Predicting foundation pit deformation is a significant challenge for foundation pit engineering. The inaccuracy of deformation prediction is increased by the intricacy of subterranean space and the variety of construction conditions. Thus, this study develops a deformation prediction model that combines the attention mechanism and bidirectional long short term memory network (BiLSTM) in order to increase the accuracy of deep foundation pit deformation prediction. Meanwhile, to enhance the generalization ability of the model, this study introduces combined regularization in the loss function and adds a Dropout mechanism in the network structure. This study takes a deep foundation pit excavation project in Guangzhou as an example. Experiments shows that the model proposed in the study can complete convergence in about 30 training rounds, and the training loss is maintained at the 0.03 level. Meanwhile, the maximum absolute error of the model in the prediction of verification data is 1.44mm, and the minimum error is 0.001mm. The mean absolute error of the model is 0.311mm, the root mean square error is 0.433mm, and the R2 is 0.906, which is better than the comparison model. The attention mechanism and BiLSTM model suggested in this study provides good generalization performance and high prediction accuracy in deep foundation pit deformation prediction, according to experimental results. Its potential for use in engineering safety management is promising.

Modeling and prediction of desalination performance in a scaled-up membrane capacitive deionization system using machine learning and deep learning.

Accurate timing prediction of surgery is essential for efficient operating room scheduling and ensuring patient care. This study proposes a two-layered Accurate Surgical Time Prediction (ASTP) framework. The first layer combines feature importance and advanced machine learning models to estimate surgical time. After preprocessing, two complementary interpretable AI methods were used: Long Short Term Memory with SHapley Additive exPlanations (SHAP) values and Random Forest permutation importance, to determine the importance of features. In the second layer, subsets of features (TOP-K) were progressively evaluated using HGBR and compared with multiple models: artificial neural networks (ANNs) and recurrent models (Long Short Term Memory, Gated Recurrent Unit, and hybrid Long Short-Term Memory + Gated Recurrent Unit). The proposed framework was evaluated on two datasets: a real-world dataset from an operating room at Nile Hospital and a public dataset from the Medical Informatics Operating Room Vitals and Events Repository (MOVER). In the Nile Hospital dataset, the results show that the Histogram Gradient Boosting Regression (HGBR) approach achieves the best balance, with a Mean Absolute Error of 8.89 min, Root Mean Square Error of 19.5 min, and R-squared of 0.26 using only four features, outperforming other methods. On the MOVER dataset, HGBR also demonstrated the strongest overall predictive behavior, achieving its best numerical result at TOP-12 and best subset at TOP-7, which preserved near-optimal performance with reduced input complexity. The proposed ASTP framework provides an interpretable and resource-efficient solution for surgical time prediction, supporting more intelligent operating room scheduling and facilitating future integration into hospital decision-making.

ABSTRACT Remembering valuable information is important in one’s daily life. Although reward has been shown to enhance true memory through value-directed remembering, its effect on false memory, especially in the long term, is not well understood. We combined the pictorial DRM paradigm with a reward-learning task to investigate the effect of reward on false memory and its persistence over time. In Experiment 1, certain lists of pictures were rewarded, while others were not, then participants’ memories were tested immediately, 1 d later, or 1 week later. In Experiment 2, lists were associated with either a high or low reward. Results from two experiments consistently showed that reward not only enhanced false recognition for lure pictures but also significantly increased false binding between lure pictures and reward, forming false rewarding memories. Importantly, this reward-enhanced false memory effect persisted even after one week, whereas true memories of reward decreased over time. Our results suggest a critical role of time in shaping both false and true episodic memories of reward experiences.

Artificial intelligence enabled fouling prediction and effect of adsorbent sources in submerged fluidized bed ceramic membrane reactor for food industry wastewater treatment.

