Raja Kamil

Millions of people around the world suffer from depression. While early diagnosis is essential for timely intervention, it remains a significant challenge due to limited access to clinically diagnosed data and privacy restrictions on mental health records. These limitations hinder the training of robust AI models for depression detection. To tackle this, this article proposes a parallel transfer learning framework for depression detection that integrates BERT and T5 through a fusion mechanism, combining the complementary advantages of these two large language models (LLMs). By integrating their semantic embeddings, the method captures a broader range of linguistic cues from transcribed speech. These embeddings are processed through a model with two parallel branches: a one-dimensional convolutional neural network and a dense neural network are used to construct each branch for preliminary prediction, which are then fused for final prediction. Evaluations on the E-DAIC dataset demonstrate that the proposed method outperforms baseline models, achieving a 3.0% increase in accuracy (91.3%), a 6.9% increase in precision (95.2%), and a 1.7% improvement in F1-score (90.0%). The experimental results verify the effectiveness of BERT and T5 fusion in enhancing depression detection performance and highlight the potential of transfer learning for scalable and privacy-conscious mental health applications.

The growing demand for flawless printed circuit boards (PCBs) necessitates highly precise and reliable defect detection technologies. However, the tiny size of defects, complex textures, and irregular geometries present major challenges to existing vision-based approaches. To overcome these limitations, this study introduces MSMD-YOLO, an enhanced detection framework developed as a next-generation extension of YOLOv11 [1] for PCB inspection. The proposed Multi-Scale Feature Merging (MSFM) module enables comprehensive fusion of structural information across multiple scales, enriching fine-grained feature representation and detail extraction. In parallel, the MDA-Attention module integrates channel-wise and spatial attention mechanisms with dilated convolutions [2], significantly improving semantic discrimination and spatial sensitivity while maintaining low computational complexity. Additionally, the Weighted IoU(WIoU) [3] loss introduces adaptive attention that prioritizes medium-quality anchors and enhances localization of small or low-quality defects. Experimental evaluation on a challenging PCB defect dataset demonstrates that MSMD-YOLO achieves a mean average precision (mAP) [4] of 92.9%, outperforming YOLOv11 by 5.0 percentage points. The model effectively minimizes false detections and missed defects, confirming its robustness and reliability as a practical solution for high-accuracy industrial inspection and automated quality control in advanced PCB manufacturing environments.

The prevalence of diabetes is rising. According to the International Diabetes Federation (IDF) predictions, the number of diabetic patients worldwide will reach 608 million in 2030, accounting for approximately 11.3% of the total number of people in the world. To monitor and predict the future 1 hour glucose have a great significance meaning for patients. This research utilizes a differential evolution (DE) algorithm, an optimized hybrid model transformer and long short-term memory (T-LSTM) technologies to analyze historical data from continuous blood glucose monitoring (CGM) systems and equipment calibration values. The aim is to predict future blood sugar levels in patients, thereby helping to prevent episodes of hypoglycemia and hyperglycemia. The study tested the model using the CGM data from 8 patients at the Suzhou Municipal Hospital in Jiangsu Province, China. Results show that this DE-optimized T-LSTM model outperforms traditional models. The model's accuracy is evaluated using mean squared error (MSE), with MSE values recorded at 15, 30, and 45 minutes being 0.96, 1.54, and 2.31, respectively.

Depression is a serious mental health disorder that poses significant challenges to individuals’ emotional well-being, daily functioning, and overall quality of life. While artificial intelligence methods offer promising solutions in diagnosing, their development typically requires large datasets, which are difficult due to privacy concerns and the complex nature of clinical diagnosis. To address these challenges, this study leverages transfer learning to improve depression detection. Specifically, the method employs the bidirectional encoder representations from transformers (BERT) pre-trained model on large-scale language data, and fine-tunes it on the Extended Distress Analysis Interview Corpus to adapt the model for depression detection. By using the BERT Tokenizer, interview transcripts were tokenized to retain critical linguistic context. These tokens were then processed by the pre-trained bert-base-uncased model to extract robust language features. The features were then passed through a 1-dimensional convolutional neural network (1DCNN) for further analysis, enabling the detection of depression-related patterns. Finally, the refined features were classified via dense layers. To investigate the effectiveness of this method, a total of 11 models-six conventional machine learning models and five neural networks-were tested using three tokenization methods for comparison. Among them, the BERT Tokenizer + 1DCNN model achieved the best performance, with an accuracy of 89.3%, F1 score of 89.4%, and AUC of 95.0%. Notably, transfer learning improved accuracy by 7.7%, highlighting its effectiveness in training neural networks on small datasets. These results demonstrate that the proposed approach not only addresses the issue of limited training data but also significantly enhances the accuracy and reliability of AI-aided depression detection. The method can be applied in computer-aided systems for improving clinical depression diagnostics and other healthcare applications where data scarcity is a barrier.

