression, robust regression, hazards regression, among others. High-dimensional inference is also thoroughly addressed and so is feature screening. The book also provides a comprehensive account on high-dimensional covariance estimation, learning latent factors and hidden structures, as well as their applications to statistical estimation, inference, prediction and machine learning problems. It also introduces thoroughly statistical machine learning theory and methods for classification, clustering, and prediction. These include CART, random forests, boosting, support vector machines, clustering algorithms, sparse PCA, and deep learning.

The research groups in computer vision, graphics, and machine learning have dedicated a substantial amount of attention to the areas of 3D object reconstruction, augmentation, and registration. Deep learning is the predominant method used in artificial intelligence for addressing computer vision challenges. However, deep learning on three-dimensional data presents distinct obstacles and is now in its nascent phase. There have been significant advancements in deep learning specifically for three-dimensional data, offering a range of ways to address these issues. This study offers a comprehensive examination of the latest advancements in deep learning methodologies. We examine many benchmark models for the tasks of 3D object registration, augmentation, and reconstruction. We thoroughly analyse their architectures, advantages, and constraints. In summary, this report provides a comprehensive overview of recent advancements in three-dimensional deep learning and highlights unresolved research areas that will need to be addressed in the future.

A survey of multimodal hybrid deep learning for computer vision: Architectures, applications, trends, and challenges

PurposeThis paper assembles the 100 most-cited journal articles and reviews on the blood supply chain to analyze the research trends and knowledge themes.Design/methodology/approachThe study conducted a bibliometric analysis of the 100 most-cited journal articles and reviews published between 2004 and 2023 in the context of the blood supply chain, as retrieved from the Scopus database. The study analyzed research trends in this area and knowledge themes through a keyword co-occurrence analysis and bibliographic coupling using VOSviewer.FindingsThe study reveals that the top 100 most-cited journal articles and reviews within the domain of the blood supply chain were published between 2004 and 2023, with a notable surge in research activity observed in the recent years. Through keyword analysis, five themes, which are “inventory management and supply chain management operations for blood products,” “stochastic and possibilistic programming for designing blood supply chain network and sustainability,” “robust optimization in blood supply chain network design under uncertainty and disaster conditions,” “Network design using multi-objective, possibilistic and two-stage stochastic programming in blood supply chains” and “perishable inventory management in blood supply chains,” were obtained. On the other hand, four themes from bibliographic coupling include “designing blood supply chain networks during the time of disasters and emergencies,” “optimizing and managing blood supply chains for enhancing efficiency and reducing wastage,” “inventory management and advanced techniques in the context blood supply chain management” and “enhancing safety, efficiency and continuous blood supply in the face of various challenges.”Originality/valueThis study uniquely analyzed 100 most-cited publications in the realm of the blood supply chain, which provides relevant knowledge themes that are crucial for future studies and contributing to the knowledge in this research domain.

Facial expression recognition (FER) is a highly active field with applications in computer vision, human-computer interaction, security, and computer graphics animation. Recent advancements in deep learning and machine learning have increased interest in utilizing these techniques for accurate facial expression classification. This paper presents a comparative study that evaluates the performance of deep learning and machine learning as classifiers in FER systems, specifically after data fusion. Data fusion techniques combine and integrate multiple sources of information, aiming to enhance the overall classification accuracy by extracting two types of features using geometrical and appearance features trained using two types of convolutional neural networks. The feature outputs of these networks are fused to create a final feature vector for the classification process. The study evaluates the performance of deep learning on two benchmark datasets, the extended Cohn-Kanade (CK+) and Oulu-CASIA datasets, to assess the performance of deep learning. As a point of comparison, the traditional machine learning approach based on the support vector machine (SVM) is also evaluated on the same datasets. Performance metrics such as classification accuracy, precision, recall, and F1-score are utilized. The results obtained from the study highlight the strengths and limitations of both deep learning and machine learning techniques when employed as classifiers in FER systems. Notably, the experimental results demonstrate that the deep learning approach significantly outperforms the baseline methods, achieving an increase in recognition accuracy of 5.22% for the CK+ and 3.07% for the Oulu-CASIA dataset.