A comprehensive review of one year that investigates the revolutionary possibilities of integrating IoT (Internet of Things) and ML (Machine Learning) for renewable energy forecasting in Indian healthcare systems, with a focus on the rural village of Vellore, Tamil Nadu. Increased patient mortality due to unpredictable power supplies from various regions around the world. Over 300 articles indexed by SCI (Science Citation Index) from 2014 to 2025 and real case studies address power outages using IoT for real-time monitoring (solar irradiance: 1000 W/m², wind speed: 5–7 m/s) and ML models (LSTM (Long Short Term Memory), CNN-LSTM (Convolutional Neural Network-Long Short Term Memory), XG Boost (Extreme Gradient), RF (Random Forest), ANN (Artificial Neural Network) for forecasting on weekly, monthly, and yearly timescales, expanding on previous research's short and medium term emphasis. A 100-kW hybrid solar-wind system (20 kW PV, 10 kW wind) with edge-cloud optimizes demand (4000–4250 kWh/day for 200 beds) and reduces peak load (200–210 kW) by 15–20%, achieving 85–95% accuracy and 15–35% efficiency. Case studies demonstrate a 5–7-year ROI (Return on Investment), 15–40% CO₂ reduction, and 95% uptime (>99.9% with IoT). NB (Narrow band)-IoT/5G microgrids, subsidies (0.05–0.07 $/kWh), and IoT-ML to raise accuracy by 10–20% are proposed to improve healthcare resilience and global sustainability. This pioneering study lays the groundwork for hospital renewable energy infrastructure assessments. • For rural hilly microgrids, the first hybrid quantum-classical framework tailored to India (QPSO + QAOA + Q-GIS). • 30% lower levelized energy costs (₹8.00 → ₹5.60/kWh) compared to traditional PSO/GA/MILP. • Annual CO 2 emissions were reduced by 25% (from 500 to 375 tons), while the percentage of renewable energy rose to 92%. • In difficult hilly terrain, Q-GIS finds 40% more suitable sites (ANOVA p < 0.05). • Supports SDGs 7 and 13 and India's 500-GW non-fossil target with NISQ-era simulation (Qiskit + PuLP).

This study addressed gaps in prior research on savant syndrome (SS) by restricting participation to individuals with diagnosed autism spectrum disorder (ASD), where SS frequently co-occurs, and focusing on a specific subtype: savant syndrome in calendar calculation (SSC). The primary objectives were to determine the prevalence, intellectual profile, and developmental trajectory of SSC and compare these characteristics with those in savant syndrome in art (SSA), savant syndrome in music (SSM), and non-savant individuals (NSS) with ASD. SSC was identified using broad classification criteria. Intellectual functioning and Wechsler Intelligence Scale indices were compared across the four groups (SSC, SSA, and SSM, and NSS). For SSC, accuracy rates and reaction times on a calendar calculation task were assessed in relation to Full-Scale Intelligence Quotient (FSIQ) and subtest scores. SSC was most frequently observed among individuals with mild intellectual disability. Some participants lost calendar calculation skills over time. Wechsler test findings showed significantly lower FSIQ in SSC than in NSS. By subtest, SSC scored highest on Digit Span, SSA and SSM on Block Design, and NSS on Similarities. All SS groups scored lowest on Comprehension, whereas NSS scored lowest on Coding. In the calendar calculation task, accuracy correlated with Digit Span scores and the discrepancy between Digit Span and Comprehension, while reaction time correlated inversely with FSIQ. Exceptional calendar calculation abilities in ASD were linked to a cognitive profile marked by superior memory (auditory working memory and long term memory) but weak Comprehension.

