The classification of clinical data is essential for optimizing clinical workflows, facilitating efficient data retrieval, and supporting research aimed at improving patient care. In particular, accurately classifying unstructured surgical reports is crucial for automating data abstraction and organizing key clinical information necessary for both practice and research. However, language models often face challenges stemming from limited and imbalanced datasets, which can result in biased performance and reduced reliability in surgical decision-making. This paper presents a language model–based approach for classifying surgical reports by surgical approach and procedure type, leveraging back translation as a data augmentation strategy to enhance data diversity and model generalization. The proposed method is applied to electronic medical records (EMRs) to distinguish between different surgical procedures and approaches. By enriching the dataset with synthetic examples, the approach improves classification accuracy and mitigates the effects of data scarcity and imbalance in healthcare-related text classification tasks. Experimental results demonstrate that back translation effectively enhances the robustness and reliability of surgical report classification.

Accurate detection and enumeration of mobile robots in indoor environments is a fundamental requirement for applications such as smart warehouses, collaborative robotics, and automated logistics systems. Conventional vision-based approaches for localization and object counting are often constrained by occlusions, illumination variations, and high computational overhead. Channel State Information (CSI) derived from commodity Wi-Fi signals provides a robust alternative, enabling passive, device-free sensing and reliable localization in GPS-denied environments. This paper introduces a CSI-driven framework for estimating the number of mobile robots in a controlled indoor setting and for localizing an individual robot within predefined spatial zones. By exploiting fine-grained channel measurements from existing Wi-Fi infrastructure, the proposed method offers a scalable, low-cost, and real-time solution that effectively mitigates the limitations of traditional vision-based sensing techniques.

Surgical site infections (SSIs) are common postoperative complications that increase patient morbidity, hospital stays, and healthcare costs. Early detection and precise documentation are critical for timely intervention and improved outcomes. While patients often submit images of their wounds through a patient portal for remote monitoring, manual review is challenging due to image variability, high volume, and subjectivity, underscoring the need for automated assessment tools. In this paper, we present Vision Transformer for CAptioning of Incision Images (ViTCAI), a vision-language model designed for automated triage and captioning of postoperative incision images. Fine-tuned on a clinically annotated dataset, ViTCAI improves descriptive accuracy in identifying surgical incisions and SSIs. Our results show that ViTCAI provides consistent, detailed captions that can support clinical decision-making, reducing workload and enhancing diagnostic efficiency in postoperative care.

Advancements in machine learning, coupled with Wi-Fi sensing using channel state information (CSI), have emerged as a powerful approach for human activity and gesture recognition. Contrary to camera-based systems that rely on capturing images of individuals, passive Wi-Fi sensing provides a non-intrusive method for detecting and interpreting human movements without the need for capturing detailed images. In this paper, subcarrier masking contrastive transfer learning (SMCTL) is proposed for human gesture recognition from CSI measurements. The unsupervised feature extractor is pretrained on a large-scale dataset of CSI measurements. Then, the trained feature extractor and a classifier are fine-tuned end-to-end on a smaller dataset collected from a different environment for cross-domain adaptation. Performance of this method is evaluated for various numbers of fine-tuning samples, for which it outperforms baseline supervised and self-supervised deep learning models.

In the realm of robot action recognition, identifying distinct but spatially proximate arm movements using vision systems in noisy environments poses a significant challenge. This paper studies robot arm action recognition in noisy environments using machine learning techniques. Specifically, a vision system is used to track the robot's movements followed by a deep learning model to extract the arm's key points. Through a comparative analysis of machine learning methods, the effectiveness and robustness of this model are assessed in noisy environments. A case study was conducted using the Tic-Tac-Toe game in a 3-by-3 grid environment, where the focus is to accurately identify the actions of the arms in selecting specific locations within this constrained environment. Experimental results show that our approach can achieve precise key point detection and action classification despite the addition of noise and uncertainties to the dataset.

Vision-based methods are commonly used in robotic arm activity recognition. These approaches typically rely on line-of-sight (LoS) and raise privacy concerns, particularly in smart home applications. Passive Wi-Fi sensing represents a new paradigm for recognizing human and robotic arm activities, utilizing channel state information (CSI) measurements to identify activities in indoor environments. In this paper, a novel machine learning approach based on discrete wavelet transform and vision transformers for robotic arm activity recognition from CSI measurements in indoor settings is proposed. This method outperforms convolutional neural network (CNN) and long short-term memory (LSTM) models in robotic arm activity recognition, particularly when LoS is obstructed by barriers, without relying on external or internal sensors or visual aids. Experiments are conducted using four different data collection scenarios and four different robotic arm activities. Performance results demonstrate that wavelet transform can significantly enhance the accuracy of visual transformer networks in robotic arms activity recognition.

