Standard Backpropagation Research Articles

Effect of Intensive Blood Pressure Lowering on the Risk of Incident Silent Myocardial Infarction: A Post Hoc Analysis of a Randomized Controlled Trial.

Silent myocardial infarction (SMI) frequently goes undetected, yet it is associated with increased cardiovascular morbidity and mortality. The impact of intensive systolic blood pressure (SBP) lowering on the risk of SMI in those with hypertension remains uncertain. In this post hoc analysis of the Systolic Blood Pressure Intervention Trial (SPRINT), participants with serial electrocardiograms (ECGs) during the trial were included. SPRINT investigated the benefit of intensive SBP lowering, aiming for < 120 mmHg compared to the standard SBP goal of < 140 mmHg. Incident SMI was defined as evidence of new MI on an ECG without adjudicated recognized myocardial infarction (RMI). During a median follow-up of 3.9 years, a total of 234 MI events (55 SMI and 179 RMI) occurred. Intensive, compared to standard, SBP lowering resulted in a lower rate of SMI (incidence rate 1.1 vs. 2.3 cases per 1000 person-years, respectively; HR [95% CI]: 0.48 [0.27-0.84]). Similarly, intensive, compared to standard, BP lowering reduced the risk of RMI (incidence rate 4.6 vs. 6.5 cases per 1000 person-years, respectively; HR [95% CI]: 0.71 [0.52-0.95]). No significant differences were noted between the strength of the association of intensive BP control on lowering the risk of SMI and RMI (p-value for HR differences = 0.23). This study shows that in adults with hypertension, the benefits of intensive SBP lowering, compared with standard BP lowering, go beyond the prevention of RMI to include the prevention of SMI. ClinicalTrials.gov Identifier: NCT01206062.

Association between intensive blood pressure lowering and stroke-free survival among patients with and without Diabetes

This study pooled data from SPRINT (Systolic Blood Pressure Intervention Trial) and ACCORD-BP (Action to Control Cardiovascular Risk in Diabetes Blood Pressure) trial to estimate the treatment effect of intensive BP on stroke prevention, and investigate whether stroke risk score impacted treatment effect. Of all the potential manifestations of the hypertension, the most severe outcomes were stroke or death. A composite endpoint of time to death or stroke (stroke-free survival [SFS]), whichever occurred first, was defined as the outcome of interest. Participants without prevalent stroke were stratified into stroke risk tertiles based on the predicted revised Framingham Stroke Risk Score. The stratified Cox model was used to calculate the hazard ratio (HR) for the intensive BP treatment. 834 (5.92%) patients had SFS events over a median follow-up of 3.68 years. A reduction in the risk for SFS was observed among the intensive BP group as compared with the standard BP group (HR: 0.76, 95% CI: 0.65, 0.89; risk difference: 0.98([0.20, 1.76]). Further analyses demonstrated the significant benefit of intensive BP treatment on SFS only among participants having a high stroke risk (risk tertile 1: 0.76 [0.52, 1.11], number needed to treat [NNT] = 861; risk tertile 2: 0.87[0.65, 1.16], NNT = 91; risk tertile 3: 0.69[0.56, 0.86], NNT = 50). Intensive BP treatment lowered the risk of SFS, particularly for those at high risk of stroke.

Modelling of Cyber Attack Detection and Response System for 5G Network Using Machine Learning Technique

The rapid increase in the adoption of 5G networks has revolutionized communication technologies, enabling high-speed data transmission and connectivity across various domains. However, the advent of 5G technology comes with an increased risk of cyber-attacks and security breaches, necessitating the development of robust defence mechanisms to safeguard network infrastructure and mitigate potential threats. The work presents a novel approach for modelling a cyber-attack response system tailored specifically for 5G networks, leveraging machine learning techniques to enhance threat detection and response capabilities. The study introduced innovative methodologies, including the integration of standard backpropagation and dropout regularization technique. Furthermore, an intelligent cyber threat classification model that proactively detects and mitigates malware threats in 5G networks was developed. Additionally, a comprehensive cyber-attack response model designed to isolate threats from the network infrastructure and mitigate potential security risks was formulated. The result of testing the response algorithm with simulation, and considering quality of service such as throughput, latency and packet loss, showed 80.05%, 24.9ms and 4.09% respectively. During system integration of the model on 5G network with stimulated malware, the throughput reported 71.81%. Also, packet loss reported loss rate of 23.18%, while latency reported 178.98ms. Our findings contribute to the advancement of cybersecurity in 5G environments and lay the foundation for the development of robust cyber defence systems to safeguard critical network infrastructure against emerging threats.