Objectives Urban road traffic exhibits complex and highly dynamic flow patterns, making real-time risk probability assessment challenging. Existing measures such as Time to Collision (TTC) rely on full sample vehicle trajectory data, which are difficult to obtain at large urban scales. To address this limitation, this study adopts a spatiotemporal grid representation and utilizes floating vehicle trajectory data to characterize traffic operational states and develop a real-time urban traffic risk probability assessment framework. Methods This study proposes an integrated spatiotemporal framework for urban traffic risk probability estimation. A Graph Attention Network Long Short Term Memory (GAT-LSTM)-based Spatiotemporal Autoencoder Neural Network (GL-SANN) is developed to extract latent risk features from vehicle operational parameters. These features are further incorporated into a Deep Clustering Spatiotemporal Network (DCSN) with K-means clustering to assess traffic risk levels. A LightGBM model is then constructed for real-time risk identification. Finally, a Spatiotemporal Graph Convolutional Risk Prediction (SGCRP) model is designed to predict future vehicle operational parameters and infer short-term risk states. Results Experiments using ride-hailing vehicle trajectory data from Xi’an demonstrate that traffic risk probability patterns can be classified into five levels, with risk probability levels increasing significantly during peak periods and decreasing during off-peak periods. Specifically, the proportion of high-risk grids reaches approximately 50%–60% during peak periods, while during off-peak periods, this proportion decreases to about 37%. Intersections and traffic-intensive grids consistently exhibit higher risk probability levels than ordinary grids. The DCSN model consistently outperforms benchmark methods. The LightGBM-based risk identification model achieves a precision of 0.984. The proposed risk prediction model improves training speed by 33.1% over the Transformer model and yields a prediction precision of 0.974. Conclusions These findings provide a method for proactive urban traffic risk prevention and contribute to the development of intelligent transportation systems.

The intersection of age- and experience-dependent processes influence learning throughout development, and developmental learning can have long-term effects on behavior. Phosphorylation of ribosomal protein S6 phosphorylation (pS6) is required in active ribosomes, and new protein synthesis is a conserved mechanism that supports long term memory formation. As such, pS6 fluctuations in brain regions processing experience can provide insight into shifts in the ability for learning and memory. Juvenile male and female zebra finch songbirds (Taeniopygia guttata) perform sensory song learning in ways that affect their adult behaviors. As adults, both sexes perform song recognition learning. Both juvenile and adult types of sensory learning invoke the auditory forebrain. Prior reports established a pS6 song response in the auditory forebrain in Posthatch day 30 juvenile males but not females, and not in younger birds. This was intriguing because the experience-dependent pS6 increase tracked with the onset of the critical period for juvenile sensory song learning in males, and behavioral data indicated that females also effectively learn at P30, though they may not have a critical period. Further, by adulthood (> Posthatch day 90), both male and female auditory forebrains showed an equivalent increase in pS6 after hearing song in patterns consistent with effective recognition learning. Here, to further test the relationships between a crucial molecular marker of active learning processes and the developmental trajectory of juvenile sensory song learning and the emergence of adult-like song recognition learning, we assessed the effect of age and sex, as well as the absence of tutor experience, a manipulation that extends the critical period for developmental learning in males, on the phosphorylation of S6 within the auditory forebrain. Outcomes highlight the complexity of molecular mechanisms across developmental learning and reveal questions to be addressed by further inquiry.

The evolving nature of the threat landscape against cloud services is outpacing the capabilities of traditional security measures. Current firewall implementations in cloud services may provide a foundational layer of security, but they have significant limitations regarding their ability to respond to recently identify zero day vulnerabilities and to protect sensitive information from potential attacks utilizing quantum computing, as well as validating audit logs in multi-tenanted, shared cloud service landscapes. In this research, we present an innovative integrated approach to addressing each of these limitations by combining AI driven Anomaly detection techniques with post quantum cryptography authentication, a Zero Trust Architecture (ZTA), and blockchain based audit logging. Our proposed AI enhanced cloud firewall uses a Long Short Term Memory (LSTM) deep learning model to analyze and classify traffic patterns across IaaS), Platform-as-a-Service (PaaS), and Software-as-a-Service (SaaS) environments and dynamically creates adaptive firewall policies with sub-second response times. Experimental testing on simulated cloud traffic sets demonstrated that our proposed framework achieved a detection rate of 94.7% and a False Positive Rate (FPR) of 2.1%, representing improvements of 24.7% and 12.9%, respectively, when compared to traditional rule-based firewalls. Additionally, the blockchain anchored audit logging mechanism will create tamper proof audit logs, and the post-quantum cryptography layer will prevent attacks using the CRYSTALS-Kyber and CRYSTALS-Dilithium algorithms. These test results demonstrate that our proposed framework is a scalable, resilient, and security hardened solution for future generations of cloud computing landscapes.