Wireless local-area network (WLAN) sensing offers advantages over other approaches to human activity recognition (HAR) for Internet of Things (IoT) applications, including privacy as well as adaptability to non-line-of-sight scenarios. This is why HAR plays an important role in the upcoming IEEE 802.11bf Wi-Fi standard, which aims to bring the adoption of WLAN sensing to a much larger scale. In this paper, we propose CapsHAR, a model based on capsule networks, which uses channel state information (CSI) from Wi-Fi signals to accurately perform human activity recognition. We evaluate the capability of the model on a variety of datasets, including large and small-scale gestures, as well as compare its performance to a variety of models and approaches. We then extend the CapsHAR model into a distributed architecture in order to eliminate the communication overhead of sending CSI data from multiple access points (AP) to a single server. We propose the use of edge computing to run CapsHAR at each AP separately, then combine the outputs of the models through a Fresnel zone-based voting scheme which makes more efficient use of spatial diversity. Overall, the CapsHAR architecture consistently achieves classification accuracy surpassing that of the state-of-the-art models, demonstrating the viability of capsule networks for reliable HAR in Wi-Fi-based IoT applications.

Photoplethysmogram (PPG) is a major indicator of a patient’s physiological status. PPG is generally studied using manually designed algorithms to detect its critical morphological points. However, existing algorithms for analyzing these signals do not serve the purpose effectively and accurately, particularly for abnormal signals. This paper proposes a multi-observation hidden semi-Markov model (HSMM) for PPG signal semantic segmentation, which leverages the information available in raw signal and its first and second derivatives simultaneously. The results indicate the feasibility of PPG signal segmentation with high accuracy using an HSMM and a small training dataset. Moreover, employing a multi-observation approach improves the accuracy significantly.

Machine learning and deep learning models for time series classification generally require a large volume of data to achieve superior performance. However, due to the lack of a sufficient amount of time series in many real-world applications, particularly health care, training these models is more challenging than expected. This paper introduces the Random Frequency Butchering (RFB) method to enhance the generalization performance of classification tasks on limited time series in health care. This approach generates a number of filters with random cutoff frequencies in the frequency domain. The concatenation of time series representations from these filters stacked with the original time series is then used to train an arbitrary time series classifier. The experimental results on the standard medical time series datasets show that the RFB time series representation can significantly enhance the classification performance of the MiniRocket, Inception-Net, and ResNet classifiers.

WiFi sensing is an important part of the new WiFi 802.11bf standard, which can detect motion and measure distances. In recent years, some machine learning methods have been proposed for human activity recognition from WiFi signals. However, to the best of our knowledge, none of these methods have explored orientation prediction of the user using WiFi signals. Orientation prediction is particularly critical for human-machine interaction in an environment with multiple smart devices. In this paper, we propose a data collection setup and machine learning models for joint human orientation and activity recognition using WiFi signals from a single access point (AP) or multiple APs. The results show feasibility of joint orientation-activity recognition in an indoor environment with a high accuracy.

Anatomical movements of the human body can change the channel state information (CSI) of wireless signals in an indoor environment. These changes in the CSI signals can be used for human activity recognition (HAR), which is a pre-dominant and unique approach due to preserving privacy and flexibility of capturing motions in non-line-of-sight environments. Existing models for HAR generally have a high computational complexity, contain very large number of trainable parameters, and require extensive computational resources. This issue is particularly important for implementation of these solutions on devices with limited resources, such as edge devices. In this paper, we propose a lightweight human activity recognition (LiteHAR) approach which, unlike the state-of-the-art deep learning models, does not require extensive training of a large number of parameters. This approach uses randomly initialized convolution kernels for feature extraction from CSI signals without training the kernels. The extracted features are then classified using Ridge regression classifier, which has a linear computational complexity and is very fast. LiteHAR is evaluated on a public benchmark dataset and the results show its high classification performance with a much lower computational complexity in comparison with the complex deep learning models.

