Currently, most traditional interactive art spaces are static, and it is difficult to achieve real-time response to the behaviors and emotions of the audience. This study proposes the human-centered design and digital twin simulation framework, which integrates human-centered design and digital twin simulation. It collects behavioral and emotional data through multimodal sensing to dynamically adjust lighting, sound, and visuals. The framework is deployed in three physical immersive art exhibitions, involving 82 participants aged 18–67, including 19 visitors with neurodiversity. The results show that the digital twin adaptation accuracy rate is 91% ± 4.2%, the interaction effectiveness is 88.4% ± 6%, the sensor integration accuracy rate is 95% ± 3.8%, the adaptation delay is reduced by 175 milliseconds ± 15 milliseconds, and the emotional recognition accuracy rate is 91% ± 5% (all p < 0.01). The four-week longitudinal data indicates that the duration of stay and emotional investment continue to increase, and the performance across art forms is superior to the baseline.
Traditional human resource management capability assessment mostly adopts static frameworks and single indicators, which are insufficient to support dynamic decision-making in recruitment, training, and talent inventory. This paper constructs a deep learning dynamic assessment model (DL-DEM) integrating structured personnel data, unstructured feedback, and collaboration traces to form a quarterly time series sequence. The model outputs continuous scores ranging from 0 to 1 and divides capabilities into four levels. Experiments were conducted using a public dataset from IBM and longitudinal quarterly data from manufacturing and service enterprises. The MAE of this model is 0.413, the RMSE is 0.287, the F1 is 0.859, and the AUC is 0.917. The fairness deviation index is controlled at a relatively low level. Compared with models such as gated recurrent unit, temporal convolutional network, and transformer, DL-DEM has better prediction accuracy, temporal consistency, and interpretability and can provide tools for dynamic talent monitoring and human resource strategy formulation.
AI-enabled smart logistics offers predictive, collaborative support for sustainable campus operations by managing distributed facility resources and utilities. This work presents EcoCollab-AVF, a leakage-safe adaptive validation fusion framework embedding multi-utility forecasting, anomaly detection and e-collaboration into a closed-loop workflow. Tested on the UNICON multi-campus dataset (216 030 hourly records), it outperformed most baselines, delivering a macro-SMAPE of 25.529% and yielding notable predictability gaps across electricity, gas and water. Explainable residual scoring produced prioritised facility-service logistics decisions. This research centres on operational-resource coordination, not conventional logistics, and makes no verified claims about collaboration, carbon or cost savings.
This research proposes a Tangent Search-tuned Conditional Generative Adversarial Network with Neural Style Transfer (TS-CGAN-NST) model for digital media aesthetic education. The proposed model integrates a Conditional Generative Adversarial Network (CGAN) with a Neural Style Transfer (NST) mechanism to generate aesthetically coherent digital artwork while preserving semantic consistency and artistic characteristics. Data preprocessing includes missing value imputation, text cleaning, one-hot encoding, and Min-Max normalization, followed by Principal Component Analysis (PCA) for informative feature extraction and dimensionality reduction. The Tangent Search (TS) algorithm optimizes generator parameters to improve convergence, visual harmony, and style consistency. The proposed TS-CGAN-NST achieves an SSIM of 0.982, PSNR of 41.86 dB, FID of 7.43, Style Consistency of 98.92%, and Aesthetic Quality of 98.15%, outperforming existing methods in generating visually coherent and aesthetically appealing digital media content.
This research presents a Deep Learning and Internet of Things-Based students' Psychological Crisis Prediction Algorithm (DL-IoT-PCPA). This algorithm combines data from wearable Internet of Things sensors, mobile interactions, and academic achievement. A Convolutional Neural Network (CNN) and a Bidirectional Long Short-Term Memory (BiLSTM) network will be used in the DL-IoT-PCPA architecture to illustrate temporal activity that is the consequence of the extraction of features from heterogeneous data streams. There is a possibility that the system would be assisted in connecting physiological and environmental data from various locations by using an Adaptive Feature Fusion (AFF) module. When compared to other machine learning models, it performed far better in a test that used a real-time collegiate dataset, with a prediction accuracy of 94.6%. According to the findings, the model can identify warning signs of impending mental health crises and provide suggestions on how assistance would be provided immediately.
