In the current era of growing interest in health-conscious eating and personalized nutrition, traditional recipe recommendation systems often fail to account for diverse user needs, ingredient availability, and practical cooking constraints. The Multimodal Artificial Intelligence (AI) Framework proposed in this study integrates and analyzes multiple data modalities—textual dietary preferences, food images, and cooking videos—to generate personalized and health-aware cooking recommendations. The framework considers individual health profiles, ingredients detected from visual inputs, and user-specific cooking skill levels inferred from video analysis to tailor recipe suggestions effectively. By leveraging multimodal deep learning algorithms, the system delivers contextually aware, precise, and adaptive recommendations. Experimental evaluations on benchmark and hybrid datasets demonstrate its effectiveness in enhancing recommendation relevance, supporting dietary compliance, and improving overall user satisfaction. These results indicate strong potential for real-world deployment in intelligent culinary assistants, personalized diet planning platforms, and smart health applications.
Online platforms have become integral to modern communication, with images playing a crucial role in conveying information and emotions. However, a significant portion of online imagery contains offensive or inappropriate content, such as hate speech, violence, and nudity, which can have detrimental social and psychological impacts. This research aims to address this challenge by leveraging the power of deep learning, specifically Convolutional Neural Networks (CNNs), to accurately categorize images as offensive or non-offensive. Traditional machine learning algorithms, including Support Vector Machines (SVM), Naïve Bayes, and Decision Trees, have shown limitations in achieving high accuracy in this complex task. These methods often struggle to effectively capture the intricate visual patterns and contextual nuances present in images. In contrast, CNNs, with their inherent ability to automatically learn and extract hierarchical features from images, offer a significant advantage. CNNs can effectively analyze visual elements, text overlays, and contextual cues within images to identify and classify offensive content with high precision. Our research demonstrates the effectiveness of a CNN-based model in classifying images as offensive or non-offensive with an accuracy exceeding 90%. This high level of accuracy significantly improves upon the limitations of traditional methods and has the potential to mitigate the negative social impact of offensive content on online platforms. By effectively identifying and filtering offensive imagery, our model can contribute to creating a safer and more inclusive online environment for all users.
Recognizing emotional intensity is a complex task that exceeds the scientific and biometric recognition of micro-expressions. The methods used by AI, including changes in neurogenerative states, are not reliable in recognizing emotional intensity because, above all, they are unable to distinguish between a highly intense emotion and a simulated emotion, while humans have the innate predisposition to emotion recognition. In fact, this innate predisposition is a necessary component to develop the ability to discern emotional intensity, which is the result of a continuous synchronicity process started in the womb with the exposure to maternal emotional variations. Successively, this capacity improves with the interaction of nature and culture, where prejudice, stereotypes, socio-cultural aspects and gender have an impact on emotional evolution. Finally, the assessment of intensity is closely linked to individual parameters such as personal history, coping responses, personality traits, and other individual factors. This study integrates the perspective of neuroscience with methods used in artificial intelligence for facial micro-expressions recognition and biometric elements. The mechanisms involved in the modulation of emotional responses are integrated here with neurophysiological evidence from profiling and computational approaches to emotion detection. Another element that is considered is free will, especially in the forensic field, highlighting how the incorrect use of AI risks compromising several fundamental rights. As highlighted in this study, human supervision of technicians specialized in profiling, is essential to ensure that purely biometric data is interpreted correctly. A multidisciplinary, human-centered approach is needed, combining robust physiological modeling, transparent algorithms, and strong ethical safeguards.
Artificial Intelligence Generative Tools (AI-GT) have been used recently in various domains, due to their efficiency, ease of use and cost effectiveness. One of these domains is education, where AI-GT can greatly assist the educational process and provide learners with effective tools to enhance their skills and produce high quality outcomes.  However, these tools could cause a threat to the educational process if miss misused, especially to students in the early stages. In this talk, we will discuss the technical details behind the AI-GT, and discuss the most widely used AI-GT that can be used in the educational process, mentioning their pros and cons and how educational institute should deal with them, such as students can benefit from them without negatively affecting their learning capabilities and process.
NIMT is Thailand’s National Metrology Institute. It is an ASEAN most advanced NMI and one of Asia’s centers of excellence in developing measurement standards and measurement methods for the benefit of industry, trade, society and science. NIMT was established by the National Metrology System Development Act B.E.2540 and was founded on 1 June 1998 as a public autonomous agency under supervision of the Ministry of Science and Technology. NIMT is entrusted by the law to establish, develop, and maintain the national measurement standards, in all disciplines, and to disseminate their accuracy and standards and norms, to measurement activities carried out in the country. Today, the 5G technology for sub6G and mmwave with antenna calibration is very important and therefore the , RF/Microwave Laboratory has to prepare the calibration and measurement system for the measured parameters of that system.
This study focuses on the decline in an Indonesian internet provider market share despite increasing revenue and number of customers. This phenomenon is interesting to study further, especially considering the importance of customer satisfaction in maintaining market share. There are few studies that specifically analyze the influence of brand image and brand awareness on continuance intention on fixed broadband products, especially in Indonesia. This study aims to fill the gap in the literature by analyzing the influence of brand image and brand awareness on continuance intention through customer satisfaction on internet provider products in Indonesia. By distributing a structured survey, we took a quantitative strategy. To analyze the data we collected, we used the SEM-PLS method. Our findings show how Brand Image has a positive and significant effect on Customer Satisfaction, Continuance Intention has a positive and significant effect on Customer Satisfaction and Brand Awareness has a positive and significant effect on Continuance Intention through Customer Satisfaction. On the other hand, Brand Awareness does not have a significant effect on Continuance Intention and Customer Satisfaction either directly or indirectly. Also, Brand Image does not significantly affect Continuance Intention.