Dropout is a well-known regularization method by sampling a sub-network from a larger deep neural network and training different sub-networks on different subsets of the data. Inspired by the dropout concept, we propose EDropout as an energy-based framework for pruning neural networks in classification tasks. In this approach, a set of binary pruning state vectors (population) represents a set of corresponding sub-networks from an arbitrary original neural network. An energy loss function assigns a scalar energy loss value to each pruning state. The energy-based model (EBM) stochastically evolves the population to find states with lower energy loss. The best pruning state is then selected and applied to the original network. Similar to dropout, the kept weights are updated using backpropagation in a probabilistic model. The EBM again searches for better pruning states and the cycle continuous. This procedure is a switching between the energy model, which manages the pruning states, and the probabilistic model, which updates the kept weights, in each iteration. The population can dynamically converge to a pruning state. This can be interpreted as dropout leading to pruning the network. From an implementation perspective, unlike most of the pruning methods, EDropout can prune neural networks without manually modifying the network architecture code. We have evaluated the proposed method on different flavors of ResNets, AlexNet,$l_{1}$pruning, ThinNet, ChannelNet, and SqueezeNet on the Kuzushiji, Fashion, CIFAR-10, CIFAR-100, Flowers, and ImageNet data sets, and compared the pruning rate and classification performance of the models. The networks trained with EDropout on average achieved a pruning rate of more than 50% of the trainable parameters with approximately < 5% and < 1% drop of Top-1 and Top-5 classification accuracy, respectively.

A typical deep neural network (DNN) has a large number of trainable parameters. Choosing a network with proper capacity is challenging and generally a larger network with excessive capacity is trained. Pruning is an established approach to reducing the number of parameters in a DNN. In this paper, we propose a framework for pruning DNNs based on a population-based global optimization method. This framework can use any pruning objective function. As a case study, we propose a simple but efficient objective function based on the concept of energy-based models. Our experiments on ResNets, AlexNet, and SqueezeNet for the CIFAR-10 and CIFAR-100 datasets show a pruning rate of more than 50% of the trainable parameters with approximately < 5% and < 1% drop of Top-1 and Top-5 classification accuracy, respectively.

Pruning is one of the major methods to compress deep neural networks. In this paper, we propose an Ising energy model within an optimization framework for pruning convolutional kernels and hidden units. This model is designed to reduce redundancy between weight kernels and detect inactive kernels/hidden units. Our experiments using ResNets, AlexNet, and SqueezeNet on CIFAR-10 and CIFAR-100 datasets show that the proposed method on average can achieve a pruning rate of more than 50% of the trainable parameters with approximately < 10% and < 5% drop of Top-1 and Top-5 classification accuracy, respectively.

Data representation can facilitate training of deep neural network when limited data is available. We have previously proposed the radial transform sampling method as a data representation technique for training neural networks. In this paper, a probabilistic framework to analyze radial transform is presented. To further elaborate it, performance of training deep neural networks on radial transform generated images for semantic segmentation of kidneys in abdominal computed tomography is evaluated. Our results show that the proposed representation method can achieve higher performance than other similar methods by translating a semantic segmentation problem to a classification problem when limited annotated images are available.

Intelligent Transportation System's main objective is to provide sustainable and optimized means of transportation for citizens of urban centers. Identifying the travel modes used by citizens, in real- time, allows these systems to better adapt transportation infrastructure and services according to user needs and possibilities of interaction. Many works have explored the use of machine learning algorithms and ensemble methods, combined with general and domain specific feature extraction techniques, for this task. More recently, several works evaluated the use of deep learning algorithms while none of them has explored the use of Recurrent Neural Networks (RNNs) in conjunction with domain feature extraction to enable the development of flexible and lightweight travel mode detection solutions based on multiple smartphone sensor readings. In this paper, we propose a deep RNN architecture for building online travel mode detection models using Long-Short Term Memory (LSTM) cells and evaluate its performance using real mobility data collected with a wide variety of sensors. The experiments show that the proposed architecture allows the generation of models with high accuracy and lower memory consumption and computation cost than state-of-the-art supervised machine learning (ML) approaches.