<sec><p indent="0mm">The 2024 Nobel Prize in Physics was awarded to John Hopfield and Geoffrey Hinton for their pioneering contributions to artificial neural networks and machine learning. Hopfield was originally trained as a condensed matter physicist, while Hinton has a background in cognitive psychology and artificial intelligence. Both of them recognized the deep connection between neural computation and statistical physics. Their pioneering work demonstrates how principles from statistical physics shaped the theoretical foundations of artificial neural networks and deep learning. This review mainly introduces their breakthrough achievements in neural networks and machine learning, with particular emphasis on the underlying physical principles. </sec><sec> The Hopfield model is one of the most significant contributions of John Hopfield, introducing a groundbreaking theoretical framework for understanding associative memory in machines. This model operates through an iterative dynamic rule, updating neuron states to minimize an energy function, which takes inspiration from spin glass systems in physics. The energy landscape concept in the Hopfield model provides crucial insights into information storage and retrieval. By demonstrating robust distributed representations, the model has inspired extensive research on attractor dynamics in both artificial neural networks and biological systems, serving as a foundational pillar for modern neural architectures and brain-inspired computing. Beyond this model, Hopfield explored time encoding in neural systems, highlighting the role of synchronized oscillations and providing new perspectives on temporal dynamics in enhancing computational capacity. He also pioneered the critical brain hypothesis, linking neural network dynamics to self-organized criticality. </sec><sec> The Boltzmann machine, developed by Geoffrey Hinton and his collaborators, serves as a key architecture bridging statistical physics and machine learning. In this model, the energy function determines the probability distribution of system states following the Boltzmann distribution, with learning based on maximum likelihood estimation. This foundational work led to subsequent innovations, including restricted Boltzmann machines (RBMs), which streamlined the architecture and improved training efficiency. Hinton further advanced deep learning through deep belief networks (DBNs), which stack RBMs into hierarchical architectures, and the contrastive divergence algorithm, which enhanced RBM training efficiency. Beyond the Boltzmann machine, Hinton pioneered advances in backpropagation, deep autoencoders, and techniques like Dropout, optimizing the training process of deep networks. He introduced t-SNE as a powerful visualization tool for high-dimensional data and developed innovative architectures like capsule networks to address limitations in convolutional networks. His forward-forward algorithm represents another significant advancement in learning mechanisms, highlighting his continuous contributions to artificial intelligence. </sec><sec> The Nobel Prize-winning contributions of Hopfield and Hinton exemplify how physical principles can guide the development of revolutionary computational paradigms. Their work has established a bidirectional interaction between the “Science of AI” and “AI for Science”, accelerating interdisciplinary integration and creating new research paradigms that transcend traditional boundaries. In the future, the integration of statistical physics and machine learning will continue to generate new theoretical frameworks for understanding deep learning systems, while also making it possible to solve complex problems in physics and other scientific fields. </sec>

Crack detection plays a vital role in ensuring the structural integrity of various infrastructures, including roads, bridges, and pipelines. Manual inspection methods are time-consuming, labor-intensive, and prone to error. Recent advances in image processing, machine learning (ML), and deep learning (DL) have facilitated the development of automated systems that can efficiently detect cracks with high precision. This paper presents an extensive review of the state-of-the-art methods used for crack detection through these technologies, highlighting their strengths, limitations, and future research directions. Crack detection is an important task in many fields, such as infrastructure inspection and maintenance. Cracks can indicate structural damage and pose safety hazards. Automating crack detection using image processing techniques has gained popularity due to its speed and cost-effectiveness compared to manual inspection methods (Bhat et al., 2020). Traditional methods often rely on manual feature engineering, which can be time-consuming and may not generalize well to different crack types and backgrounds. However, recent advances in deep learning, particularly convolutional neural networks, have shown promising results in automating crack detection (Fei et al., 2023). CNNs can automatically learn hierarchical features from images, making them suitable for detecting cracks with varying shapes, sizes, and textures. Despite the progress, challenges remain in crack detection, such as accurately detecting thin cracks with sub-pixel widths (Pushing the Envelope of Thin Crack Detection, 2021), handling intensity inhomogeneity, and distinguishing cracks from noise and other background clutter (CrackFormer: Transformer Network for Fine-Grained Crack Detection, 2021). Researchers are actively developing more robust and accurate crack detection algorithms using advanced deep learning architectures like Transformers (CrackFormer: Transformer Network for Fine-Grained Crack Detection, 2021) to address these challenges.

Blood bank has a major task to collect blood from donors, monitor blood quality and supply, and distribute blood and blood components to hospitals within the network. Blood distribution is an important activity within this blood supply chain. If the blood bank is able to deliver blood supply to its respective demand in a timely manner, patients’ lives will be saved. Nowadays, many regional blood banks in Thailand confront with ineffective communication channel and insufficient information to fulfill its obligation. Thus, this leads to an inaccurate blood distribution and a waste of time, which can be harmful to patients with critical conditions. It is our goal to develop a web-based system to manage blood requisition within the blood supply chain. The system was designed to cope with this problem. The main objective is to improve the efficiency of data communication within the supply chain to reduce response time for each blood demand request. We also focused on managing blood inventory at each blood bank effectively. The results have shown that the proposed system helps enhancing the communication among blood partners within the supply chain network. The blood bank staffs are able to fulfill blood demand request in a timely manner.