Forgetting is an everyday part of life but its precise mechanisms are incompletely understood. Recall errors are often not random. Rather, people often incorrectly report information about the wrong object in memory. In short term memory, this has been referred to as “misbinding”. Here, it has commonly been assumed that the features of an object get swapped around in mind. However, an alternative mechanism is that information about a feature of one object might be lost, and replaced by another object’s feature. Commonly-used cued recall approaches are blind to this distinction, but testing multiple objects on the same trial has the power to detect this. We asked people to report all features of all objects from an array (multifeature whole-report) in any order (free recall). This enabled us to directly quantify these subtypes of misbinding in memory. We introduce a probabilistic model that shows that misbinding actually includes a mixture of symmetric swaps and asymmetric misattributions where a forgotten feature gets replaced by a feature of another object in memory without a reciprocal exchange. This distinction is observed even when memory objects are encoded sequentially.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-52649-7.

Diabetes is a common chronic disease that needs early diagnosis and proper management to avoid severe complications. While current Artificial Intelligence (AI) tools generate predictive information, they often lack an integrated element for post-diagnosis support in order to fill in this critical gap in patient self-management. This research proposes and validates a hybrid System which aims to bridge this gap. The methodology is based on a novel, fused dataset (PIMA and Type 2 Diabetes) that was carefully preprocessed following a leakage-safe protocol in order to increase generalizability. The system architecture is a combination of two different critical components: Strong Bidirectional Long Short term Memory (BiLSTM) model for prediction and rule based engine for creating personalized lifestyle recommendations. In order to validate the efficacy of the BiLSTM model, seven traditional machine learning (ML) models and standard deep learning (DL) models have been comparatively tested, in which BiLSTM model has demonstrated a better generalization and prediction performance. Rigorous 10-fold cross validation was used to validate the system, which came up with an accuracy of 84.02%, precision of 87.89%, and recall of 80.50%. This research concludes that by successfully combining the high-performance predictive engine and real-time guidance module, it is possible to develop a holistic clinically relevant tool to close the loop between diagnosis and proactive self-management.

Super Smart Grids (SSG) aim to provide large-scale, multi-zonal electricity access while dynamically balancing supply and demand. However, their implementation faces multidisciplinary challenges that range from ensuring grid stability to avoiding structural injustices in their design. Existing load forecasting approaches are unsuitable for SSG planning or deployment due to an over-reliance on regional specificity, time-series data, and a lack of social understanding. This paper proposes a hybrid approach to load forecasting that combines time-series power consumption and socioeconomic metrics, coupled with a novel deep learning algorithm that integrates Artificial Neural Network (ANN) and Luong's Attention Mechanism (LAM). First, two parallel ANNs are used to extract the main features of load demand behavior. Then, LAM fuses both ANNs using an attention score function that dynamically selects the most relevant characteristics per sample to improve the generalization abilities of the model. The SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) frameworks then interpret this algorithm to thoroughly comprehend its load forecast decision-making. Evaluated on 92 suburban zones in Australia, the proposed method achieves a Mean Absolute Percentage Error (MAPE) of 1.78%, outperforming Bidirectional Long-Short Term Memory (BiLSTM), Long-Short Term Memory (LSTM) and Recurrent Neural Network (RNN) when socioeconomic metrics are present during training. By merging high-resolution forecasting with socioeconomic awareness, this approach enhances demand and supply management, optimizes pricing strategies, and ensures equitable energy distribution-critical requirements for SSG deployment.

The G protein-coupled estrogen receptor (GPER) participates in nociception. The GPER inverse agonist, a tetrapeptide (PLMI), was studied on pain-like symptoms, in murine models of chemotherapy-induced peripheral neuropathy. All experiments were performed in mice. We used the PLMI and the GPER antagonist G15 to study the role of GPER in mechanical allodynia in the model of paclitaxel-induced peripheral neuropathy. Sites of GPER/PLMI actions were explored by using nociceptors and dorsal horn GPER knockouts. The effect of PLMI and/or G15 was assessed in dorsal root ganglia primary cultures to explore the role of GPER in neuronal calcium flux. After a chronic administration of PLMI, the antinociceptive and neuroprotective effects were investigated in the paclitaxel-induced neuropathy model. Short term memory, reward-related conditioning and acute effects in bortezomib- and oxaliplatin-induced nociception were evaluated. NMR and CD spectroscopy were used to determine the conformation of the peptide PLMI, in solution. In our paclitaxel-induced pain-like symptoms model, peripheral, spinal and supraspinal GPER participates in nociception. The peptide PLMI decreases nociception by lowering intraneuronal free calcium flux. Chronic PLMI treatment reduces pain-like behaviours and protects against nerve conduction velocity deficits, without causing cognitive impairments or addiction. PLMI alleviates oxaliplatin- and bortezomib-induced neuropathic pain. The peptide PLMI adopts a turn conformation. Our results suggest that GPER inverse agonists could be used to alleviate nociception and to protect against paclitaxel-induced peripheral neuropathy. The turn conformation of the PLMI peptide in solution is in favour of a bioactive GPCR-interacting peptide.