Overfitting is a major problem in training machine learning models, specifically deep neural networks. This problem may be caused by imbalanced datasets and initialization of the model parameters, which conforms the model too closely to the training data and negatively affects the generalization performance of the model for unseen data. The original dropout is a regularization technique to drop hidden units randomly during training. In this paper, we propose an adaptive technique to wisely drop the visible and hidden units in a deep neural network using Ising energy of the network. The preliminary results show that the proposed approach can keep the classification performance competitive to the original network while eliminating optimization of unnecessary network parameters in each training cycle. The dropout state of units can also be applied to the trained (inference) model. This technique could compress the number of parameters up to 41.18% and 55.86% for the classification task on the MNIST and Fashion-MNIST datasets, respectively.

Medical datasets are often highly imbalanced with over-representation of prevalent conditions and poor representation of rare medical conditions. Due to privacy concerns, it is challenging to aggregate large datasets between health care institutions. We propose synthesizing pathology in medical images as a means to overcome these challenges. We implement a deep convolutional generative adversarial network (DCGAN) to create synthesized chest X-rays based upon a modest sized labeled dataset. We used a combination of real and synthesized images to train deep convolutional neural networks (DCNNs) to detect pathology across five classes of chest X-rays. The comparative study of DCNNs trained with the combination of real and synthesized images showed that these networks can outperform similar networks trained solely with real images in pathology classification. This improved performance is largely attributable to the balancing of the dataset using DCGAN synthesized images, where classes that are lacking in example images are preferentially augmented.

Deep learning models have a large number of free parameters that must be estimated by efficient training of the models on a large number of training data samples to increase their generalization performance. In real-world applications, the data available to train these networks is often limited or imbalanced. We propose a sampling method based on the radial transform in a polar coordinate system for image augmentation to facilitate the training of deep learning models from limited source data. This pixel-wise transform provides representations of the original image in the polar coordinate system by generating a new image from each pixel. This technique can generate radial transformed images up to the number of pixels in the original image to increase the diversity of poorly represented image classes. Our experiments show improved generalization performance in training deep convolutional neural networks with radial transformed images.

Medical datasets are often highly imbalanced with over-representation of common medical problems and a paucity of data from rare conditions. We propose simulation of pathology in images to overcome the above limitations. Using chest X-rays as a model medical image, we implement a generative adversarial network (GAN) to create artificial images based upon a modest sized labeled dataset. We employ a combination of real and artificial images to train a deep convolutional neural network (DCNN) to detect pathology across five classes of chest X-rays. Furthermore, we demonstrate that augmenting the original imbalanced dataset with GAN generated images improves performance of chest pathology classification using the proposed DCNN in comparison to the same DCNN trained with the original dataset alone. This improved performance is largely attributed to balancing of the dataset using GAN generated images, where image classes that are lacking in example images are preferentially augmented.

Pathfinding in hospitals is challenging for patients, visitors, and even employees. Many people have experienced getting lost due to lack of clear guidance, large footprint of hospitals, and confusing array of hospital wings. In this paper, we propose Halo; An indoor navigation application based on voice-user interaction to help provide directions for users without assistance of a localization system. The main challenge is accurate detection of origin and destination search terms. A custom convolutional neural network (CNN) is proposed to detect origin and destination search terms from transcription of a submitted speech query. The CNN is trained based on a set of queries tailored specifically for hospital and clinic environments. Performance of the proposed model is studied and compared with Levenshtein distance-based word matching.

Differential evolution (DE) is a high performance and easy to implement evolutionary algorithm. The DE algorithm with small population size (i.e., micro-DE) can further increase the efficiency of the algorithm. However, it also decreases its exploration capability, causing stagnation and pre-mature convergence. In this paper, the idea of exploration enhancement at the mutation level is proposed. The proposed algorithm randomly generates the mutation scale factor for each individual and each dimension of the problem using a uniform distribution. Each individual can select a mutation scheme uniformly and randomly from a pool of mutation schemes in each generation, instead of using a fixed mutation scheme for all individuals during generations. The proposed idea is simple and easy to implement, without changing the algorithm complexity or adding overhead running time. This approach relaxes setting of mutation scheme control parameter. In this paper, we provide a detail analysis about the exploration capability of four variants of micro-DE versions, namely classical micro-DE, micro-DE with vectorized random mutation factor, micro-DE with ensemble mutation scheme, and micro-DE with vectorized random mutation factor and ensemble mutation scheme. Experimental results for various dimensions between 30 to 1000 on the CEC BlackBox Optimization Benchmarking 2015 (CEC-BBOB 2015) show superior performance of the proposed approach compared to the micro-DE and micro-DE with randomized mutation factor algorithms.

