Improved Information-Theoretic Generalization Bounds for Distributed,Federated, and Iterative Learning

We consider information-theoretic bounds on the expected generalization error for statistical learning problems in a network setting. In this setting, there are K nodes, each with its own independent dataset, and the models from the K nodes have to be aggregated into a final centralized model. We consider both simple averaging of the models as well as more complicated multi-round algorithms. We give upper bounds on the expected generalization error for a variety of problems, such as those with Bregman divergence or Lipschitz continuous losses, that demonstrate an improved dependence of 1/K on the number of nodes. These “per node” bounds are in terms of the mutual information between the training dataset and the trained weights at each node and are therefore useful in describing the generalization properties inherent to having communication or privacy constraints at each node.     

高中物理“教-学-评一体化”课堂教学探索——以“示波器的原理——带电粒子在电场中的偏转”为例

"教-学-评一体化"的课堂是指围绕教学目标,教师的教学、学生的学习以及教师对学生的评价组成一个有机的、整体的有效课堂.教学"有效"的唯一证据在于目标的达成,在于学生学习结果的质量,在于何以证明学生学会了什么.因此,教学中要关注对学生的评价.本文以"示波器的原理——带电粒子在电场中的偏转"为例,论述在"教-学-评一体化"的课堂中如何用评价促进学生思维发展.     

Effects of Silk Fibroin Enzyme Hydrolysates on Memory and Learning: A Review

Silk protein products have been used for a wide range of applications. This review focuses on the studies conducted relative to cognitive functions with silk fibroin enzyme hydrolysates (FEH) in humans and animals. All known studies reported in PubMed and Google Scholar have been included. Studies have been conducted on children, high school and college students, adults and seniors, ranging in ages from 7–92 years. Doses of 200–600 mg silk FEH per day for three weeks to 16 weeks have been used. Based on these studies, it can be concluded that silk FEH exhibit beneficial cognitive effects with respect to memory and learning, attention, mental focus, accuracy, memory recall, and overall memory and concentration. These conclusions are supported by studies in rats and mice. Mechanistic studies that have been conducted in animals and cell culture systems are also reviewed. These studies indicate that silk FEH exerts its positive effects on memory and learning by providing neuroprotection via a complex mechanism involving its potent antioxidant and inflammation-inhibiting activities. Acetylcholine (ACh) is secreted by cholinergic neurons, and plays a role in encoding new information. Silk FEH were shown to decrease the levels of the pro-oxidant and pro-inflammatory mediators interlukin-1 (IL-1β), IL-6 and tumor necrosis factor-alpha (TNF-α), protecting the cholinergic system from oxidative stress, thus enhancing ACh levels in the brain, which is known to promote cognitive functions. In addition, the expression of brain-derived neurotrophic factor (BNDF), which is involved in the survival of neurons, is enhanced, and an increase in the expression of the phosphorylated cAMP response element-binding protein (p-CREB) occurs, which is known to play a positive role in cognitive functions. No adverse effects have been reported in association with the use of silk FEH.     

High-Throughput Preparation and Machine Learning Screening of a Blue-Phase Liquid Crystal Based on Inkjet Printing

Blue-phase liquid crystal (BPLC) is considered as the next-generation liquid crystal display material, but its practical application is seriously affected by a narrow temperature range and a long research period. In this paper, we used inkjet printing technology to prepare BPLC materials with high throughput, and try to use machine vision technology to test BPLC with high throughput. The “standard curve method” for establishing each printing channel and the “vector matching method” for searching the chromaticity value of the minimum distance were proposed to improve the accuracy of inkjet printing BPLC materials. For a large number of sample-phase images, we propose a machine learning method to identify the liquid crystal phase. In this paper, for the first time, the high-throughput preparation and high-throughput detection of 1080 BPLC samples with five common components by a comprehensive experimental method has been successfully realized. The results are helpful to improve the research efficiency of blue-phase materials and provide a theoretical basis and practical guidance for rapid screening of multi-component BPLC materials.     

DAD-Net: Classification of Alzheimer’s Disease Using ADASYN Oversampling Technique and Optimized Neural Network

Alzheimer’s Disease (AD) is a neurological brain disorder that causes dementia and neurological dysfunction, affecting memory, behavior, and cognition. Deep Learning (DL), a kind of Artificial Intelligence (AI), has paved the way for new AD detection and automation methods. The DL model’s prediction accuracy depends on the dataset’s size. The DL models lose their accuracy when the dataset has an imbalanced class problem. This study aims to use the deep Convolutional Neural Network (CNN) to develop a reliable and efficient method for identifying Alzheimer’s disease using MRI. In this study, we offer a new CNN architecture for diagnosing Alzheimer’s disease with a modest number of parameters, making it perfect for training a smaller dataset. This proposed model correctly separates the early stages of Alzheimer’s disease and displays class activation patterns on the brain as a heat map. The proposed Detection of Alzheimer’s Disease Network (DAD-Net) is developed from scratch to correctly classify the phases of Alzheimer’s disease while reducing parameters and computation costs. The Kaggle MRI image dataset has a severe problem with class imbalance. Therefore, we used a synthetic oversampling technique to distribute the image throughout the classes and avoid the problem. Precision, recall, F1-score, Area Under the Curve (AUC), and loss are all used to compare the proposed DAD-Net against DEMENET and CNN Model. For accuracy, AUC, F1-score, precision, and recall, the DAD-Net achieved the following values for evaluation metrics: 99.22%, 99.91%, 99.19%, 99.30%, and 99.14%, respectively. The presented DAD-Net outperforms other state-of-the-art models in all evaluation metrics, according to the simulation results.     