ABSTRACT Integrating the intrinsic in‐memory computing and temporal dynamic response characteristics of memristors directly into the design of time‐varying information processing and hardware security is pivotal for boosting the efficiency, security, and dynamic adaptability of big data‐driven artificial intelligence. Herein, we report an Ag/AgBiS 2 /Mo synaptic memristor featuring a high switching ratio (&gt;10 5 ) and long‐term retention (&gt;10 4 s), which exhibits electrically and optically modulated diverse synaptic plasticity behaviors, including excitatory postsynaptic current (EPSC), paired‐pulse facilitation/depression (PPF/D), long‐term potentiation/depression (LTP/D), short/long‐term memory (S/LTM) spike‐timing/rate/duration/voltage‐dependent plasticity (STDP, SRDP, SDDP, SVDP). Notably, this AgBiS 2 ‐based memristor enables co‐integrated two core functionalities: 1) Model‐level encryption and authorized inference. Device‐based physically unclonable constraints yield an inference accuracy of 88.5% under a valid authorization key, whereas the accuracy plummets to 18.9% with an invalid key. 2) Time‐fading trajectory (TF‐Traj) modeling and dynamic perception. A trajectory classification accuracy of 95.2% is achieved on the constructed 2D TF‐Traj with significantly enhanced accuracy and stability of single‐step trajectory prediction. The results directly bridge the device physics, dynamic intelligent perception, and information security, highlighting the promise of memristor's intrinsic dynamics for advancing trustworthy, multi‐scenario neuromorphic computing and high‐confidence intelligent systems.

How do neural circuits change to incorporate newly learned events? After decades of research, we now have a good understanding of the diversity and duration of memory types, the context-specifics of learning regimes, and the brain areas that are likely to be involved. However, we are still far from a mechanistic understanding of the neural activity required to transform new behavioral experiences into long term memory accessed during recall events. What kind of network activity is required to affect these wide-spread changes in neuronal circuitries? Among current theories of memory in mammals, one of the most intriguing is concerned with the role of large-scale synchronous neural activity, which is thought to enable the transfer of information from the hippocampus to numerous regions of the neocortex. In this contribution, we first review sleep, learning, and memory in birds before highlighting evidence that large-scale synchronous neural activity also exists in the avian brain, making the case that by examining these questions from a comparative research perspective, we gain important insight into the canonical features of memory consolidation.

In the modern era of working, Musculoskeletal Disorders (MSDs) are increasing drastically. One of the leading causes of MSD is Low Back Pain (LBP). Patient health monitoring technology is paramount to the investigators, enabling remote recovery services via cutting-edge technologies that lower the barrier between clinicians and patients. This work provides a low-cost, efficient, and user-friendly visual capture recovery system for the administration of Low Back Pain (LBP). This study proposes a unique computer vision and deep learning method for remotely monitoring patients' joint angles during physiotherapy rehabilitation. A single long-short term memory layer with 64-unit lightweight model with dense neurons was used to identify the correct postures for LBP recovery exercises in real-time video. The proposed system exploits a 3D human skeleton representation for calculating angles on three landmarks to recognize the angle deviations and classify the nine LBP recuperation exercise poses with high cross-validation accuracy, low computational cost, real-time exercise correction feedback, and minimal latency to process frames. The suggested approach successfully predicts and provides feedback on LBP exercise postures from real-time video feeds captured by common RGB cameras, without additional hardware or specialist cameras, thereby improving the quality of life for people around the globe.