Optimized Design of Low-Carbon Mix Ratio for Non-Dominated Sorting Genetic Algorithm II Concrete Based on Genetic Algorithm-Improved Back Propagation.

In order to achieve low-carbon optimization in the intelligent mix ratio design of concrete materials, this work first constructs a concrete mix ratio database and performs a statistical characteristics analysis. Secondly, it employs a standard back propagation (BP) and a genetic algorithm-improved BP (GA-BP) to predict the concrete mix ratio. The NSGA-II algorithm is then used to optimize the mix ratio. Finally, the method's accuracy is validated through experiments. The study's results indicate that the statistical characteristics of the concrete mix ratio data show a wide distribution range and good representativeness. Compared to the standard BP, the fitting accuracies of each GA-BP set are improved by 4.9%, 0.3%, 16.7%, and 4.6%, respectively. According to the Fast Non-Dominated Sorting Genetic Algorithm II (NSGA-II) optimization for meeting C50 concrete strength requirements, the optimal concrete mix ratio is as follows: cement 331.3 kg/m3, sand 639.4 kg/m3, stone 1039 kg/m3, fly ash 56 kg/m3, water 153 kg/m3, and water-reducing agent 0.632 kg/m3. The 28-day compressive strength, material cost, and carbon emissions show relative errors of 2.1%, 0.6%, and 2.9%, respectively. Compared with commercial concrete of the same strength grade, costs and carbon emissions are reduced by 7.2% and 15.9%, respectively. The methodology used in this study not only significantly improves the accuracy of concrete design but also considers the carbon emissions involved in the concrete preparation process, reflecting the strength, economic, and environmental impacts of material design. Practitioners are encouraged to explore integrated low-carbon research that spans from material selection to structural optimization.

Intelligent risk identification for drilling lost circulation incidents using data-driven machine learning

In drilling operations, lost circulation (LC) poses a significant challenge, potentially leading to severe accidents such as overflow or blowouts if not accurately identified and controlled. This study introduces a method for identifying LC risk severity using data-driven machine learning and statistical algorithms, specifically demonstrated in the Ying-Qiong Basin of the South China Sea. The method begins with variance filtering to eliminate invalid data, followed by the application of the Pearson correlation coefficient to remove redundant data. This process results in the selection of 12 logging parameters out of the initial 38. Combined with indirect judgments of LC severity based on relevant prior knowledge, the data is labeled to construct the original dataset. A feature extraction and optimization method is proposed, which employs sliding time windows to extract 21 features. These features are then selected and optimized using a random forest algorithm to construct the optimal feature set. To enhance the controllability of the Extreme Learning Machine (ELM) model, the sparrow search algorithm is integrated. Additionally, several fitness functions are introduced for both regression and classification models, resulting in the development of the improved ELM (IELM) model. The dataset is split into training and test sets, and the performance of the IELM model is compared to that of the standard ELM and BP neural network models. The results indicate that the IELM model outperforms the other two, achieving an impressive test F1 score of 97.22 % for accurately identifying LC severity. Moreover, the model trained using the optimal feature set demonstrates higher accuracy and faster convergence. The proposed feature extraction and optimization method effectively increases the amount of information contained in the data, thereby enhancing the model's performance.