In this paper, we present a Reinforcement Learning (RL) based strategy for placing optimal charging stations (CS) of electric vehicles (EVs) in the case of Urban planning and smart city development under digital twin. The objective is to minimize the energy required by EVs to reach the CS for recharging. Our approach shows the efficacy of computationally identified CS placement over random placement. Extensive research has demonstrated that an RL-based strategy yields better results in identifying suitable CS locations than random positioning. Based on our investigation, the proposed method finds the most effective positions and some alternative locations for the placement of CS. This study presents a novel approach with 13.15 % enhancement in energy efficiency compared to related research findings. Furthermore, our proposed approach demonstrates expedited attainment of an optimal policy, outperforming existing literature.
There are significant cybersecurity challenges that face wireless sensor networks (WSNs) as a result of their decentralized nature and limited resources although they are highly important in most fields. Traditional security mechanisms frequently fail to cope with the changing and diverse conditions in WSNs. To reduce data transfer but maintain WSNs sensor saturation and data security, this work proposes a prediction-based data fusion and sensing strategy. The suggested method called the ARIMA-SK-EELM system which is made up of Autoregressive Integrated Moving Average (ARIMA), Stable Kernel-Enhanced Extreme Learning Machine (SK-EELM), and threefish algorithm (TFA). In the procedure on data sensing and fusion, ARIMA predicts initially from a few data elements, SK-EELM for precise accuracy on initial expected value similar to actual value while TFA is used during transmissions for both encoded and decoded data. This paper introduces an ARIMA-SK-EELM model with high predictability, low interferences, strong scalability, and secrecy. The results of simulation show that this technique suggested can be effective in reducing unnecessary transfers by accurate forecasting.
Knowledge distillation, particularly in multi-teacher settings, presents significant challenges in effectively transferring knowledge from multiple complex models to a more compact student model. Traditional approaches often fall short in capturing the full spectrum of useful information. In this paper, we propose a novel method that integrates local and global frequency attention mechanisms to enhance the multi-teacher knowledge distillation process. By simultaneously addressing both fine-grained local details and broad global patterns, our approach improves the student model's ability to assimilate and generalize from the diverse knowledge provided by multiple teachers. Experimental evaluations on standard benchmarks demonstrate that our method consistently outperforms existing multi-teacher distillation techniques, achieving superior accuracy and robustness. Our results suggest that incorporating frequency-based attention mechanisms can significantly advance the effectiveness of knowledge distillation in multi-teacher scenarios, offering new insights and techniques for model compression and transfer learning.
Breast cancer, a predominant health concern globally, necessitates advanced diagnostic tools for timely and precise detection. This study endeavored to amalgamate the capabilities of magnetic resonance imaging (MRI) scans with machine learning (ML) to foster enhanced diagnostic accuracy. Employing a comprehensive dataset sourced from three major hospitals, our approach utilized preprocessing techniques to refine MRI image quality, followed by intricate feature extraction focusing on shape, texture, and intensity. Three ML models were implemented, with the Random Forests model emerging as the standout, achieving an impressive accuracy of 92%. This represents a notable improvement over traditional MRI analysis, which registered an accuracy of 84%. When benchmarked against contemporary methods like Deep Learning ConvNets at 88% and Gradient Boosted Trees at 87%, our method consistently outperformed. The results underscore the potential of integrating advanced computational models with medical imaging, promising more reliable and early breast cancer detection. This research serves as a testament to the profound impact of technology on medical diagnostics, offering a promising direction for future endeavors in the realm of breast cancer detection.
In the ever-evolving domain of medical imaging, the integration of deep learning techniques holds the promise of transformative advancements. This research delved into the potential of employing data transfer within deep learning architectures for the automated detection of three distinct lung cancer types. Leveraging sophisticated methodologies like linear discriminant analysis (LDA), t-SNE, and PCA, the study aimed to enhance accuracy and efficiency in detecting malignancies from lung CT scan images. On rigorous evaluation, the models demonstrated compelling accuracy rates: salivary gland-type lung tumors at 90.5%, pleomorphic (spindle/giant cell) carcinoma at 88.2%, and primary pulmonary sarcomas at 91.3%. Additionally, ROC curve analysis further highlighted the robust discriminative capability of the models across varied decision thresholds. The promising results accentuate the potential of integrating data transfer techniques with deep learning in a clinical setting. This research not only exhibits a significant stride in lung cancer detection but also paves the path for further innovations in automated medical image analysis.
A novel monopolar microstrip antenna with symmetric ring-shaped trapezoid ground slots is proposed in this article. The center patch facilitates gap-coupling feeding directly connected with 50 Ohm coaxial line, while six gap-coupled radiators arranged in quasi-circle configuration. The trapezoid ground-slots beneath the six radiators serve to adjust the impedance bandwidth and reduce the overall antenna size. The proposed antenna exhibits good omni-directional radiation pattern with broad bandwidth. Consequently, the antenna supports degenerated non-fundamental TM02 and TM31 modes and demonstrates an impressive impedance bandwidth of 760 MHz from 5.44 GHz to 6.2 GHz with respect to 13% fractional bandwidth. Compared to recent monopolar microstrip antenna research, the proposed antenna shows a smaller size and a thinner substrate. This outcome makes the proposed antenna as ultra-low-profile, and our design stands out by providing a comparable bandwidth.