The rapid expansion of Electronic Medical Record (EMR) data has advanced AI-driven patient similarity computation, a key technology for intelligent healthcare. However, the handling of heterogeneous EMR formats and the integration of domain knowledge constrain existing methods. While graph-based approaches show promise, they still struggle with these issues. To address this, we propose a Large Language Model-Dynamically Enhanced Learning Network (LELN), leveraging LLMs' commonsense knowledge and reasoning to dynamically structure EMR data and enhance medical knowledge integration. LELN in tegrates two LLM-basedmodules:DS-EE(DeepSeek-Event Extraction) extracts medical events to construct structured EMR event graphs, and DS-KB (DeepSeek-Knowledge Base) infers disease-relevant knowledge to augment feature representations. The model employs a dual-stage spatial-temporal feature aggregation strategy: a Graph Attention Network captures intra- and inter-event dependencies, followed by a Bidirectional Long-Short Term Memory (BiLSTM) with attention to model temporal disease progression. Additionally, a clinical prior-guided attention mechanism emphasizes discriminative diagnostic features, improving clinical relevance. Extensive experiments on heterogeneous datasets-a real-world Chinese dataset and public MIMIC-III-show LELN outperforms baselines, achieving F1 scores of 87.66% and 85.95%, demonstrating robustness and accuracy.

Machine translation has increasingly shifted toward Neural Machine Translation (NMT) because of its ability to handle input and output sequences of varying lengths. The incorporation of attention mechanisms in NMT systems enables the model to focus on the most relevant parts of the source sentence, rather than relying solely on a fixed representation of the entire input. While NMT improves translation quality by addressing long-range dependencies and contextual understanding, it also requires a large parallel corpus for training, which is a challenge for languages with less resources. The main focus of this research is to give solution for the unique challenges of translating Ayurvedic texts using NMT. Ayurvedic texts have collection of special and scientific words related to medicines and treatments. This makes the translation process more complex and needs very efficient approach for accurate translations. Also, the content of ayurvedic text books is in the form shlokas which is formed using very complex and compound words. In order to simplify the translation process efficiently this work uses a sandhi splitter module and an Anvaya Generator/ word reordering module. In order to develop NMT system for low resource language pair Sanskrit-Malayalam, there is a need of developing a parallel corpus especially for Ayurvedic text books. Also, as the NMT model is proposed for translation it requires a minimum amount of parallel data in the corpus. So, a number of general domain Sanskrit text books with verses, called shlokas, were also considered for developing parallel corpora. The authors developed a parallel corpus for Anvaya Generator, sandhi splitter and translation. Mainly four NMT models were developed trained and tested especially for shlokas as input. The two models are basic transformer model with attention and an encoder-decoder model using Long-Short term Memory (LSTM) with attention. The other two are developed by adding two modules called Sandhi Splitter and Anvaya Generator in the pre-processing stages of the earlier models- Transformer based model and LSTM based model. The limitations of low resources and richness in grammatical structure of Sanskrit- Malayalam language pair are overcome by the concepts of deep learning and the additional modules used in preprocessing stages for developing the models. The models were tested with and without sandhi splitter and Anvaya Generator modules. The transformer-based model integrated with sandhi splitter and Anvaya Generator system achieved a higher average BLEU score of 73.11 and a uni-gram BLEU score of 76.93 for Sanskrit verses to Malayalam translation.

In order to solve the problem that traditional fake news detection methods cannot extract the semantic information of short news well. We propose a fake news detection method based on the fusion of differential features from multi‐ perspective. Firstly, different pre‐trained models were used to extract features from different perspectives, and siamese convolutional neural networks were used to obtain text structure differential features. Meanwhile, siamese bidirectional long term memory networks were further used to obtain text sequence differential features. Both features were integrated to achieve fake news detection. The experimental results on public Chinese and English datasets show that the model used has improved on various indicators compared with the traditional fake news detection model. The integration of differential features from multiple perspectives can effectively extract the semantic information of short news and improve the detection ability of fake news. © 2026 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.