Self-Knowledge Distillation via Progressive Associative Learning

As a specific form of knowledge distillation (KD), self-knowledge distillation enables a student network to progressively distill its own knowledge without relying on a pretrained, complex teacher network; however, recent studies of self-KD have discovered that additional dark knowledge captured by auxiliary architecture or data augmentation could create better soft targets for enhancing the network but at the cost of significantly more computations and/or parameters. Moreover, most existing self-KD methods extract the soft label as a supervisory signal from individual input samples, which overlooks the knowledge of relationships among categories. Inspired by human associative learning, we propose a simple yet effective self-KD method named associative learning for self-distillation (ALSD), which progressively distills richer knowledge regarding the relationships between categories across independent samples. Specifically, in the process of distillation, the propagation of knowledge is weighted based on the intersample relationship between associated samples generated in different minibatches, which are progressively estimated with the current network. In this way, our ALSD framework achieves knowledge ensembling progressively across multiple samples using a single network, resulting in minimal computational and memory overhead compared to existing ensembling methods. Extensive experiments demonstrate that our ALSD method consistently boosts the classification performance of various architectures on multiple datasets. Notably, ALSD pushes forward the self-KD performance to 80.10% on CIFAR-100, which exceeds the standard backpropagation by 4.81%. Furthermore, we observe that the proposed method shows comparable performance with the state-of-the-art knowledge distillation methods without the pretrained teacher network.

Prediction of dredged soil settlement based on improved BP neural network

Based on the measured settlement data from the Reclamation Project of Eastern Hengsha Shoal, two improved BP neural networks optimized by the particle swarm optimization (PSO) algorithm and genetic algorithm (GA) respectively are employed to predict the long-term consolidation settlement of the dredged soil in the reclamation area. By comparing the prediction results of the improved neural networks with those obtained from standard BP neural network and various curve fitting methods including hyperbola method, Hoshino method, and Asaoka method, as well as field-measured data, the improved neural network methods are validated to have a better predictive performance. The error analysis results indicate that compared with the standard BP neural network and curve-fitting prediction methods, the improved BP neural networks exhibit smaller prediction errors in terms of MAE, MSE, RMSE, MAPE, etc., and have a closer match between the predicted settlement values and the actual values. The selected GA-BP and PSO-BP neural network prediction methods in this study demonstrate higher prediction accuracy. Meanwhile, GA-BP demonstrates higher prediction accuracy and generalization ability than PSO-BP, but PSO-BP shows faster convergence in terms of prediction speed. The preferred methods presented in this paper exhibit good robustness and high reliability, which can achieve accurate prediction of dredged soil settlement. Additionally, unlike the traditional curve-fitting methods, the selected methods can effectively reduce the influence of subjective factors during the settlement prediction, and provide a new approach for settlement prediction, which can be effectively utilized as a substitute for long-term settlement monitoring and give reference for the design of settlement control of foundation of further construction projects.

Defect Recognition Method for Magnetic Leakage Detection in Oil and Gas Steel Pipes Based on Improved Neural Networks

The aging infrastructure of petroleum and natural gas pipelines poses a threat to national economies, necessitating precise defect detection for safety and efficiency. To enhance the accuracy of predicting pipeline defect sizes, this study introduces a magnetic leakage detection system, employing Backpropagation (BP) neural networks optimized with genetic algorithms. Traditional BP networks face challenges, including parameter determination and slow convergence, addressed through genetic algorithms' global search capabilities. Simulated data are generated using ANSYS software by using models of semi-circular defects in steel pipes, producing magnetic leakage signals of varying intensities. MATLAB was used to construct both standard BP and genetically optimized BP neural networks. Results show that the latter significantly reduces computational errors, demonstrating improved accuracy in defect dimension prediction. The approach contributes to overcoming non-uniqueness in the recognition process and the complex nonlinear relationship between magnetic signals and defect size parameters. The study offers a guided approach for selecting BP neural network parameters, enhancing practicality. Simulations validate the method's effectiveness, indicating low workload and high reliability. This research provides a meaningful advancement in the detection of defects in long-distance pipelines, impacting the safety and efficiency of petroleum and natural gas transportation.

Intensive Blood Pressure Lowering in Individuals With Low Diastolic Blood Pressure and Elevated Troponin Levels in SPRINT.