Zebrafish: A Pharmacological Model for Learning and Memory Research

Learning and memory are essential to organism survival and are conserved across various species, especially vertebrates. Cognitive studies involving learning and memory require using appropriate model organisms to translate relevant findings to humans. Zebrafish are becoming increasingly popular as one of the animal models for neurodegenerative diseases due to their low maintenance cost, prolific nature and amenability to genetic manipulation. More importantly, zebrafish exhibit a repertoire of neurobehaviors comparable to humans. In this review, we discuss the forms of learning and memory abilities in zebrafish and the tests used to evaluate the neurobehaviors in this species. In addition, the pharmacological studies that used zebrafish as models to screen for the effects of neuroprotective and neurotoxic compounds on cognitive performance will be summarized here. Lastly, we discuss the challenges and perspectives in establishing zebrafish as a robust model for cognitive research involving learning and memory. Zebrafish are becoming an indispensable model in learning and memory research for screening neuroprotective agents against cognitive impairment.     

Quantitative Analysis of Solar Photovoltaic Panel Performance with Size-Varied Dust Pollutants Deposition Using Different Machine Learning Approaches

In this paper, the impact of dust deposition on solar photovoltaic (PV) panels was examined, using experimental and machine learning (ML) approaches for different sizes of dust pollutants. The experimental investigation was performed using five different sizes of dust pollutants with a deposition density of 33.48 g/m² on the panel surface. It has been noted that the zero-resistance current of the PV panel is reduced by up to 49.01% due to the presence of small-size particles and 15.68% for large-size (ranging from 600 µ to 850 µ). In addition, a significant reduction of nearly 40% in sunlight penetration into the PV panel surface was observed due to the deposition of a smaller size of dust pollutants compared to the larger size. Subsequently, different ML regression models, namely support vector machine (SVMR), multiple linear (MLR) and Gaussian (GR), were considered and compared to predict the output power of solar PV panels under the varied size of dust deposition. The outcomes of the ML approach showed that the SVMR algorithms provide optimal performance with MAE, MSE and R² values of 0.1589, 0.0328 and 0.9919, respectively; while GR had the worst performance. The predicted output power values are in good agreement with the experimental values, showing that the proposed ML approaches are suitable for predicting the output power in any harsh and dusty environment.     

Solubility Optimization of Loxoprofen as a Nonsteroidal Anti-Inflammatory Drug: Statistical Modeling and Optimization

Industrial-based application of supercritical CO₂ (SCCO₂) has emerged as a promising technology in numerous scientific fields due to offering brilliant advantages, such as simplicity of application, eco-friendliness, and high performance. Loxoprofen sodium (chemical formula C₁₅H₁₈O₃) is known as an efficient nonsteroidal anti-inflammatory drug (NSAID), which has been long propounded as an effective alleviator for various painful disorders like musculoskeletal conditions. Although experimental research plays an important role in obtaining drug solubility in SCCO₂, the emergence of operational disadvantages such as high cost and long-time process duration has motivated the researchers to develop mathematical models based on artificial intelligence (AI) to predict this important parameter. Three distinct models have been used on the data in this work, all of which were based on decision trees: K-nearest neighbors (KNN), NU support vector machine (NU-SVR), and Gaussian process regression (GPR). The data set has two input characteristics, P (pressure) and T (temperature), and a single output, Y = solubility. After implementing and fine-tuning to the hyperparameters of these ensemble models, their performance has been evaluated using a variety of measures. The R-squared scores of all three models are greater than 0.9, however, the RMSE error rates are 1.879 × 10⁻⁴, 7.814 × 10⁻⁵, and 1.664 × 10⁻⁴ for the KNN, NU-SVR, and GPR models, respectively. MAE metrics of 1.116 × 10⁻⁴, 6.197 × 10⁻⁵, and 8.777 × 10⁻⁵errors were also discovered for the KNN, NU-SVR, and GPR models, respectively. A study was also carried out to determine the best quantity of solubility, which can be referred to as the (x₁ = 40.0, x₂ = 338.0, Y = 1.27 × 10⁻³) vector.     

Deep Image Steganography Using Transformer and Recursive Permutation

Image steganography, which usually hides a small image (hidden image or secret image) in a large image (carrier) so that the crackers cannot feel the existence of the hidden image in the carrier, has become a hot topic in the community of image security. Recent deep-learning techniques have promoted image steganography to a new stage. To improve the performance of steganography, this paper proposes a novel scheme that uses the Transformer for feature extraction in steganography. In addition, an image encryption algorithm using recursive permutation is proposed to further enhance the security of secret images. We conduct extensive experiments to demonstrate the effectiveness of the proposed scheme. We reveal that the Transformer is superior to the compared state-of-the-art deep-learning models in feature extraction for steganography. In addition, the proposed image encryption algorithm has good attributes for image security, which further enhances the performance of the proposed scheme of steganography.