The capabilities of evolutionary algorithms (EAs) in solving nonlinear and non-convex optimization problems are significant. Among the many types of methods, differential evolution (DE) is an effective population-based stochastic algorithm, which has emerged as very competitive. Since its inception in 1995, many variants of DE to improve the performance of its predecessor have been introduced. In this context, opposition-based differential evolution (ODE) established a novel concept in which, each individual must compete with its opposite in terms of the fitness value in order to make an entry in the next generation. The generation of opposite points is based on the population's current extreme points (i.e., maximum and minimum) in the search space; these extreme points are not proper representatives for whole population, compared to centroid point which is inclusive regarding all individuals in the population. This paper develops a new scheme that utilizes the centroid point of a population to calculate opposite individuals. Therefore, the classical scheme of an opposite point is modified accordingly. Incorporating this new scheme into ODE leads to an enhanced ODE that is identified as centroid opposition-based differential evolution (CODE). The performance of the CODE algorithm is comprehensively evaluated on well-known complex benchmark functions and compared with the performance of conventional DE, ODE, and some other state-of-the-art algorithms (such as SaDE, ADE, SDE, and jDE) in terms of solution accuracy. The results for CODE are promising.

The concept of opposition-based learning (OBL) can be categorized into Type-I and Type-II OBL methodologies. The Type-I OBL is based on the opposite points in the variable space while the Type-II OBL considers the opposite of function value on the landscape. In the past few years, many research works have been conducted on development of Type-I OBL-based approaches with application in science and engineering, such as opposition-based differential evolution (ODE). However, compared to Type-I OBL, which cannot address a real sense of opposition in term of objective value, the Type-II OBL is capable to discover more meaningful knowledge about problem's landscape. Due to natural difficulty of proposing a Type-II-based approach, very limited research has been reported in that direction. In this paper, for the first time, the concept of Type-II OBL has been investigated in detail in optimization; also it is applied on the DE algorithm as a case study. The proposed algorithm is called opposition-based differential evolution Type-II (ODE-II) algorithm; it is validated on the testbed proposed for the IEEE Congress on Evolutionary Computation 2013 (IEEE CEC-2013) contest with 28 benchmark functions. Simulation results on the benchmark functions demonstrate the effectiveness of the proposed method as the first step for further developments in Type-II OBL-based schemes.

One of the main disadvantages of population-based evolutionary algorithms (EAs) is their high computational cost due to the nature of evaluation, specially when the population size is large. The micro-algorithms employ a very small number of individuals, which can accelerate the convergence speed of algorithms dramatically, while it highly increases the stagnation risk. One approach to overcome the stagnation problem can be increasing the diversity of the population. To do so, a micro-differential evolution with vectorized random mutation factor (MDEVM) algorithm is proposed in this paper, which utilizes the small size population benefit while preventing stagnation through diversification of the population. The proposed algorithm is tested on the 28 benchmark functions provided at the IEEE congress on evolutionary computation 2013 (CEC-2013). Simulation results on the benchmark functions demonstrate that the proposed algorithm improves the convergence speed of its parent algorithm.

Most of the recent proposed approaches for sen-sor(mote) localization are focused on 2-D environments with limited functionalities. This is mostly due to the nature of problem which is non-linear, large-scale, and has limited hardware resources. The micro-evolutionary algorithms (MEAs) utilize a small-size population to solve optimization problems. Therefore, such algorithms require much less processing time and memory than standard evolutionary algorithms (EA), suitable for implementation on embedded systems. In this paper, a novel protocol for localization of motes in 3-D environments is proposed, simulated, and discussed. The localization problem is modeled as an optimization problem. The proposed model is based on a realistic approach to the localization problem, where possible errors and noises in the localization procedure such as signal strength detection are addressed. To present a suitable approach to solve the proposed optimization model, a comparative study on performance of the micro-differential evolution (MDE) algorithms is performed and the results are discussed.

With recent increase in energy consumption as well as reduction of fossil fuels, employing new methods for generation of green energy in smart grids, such as wind energy, is of great interest for governments. That is why expanding of wind turbine farms is a priority in many countries. One of the most important parameters in design and implementation of such farms is optimum selection of wind turbine farm location in a way that the corresponding constraints are met. This paper introduces a new optimization algorithm based on the opposition based ant colony optimization (OACO) algorithm for this aim. Analyzes of simulation results demonstrate performance of the proposed method for optimum localization of wind turbine farms in Saudi Arabia case study.