Among individuals with hypertension and low diastolic blood pressure (DBP), the optimal BP target remains controversial due to concerns that BP lowering may reduce coronary perfusion. We determined the impact of intensive BP control among individuals with elevated systolic BP who have low DBP and elevated hs-cTnT (high-sensitivity cardiac troponin T) levels. A total of 8828 participants in SPRINT (Systolic Blood Pressure Intervention Trial) were stratified by baseline DBP. Those with low DBP (<70 mm Hg) were further stratified by elevated hs-cTnT (≥14 ng/L) at baseline. The effects of intensive versus standard BP lowering on a cardiovascular disease composite end point, all-cause death, and 1-year change in hs-cTnT were determined. The combination of low DBP/high hs-cTnT was independently associated with a higher risk for cardiovascular disease and all-cause death, as well as greater 1-year increases in hs-cTnT, compared with DBP ≥70 mm Hg. However, randomization to intensive versus standard BP lowering led to similar reductions in cardiovascular disease risk among individuals with low DBP/high hs-cTnT (hazard ratio [HR], 0.82 [95% CI, 0.57-1.19]), low DBP/low hs-cTnT (HR, 0.48 [95% CI, 0.29-0.79]), and DBP ≥70 mm Hg (HR, 0.73 [95% CI, 0.60-0.89]; P for interaction=0.20). Intensive BP lowering also led to a reduction in all-cause death that was similar across groups (P for interaction=0.57). In this nonprespecified subgroup analysis of SPRINT, individuals with low DBP and elevated hs-cTnT, low DBP and nonelevated hs-cTnT, and DBP ≥70 mm Hg derived similar cardiovascular disease and mortality benefits from intensive BP lowering. These findings warrant confirmation in other studies.

Auto-Differentiable Transfer Mapping Architecture for Physics-Infused Learning of Acoustic Field

Opportunistic Physics-mining Transfer Mapping Architecture (OPTMA) is a hybrid architecture that combines fast simplified physics models with neural networks in order to provide significantly improved generalizability and explainability compared to pure data-driven machine learning (ML) models. However, training OPTMA remains computationally inefficient due to its dependence on gradient-free solvers or back-propagation with supervised learning over expensively pre-generated labels. This paper presents two extensions of OPTMA that are not only more efficient to train through standard back-propagation but are readily deployable through the state-of-the-art library, PyTorch. The first extension, OPTMA-Net, presents novel manual reprogramming of the simplified physics model, expressing it in Torch tensor compatible form, thus naturally enabling PyTorch's in-built Auto-Differentiation to be used for training. Since manual reprogramming can be tedious for some physics models, a second extension called OPTMA-Dual is presented, where a highly accurate internal neural net is trained apriori on the fast simplified physics model (which can be generously sampled), and integrated with the transfer model. Both new architectures are tested on analytical test problems and the problem of predicting the acoustic field of an unmanned aerial vehicle. The interference of the acoustic pressure waves produced by multiple monopoles form the basis of the simplified physics for this problem statement. An indoor noise monitoring setup in motion capture environment provided the ground truth for target data. Compared to sequential hybrid and pure ML models, OPTMA-Net/Dual demonstrate several fold improvement in performing extrapolation, while providing orders of magnitude faster training times compared to the original OPTMA.

PReFormer: A memory-efficient transformer for point cloud semantic segmentation

The success of transformer networks in the natural language processing and 2D vision domains has encouraged the adaptation of transformers to 3D computer vision tasks. However, most of the existing approaches employ standard backpropagation (SBP). SBP requires the storage of model activations on a forward pass for use during the backward pass, making their memory complexity linearly proportional to model depth, hence, inefficient. Furthermore, most 3D point transformers use the classic QKV matrix multiplication design which comes with a memory bottleneck. To address these issues, we propose a memory-efficient point transformer that makes use of reversible functions and linearized self-attention to minimize SBP and transformer memory complexities, respectively. Additionally, rather than the usual UNet architectural design for segmentation, we adopt a ∇-shaped design to capture multi-size/resolution feature representations towards a finely detailed segmentation. Experimental results on benchmark datasets (Toronto3D, DALES, and CSPC) from different sensor platforms (vehicle, aerial, and backpack) show that our approach uses less than half the number of model parameters compared to its SBP counterpart. It also takes more than twice the input sequence and uses less than half the memory compared to most of the traditional approaches. Additionally, our use of a ∇-shaped architectural design yielded about a 5% increment in model performance over the U-shaped approach. Overall, the proposed PReFormer attained competitive performance compared to the state-of-the-art.