With growing size of resting state fMRI datasets and advances in deep learning methods, there are ever increasing opportunities to leverage progress in deep learning to solve challenging tasks in neuroimaging. In this work, we build upon recent advances in deep metric learning, to learn embeddings of rs-fMRI data, which can then be potentially used for several downstream tasks. We propose an efficient training method for our model and compare our method with other widely used models. Our experimental results indicate that deep metric learning can be used as an additional refinement step to learn representations of fMRI data, that significantly improves performance on downstream modeling tasks.

Deep learning has become increasingly popular due to its remarkable ability to learn high-dimensional feature representations. Numerous algorithms and models have been developed to enhance the application of deep learning across various real-world tasks, including image classification, natural language processing, and autonomous driving. However, deep learning models are susceptible to backdoor threats, where an attacker manipulates the training process or data to cause incorrect predictions on malicious samples containing specific triggers, while maintaining normal performance on benign samples. With the advancement of deep learning, including evolving training schemes and the need for large-scale training data, new threats in the backdoor domain continue to emerge. Conversely, backdoors can also be leveraged to protect deep learning models, such as through watermarking techniques. In this thesis, we conduct an in-depth investigation into backdoor techniques from three novel perspectives. In the first part of this thesis, we demonstrate that emerging deep learning training schemes can introduce new backdoor risks. Specifically, pre-trained Natural Language Processing (NLP) models can be easily adapted to a variety of downstream language tasks, significantly accelerating the development of language models. However, the pre-trained model becomes a single point of failure for these downstream models. We propose a novel task-agnostic backdoor attack against pre-trained NLP models, wherein the adversary does not need prior information about the downstream tasks when implanting the backdoor into the pre-trained model. Any downstream models transferred from this malicious model will inherit the backdoor, even after extensive transfer learning, revealing the severe vulnerability of pre-trained foundation models to backdoor attacks. In the second part of this thesis, we develop novel backdoor attack methods suited to new threat scenarios. The rapid expansion of deep learning models necessitates large-scale training data, much of which is unlabeled and outsourced to third parties for annotation. To ensure data security, most datasets are read-only for training samples, preventing the addition of input triggers. Consequently, attackers can only achieve data poisoning by uploading malicious annotations. In this practical scenario, all existing data poisoning methods that add triggers to the input are infeasible. Therefore, we propose new backdoor attack methods that involve poisoning only the labels without modifying any input samples. In the third part of this thesis, we utilize the backdoor technique to proactively protect our deep learning models, specifically for intellectual property protection. Considering the complexity of deep learning tasks, generating a well-trained deep learning model requires substantial computational resources, training data, and expertise. Therefore, it is essential to protect these assets and prevent copyright infringement. Inspired by backdoor attacks that can induce specific behaviors in target models through carefully designed samples, several watermarking methods have been proposed to protect the intellectual property of deep learning models. Model owners can train their models to produce unique outputs for certain crafted samples and use these samples for ownership verification. While various extraction techniques have been designed for supervised deep learning models, challenges arise when applying them to deep reinforcement learning models due to differences in model features and scenarios. Therefore, we propose a novel watermarking scheme to protect deep reinforcement learning models from unauthorized distribution. Instead of using spatial watermarks as in conventional deep learning models, we design temporal watermarks that minimize potential impact and damage to the protected deep reinforcement learning model while achieving high-fidelity ownership verification. In summary, this thesis investigates the evolving landscape of backdoor threats during the development of deep learning techniques and the use of backdoors for beneficial purposes in intellectual property protection.

This review paper investigates applications of machine learning and deep learning in trading, with a particular emphasis on recent advances in deep learning. It provides an overview of algorithms, including support vector machines (SVMs), random forests, deep neural networks (DNNs), long short-term memory networks (LSTM networks), and deep reinforcement learning (DRL). Findings show that while machine learning and deep learning models were able to surpass traditional strategies in general in terms of profitability, they were also better at risk management. However, despite their performances showing superiority, their performances varied significantly under different market conditions, including markets during periods of high and low volatility. In particular, LSTM networks and random forests can generate substantial returns, higher Sharpe ratios, and lower drawdowns compared to the benchmarks, whereas DNNs struggled during highly volatile periods, as reflected in the returns in the periods. Moreover, the improved DRL agent TradeNet-CR can manage risk significantly better than another, despite not surpassing the original TradeNet-CR model.

Deep self-taught learning for facial beauty prediction