This study enhances Robotics Flexible Manipulator (RFM) performance by developing a dynamic Mechatronics Test Rig System (MTRS) and a controller for a Combined Single-Link Robotics Flexible Manipulator (CSLRFM). RFMs offer benefits such as lighter weight and faster response but present challenges in motion and vibration control under disturbances. To address this, a Modified Differential Evolution Optimization Algorithm (MDEOA) tunes a PID controller to improve vibration suppression. Additionally, a New Gyroscopic Combined Actuator (NGCA) is introduced for Intelligent Active Vibration Control (IAVC). Comparing (AVC-DEO-PID) and (AVC-FL-PID) controllers with traditional PID, Fuzzy, and (Fuzzy-PID) controllers, the results show (AVC-DEO-PID) performs best, achieving up to about 93.471% faster rise and settling times, eliminating maximum overshoot, and reducing steady-state error by about 90-100%. It is 49-78% faster in peak time, with a 92% reduction in settling time, and shows 90-100% consistency across simulations and experiments. The modified algorithm demonstrates its potential in optimizing CSLRFM performance, making it a robust solution for real-time applications.

There have been several research conducted worldwide on analysis, modelling, and controlling for flexible manipulator architectures. In this paper, one of the most interesting areas in the science of rotational dynamics as the study of spinning solid objects, such as Gyroscopes (G) has been investigated as a new approach for the improvement of the gyroscope non-linear behaviour which controls the magnitude and the direction of the torque separately in the frequency domain by describing a New Graphical Combined Actuator Model (NGCAM) for the Intelligent Active Vibration Control (IAVC) of a Combined Plant (CP) as a linear-nonlinear combined single link Robotics Flexible Manipulator (RFM) structure which rotates in the horizontal plane with two degrees of freedom in order to have a performance assessment and suppressing vibrations. The effect of the parameter variation on the stability and dynamic nonlinear behaviour of the gyroscope in gimbals with a feedback sensor control system has been formed and optimized by the Modified Fuzzy Logic (MFL) algorithm as an artificial intelligence control technique and the Proportional, Integral and Derivative (PID) tuning by the modified Ziegler Nichols (ZN) method. To validate the gyroscope stabilization to be implemented into the RFM for controlling the vibration, the proposed method combines Active Force Control (AFC) technique with Fuzzy Logic (FL) and PID as the (AVC-FL-PID) controller for the NGCAM, CP and RFM by using MATLAB & SIMULINK program is created. For a linear-non-linear combination system, it is better to use a linear-non-linear combination controller. Simulation results illustrate the effectiveness of the proposed strategy which is significantly and quite satisfactory about 25 (%) generally better than compared to the other systems’ performance criteria and is so clear and significant optimization is visible than compared to the other conventional actuators such as PZTs and other control strategies in improvement stabilization and vibration control of RFM structures. The advantages of the proposed method and the possibilities of further improvements are discussed.

This paper introduces some of the various techniques of vibration control and optimization for the purpose of vibration reduction and balancing. Here, in this research by comprising three of the most effective variational techniques now, a Modified Differential Evolutionary Optimization Algorithm (MDEOA) method is suggested to handle the challenge of adjusting the PID controller parameters for the Intelligent Active Vibration Control (IAVC) of a Combined Single Link Robotics Flexible Manipulator (CSLRFM) in order to reduce the undesired effects of vibration. The Crossover Probability Factor (CPF) as the Certain Ratio (CR) and the Mutation Factor (MF) of the algorithm are gradually altered during algorithm iteration to enhance the method's performance during optimization. On this foundation, the PID controller parameter tuning and the issue of CSLRFM mechanical vibrations are addressed using the MDEOA method. This research suggests an evolutionary algorithm that incorporates the variational techniques mentioned above, which will be combined by a certain ratio, and the specific computational procedure. In this strategy, a Strictly Bearish Distributed Exponential Function (SBDEF) has been used as the main target and the criteria and indicators for evaluating and measuring the optimal performance of differential evolution are the Integral Absolute Error (IAE) rate and the PID controller parameter values. According to simulation findings, the technique can be used to optimize the PID controller parameters settings for the IAVC of the CSLRFM and a reduction in the mechanical vibrations. Simulation results illustrate the effectiveness of the proposed MDEOA strategy which is significantly and quite satisfactory about 25 to 30 (%) better than comparing to the other algorithms in improvement stabilization and vibration control of CSLRFM.

The global prevalence of diabetes is escalating, with estimates indicating that over 536.6 million individuals were afflicted by 2021, accounting for approximately 10.5% of the world’s population. Effective management of diabetes, particularly monitoring and prediction of blood glucose levels, remains a significant challenge due to the severe health risks associated with inaccuracies, such as hypoglycemia and hyperglycemia. This study addresses this critical issue by employing a hybrid Transformer-LSTM (Long Short-Term Memory) model designed to enhance the accuracy of future glucose level predictions based on data from Continuous Glucose Monitoring (CGM) systems. This innovative approach aims to reduce the risk of diabetic complications and improve patient outcomes. We utilized a dataset which contain more than 32000 data points comprising CGM data from eight patients collected by Suzhou Municipal Hospital in Jiangsu Province, China. This dataset includes historical glucose readings and equipment calibration values, making it highly suitable for developing predictive models due to its richness and real-time applicability. Our findings demonstrate that the hybrid Transformer-LSTM model significantly outperforms the standard LSTM model, achieving Mean Square Error (MSE) values of 1.18, 1.70, and 2.00 at forecasting intervals of 15, 30, and 45 minutes, respectively. This research underscores the potential of advanced machine learning techniques in the proactive management of diabetes, a critical step toward mitigating its impact.