Supervised Local Training With Backward Links for Deep Neural Networks

The restricted training pattern in the standard BP requires end-to-end error propagation, causing large memory costs and prohibiting model parallelization. Existing local training methods aim to resolve the training obstacles by completely cutting off the backward path between modules and isolating their gradients. These methods prevent information exchange between modules and result in inferior performance. This work proposes a novel local training algorithm, BackLink, which introduces inter-module backward dependency and facilitates information to flow backward along with the network. To preserve the computational advantage of local training, BackLink restricts the error propagation length within the module. Extensive experiments performed in various deep convolutional neural networks demonstrate that our method consistently improves the classification performance of local training algorithms over other methods. For example, our method can surpass the conventional greedy local training method by 6.45% in accuracy in ResNet32 classifying CIFAR100 and recent work by 2.58% in ResNet110 classifying STL-10 with much lower complexity, respectively. Analysis of computational costs reveals that small overheads are incurred in GPU memory costs and runtime on multiple GPUs. Our method can lead up to a 79% reduction in memory cost and 52% in simulation runtime in ResNet110 compared to the standard BP. Therefore, our method could create new opportunities for improving training algorithms towards better efficiency for real-time learning applications.

Short-term photovoltaic output power prediction based on similar day and optimized BP neural network

Abstract The photovoltaic (PV) output power is affected by the ambient temperature, seasons, weather and other factors, which makes the PV output power very unstable. Therefore, accurate prediction of the PV output power is highly beneficial. This paper is dedicated to finding a simple and reliable PV short-term output power prediction method. First, we choose four key parameters, season, solar irradiance, temperature and relative humidity, to predict PV output power by using the similar day theory, which is mainly because the above four parameters are decisive for PV output power, although more parameters being taken into account will make the prediction accuracy higher, but it brings along with it an increase in the complexity; secondly, we choose the backpropagation (BP) neural network because it is very suitable for the PV output power prediction due to its excellent learning ability; finally, we optimize the standard BP neural network in loss functions, activation functions and optimizers to further improve its prediction accuracy. We validate the proposed method in different seasons and under other weather conditions. The results show that the proposed method has better prediction results, the optimized BP neural network has better performance compared with the standard BP neural network, and the standard deviation of the prediction is improved from ~1382, ~1571, ~1457, ~989 to ~903, ~792, ~333 and ~409.

Review of neuromorphic computing based on NAND flash memory.

The proliferation of data has facilitated global accessibility, which demands escalating amounts of power for data storage and processing purposes. In recent years, there has been a rise in research in the field of neuromorphic electronics, which draws inspiration from biological neurons and synapses. These electronics possess the ability to perform in-memory computing, which helps alleviate the limitations imposed by the 'von Neumann bottleneck' that exists between the memory and processor in the traditional von Neumann architecture. By leveraging their multi-bit non-volatility, characteristics that mimic biology, and Kirchhoff's law, neuromorphic electronics offer a promising solution to reduce the power consumption in processing vector-matrix multiplication tasks. Among all the existing nonvolatile memory technologies, NAND flash memory is one of the most competitive integrated solutions for the storage of large volumes of data. This work provides a comprehensive overview of the recent developments in neuromorphic computing based on NAND flash memory. Neuromorphic architectures using NAND flash memory for off-chip learning are presented with various quantization levels of input and weight. Next, neuromorphic architectures for on-chip learning are presented using standard backpropagation and feedback alignment algorithms. The array architecture, operation scheme, and electrical characteristics of NAND flash memory are discussed with a focus on the use of NAND flash memory in various neural network structures. Furthermore, the discrepancy of array architecture between on-chip learning and off-chip learning is addressed. This review article provides a foundation for understanding the neuromorphic computing based on the NAND flash memory and methods to utilize it based on application requirements.

Artificial Neural Network-Based Modeling of Performance Spark Ignition Engine Fuelled with Bioethanol and Gasoline

Machine learning technology can distinguish the relationship between engine characteristics and performances. Therefore, the goal of the present work is to predict the performance parameters of a single-cylinder 4-stroke gasoline engine at different ignition timings using a blended mixture of gasoline and bioethanol by an artificial neural network (ANN). Experimental data for training and testing in the proposed ANN was obtained at a dynamic speed and full load condition. An ANN model was developed based on standard Back-Propagation algorithm for the spark ignition engine. Multi-layer perception network (MLP) was used for non-linear mapping between the input and output parameters. An optimizer in the family of quasi-Newton methods (lbfgs) and the rectified linear unit function were used to assess the percentage error between the desired and the predicted values. The network input parameters are engine speed, fuel, and ignition timing. Furthermore, torque, power, specific fuel consumption (SFC), thermal efficiency (ηth), and energy consumption (EC) are taken as output parameters. The results show that ANN is the proper method for predicting SIE performance because it has accurate prediction results that are very similar to experimental results. Moreover, from the observation results, the ANN model can predict the engine performance quite well with correlation coefficient (R)=0.962139 and MSE=0.003967 for data testing.

SpikoPoniC: A Low-Cost Spiking Neuromorphic Computer for Smart Aquaponics

Aquaponics is an emerging area of agricultural sciences that combines aquaculture and hydroponics in a symbiotic way to enhance crop production. A stable smart aquaponic system requires estimating the fish size in real time. Though deep learning has shown promise in the context of smart aquaponics, most smart systems are extremely slow and costly and cannot be deployed on a large scale. Therefore, we design and present a novel neuromorphic computer that uses spiking neural networks (SNNs) for estimating not only the length but also the weight of the fish. To train the SNN, we present a novel hybrid scheme in which some of the neural layers are trained using direct SNN backpropagation, while others are trained using standard backpropagation. By doing this, a blend of high hardware efficiency and accuracy can be achieved. The proposed computer SpikoPoniC can classify more than 84 million fish samples in a second, achieving a speedup of at least 3369× over traditional general-purpose computers. The SpikoPoniC consumes less than 1100 slice registers on Virtex 6 and is much cheaper than most SNN-based hardware systems. To the best of our knowledge, this is the first SNN-based neuromorphic system that performs smart real-time aquaponic monitoring.

Effect of ultrasound‐assisted thin bed drying for retaining the quality of red bell pepper and compare the predictive ability of the mathematical model with artificial neural network

AbstractRed bell peppers (Capsicum annuum L.) are low in calories and high in nutrients, including vitamins A and C. Various factors such as weight loss, senescence, and microbial influence affect their quality during storage. To address this, the present study aims to preserve and retain the quality of red bell peppers using ultrasound‐assisted thin‐bed drying. The results were analyzed using an Artificial Neural Network (ANN) for more accurate prediction of variables involved in the process. Variations in moisture ratio and moisture content during drying were calculated and predicted. The Midilli model provided satisfactory curve fitting at an air velocity of 1.5 m/s, with R2 = 0.9993, χ2 = 0.0002, and RMSE = 0.0134. The two‐term and modified page models fit better with drying curves at air velocities of 2 and 2.5 m/s, with R2 = 0.9995, χ2 = 0.002, RMSE = 0.0016 and R2 = 0.9996, χ2 = 0.00003, RMSE = 0.00003, respectively. However, the trained standard backpropagation ANN algorithm demonstrated excellent predictive ability, outperforming the mathematical models with R2 = 0.9989 (training), MSE = 0.0001 (training), and R2 = 0.9996 (testing), MSE = 0.0002 (testing). Most importantly, the ultrasound‐assisted drying process retains the essential nutrients in red bell peppers, including vitamin C, carotenoids, polyphenols, and flavonoids, across various conditions. The antioxidant potential, as measured by DPPH and FRAP assays, remains largely unchanged compared to untreated samples. However, ABTS activity shows a significant difference at an air velocity of 1.5 m/s and temperatures of 60 and 70°C compared to the control sample.Practical applicationsIn terms of vitamins and antioxidants, red bell peppers are a great provider. Interestingly, capsaicin, the substance responsible for spiciness, is found in very little to no quantity in red bell peppers making it appealable to consumers. To achieve a certain level of processing, the quality of the dried product needs to be monitored while fulfilling the needs of consumers. Particularly, the quality‐determining parameters were vitamin C, total carotenoids, total polyphenols, total flavonoid content, DPPH, ABTS, and FRAP activity. These qualities were retained after drying with the potential novel combination of ultrasound with thin bed drying at low temperatures and air velocity. This means that the bioactive compounds responsible for improving the health of humans are readily available in dried form all year round. The drying kinetics were modeled, predicted, and compared with mathematical and ANN models.

Semantic Knowledge Guided Class-Incremental Learning

Driven by practical needs, research on Class-Incremental Learning (CIL) has received more and more attentions in recent years. A technical challenge to be conquered by CIL methods is the catastrophic forgetting problem, where the model’s performance improves rapidly on new classes while deteriorates drastically on old ones. The main causes behind catastrophic forgetting include network drifts, inter-class confusions, etc. In this paper, we propose a novel CIL method that solves the catastrophic forgetting problem from two aspects. First, to solve the inter-class confusion problem, we propose a novel Semantic knOwledge gUided ciL framework (SOUL) that consists of a CNN feature extractor and a Bi-GCN (Graph Convolutional Network) classifier. In each CIL session, we use the semantic knowledge extracted from the class labels to build two inter-class relation graphs among all the encountered old and new classes. Using these two relation graphs, we develop a Bi-GCN classifier to fuse two kinds of semantic relations in a balanced way, and then to transfer the inter-class relations from semantic modality to image classification weights. The entire SOUL framework is trained end-to-end by the standard BP algorithm, which optimizes the Bi-GCN classifier and the CNN feature extractor jointly. Second, to prevent the network drift, we develop the local topology preserving strategy that divides the global topological structure of the learned feature space into a set of local topological relations, and maintains these local relations at CIL session. Experimental evaluations demonstrate the state-of-the-art performance accuracies on benchmark image classification datasets.

Bio-inspired, task-free continual learning through activity regularization.

The ability to sequentially learn multiple tasks without forgetting is a key skill of biological brains, whereas it represents a major challenge to the field of deep learning. To avoid catastrophic forgetting, various continual learning (CL) approaches have been devised. However, these usually require discrete task boundaries. This requirement seems biologically implausible and often limits the application of CL methods in the real world where tasks are not always well defined. Here, we take inspiration from neuroscience, where sparse, non-overlapping neuronal representations have been suggested to prevent catastrophic forgetting. As in the brain, we argue that these sparse representations should be chosen on the basis of feed forward (stimulus-specific) as well as top-down (context-specific) information. To implement such selective sparsity, we use a bio-plausible form of hierarchical credit assignment known as Deep Feedback Control (DFC) and combine it with a winner-take-all sparsity mechanism. In addition to sparsity, we introduce lateral recurrent connections within each layer to further protect previously learned representations. We evaluate the new sparse-recurrent version of DFC on the split-MNIST computer vision benchmark and show that only the combination of sparsity and intra-layer recurrent connections improves CL performance with respect to standard backpropagation. Our method achieves similar performance to well-known CL methods, such as Elastic Weight Consolidation and Synaptic Intelligence, without requiring information about task boundaries. Overall, we showcase the idea of adopting computational principles from the brain to derive new, task-free learning algorithms for CL.

A Succinct Analysis of Hypertension and its Management Strategies

Hypertension, which is generally known as high blood pressure, holds the utmost significance as a flexible risk factor for overall population worldwide. It is directly related with an increased risk of cardiac disorders. Unfortunately, under half of the sufferer of high BP don’t know about their condition, and the one who know, they are either untreated or not treated properly. However, successful management of this disease has the potential to crucially reduce the worldwide burden of disease and deaths due to hypertension. The cause of hypertension includes a typical reciprocity of natural and physiopathological factors that influence many structures, in addition to genetic predilection. When patients of hypertension are being evaluated with hypertension, this is important to conduct exact and standard BP measurements, assess their predicted risk other heart related problems, determine if there is any indication of organ damage, identify if there is another reason for hypertension, and consider the simultaneous existence of two diseases like cardiac and renal disease. Healthy life management plays a crucial part in lowering BP and inhibiting the occurrence of hypertension and its associated cardiovascular sequelae. These modifications encompass dietary adjustments, increased physical activity, weight management, and other healthy lifestyle choices. Drug therapy or medication is highly effectual in lowering blood pressure and hold back the adverse cardiovascular outcomes in a large no. of patients.