3031 papers
This paper proposes a secure, AI-driven, voice-based email system designed for visually impaired users that eliminates graphical interfaces by integrating FaceNet for biometric authentication, Whisper for high-accuracy speech recognition, and transformer models for email summarization and translation via Gmail API.
This study demonstrates that 3D Swin Transformers significantly outperform convolutional neural networks in cross-scanner generalization for hippocampal segmentation, achieving near-invariant performance across diverse multi-center datasets without fine-tuning and establishing a new standard for clinical deployment.
This paper demonstrates that while both opinion and structural similarity drive polarization in social networks, strategically combining these mechanisms in link-recommendation algorithms can prevent network fragmentation and foster moderate opinion coexistence.
This paper proposes an explainable multimodal hybrid deep learning framework that integrates smartphone and wearable sensor data to achieve high-accuracy early depression detection by leveraging advanced feature engineering and attention mechanisms to identify key behavioral indicators such as sleep irregularity, decreased mobility, social withdrawal, and increased nighttime smartphone use.
This paper presents a Karakalpak automatic speech recognition system based on the wav2vec 2.0 XLS-R architecture, which leverages a newly created 50-hour Karakalpak Speech Corpus to achieve a 21.10% word error rate and demonstrates the impact of labeled data volume on low-resource language recognition performance.
This paper proposes an AI-guided assembly framework that integrates real-time, context-aware instructions directly into product design to replace static manuals, demonstrating through a user study that this approach significantly reduces assembly time, errors, hesitation, and cognitive load compared to traditional manual guidance.
This study proposes a Time-Decay Gated Recurrent Unit (TD-GRU) model that confines temporal decay to the hidden state to preserve raw clinical values, demonstrating statistically credible net clinical benefit and superior performance over several baselines for early Sepsis-3 prediction in sparse, irregularly sampled ICU trajectories.
This paper proposes CCPP-LPC, a complete-coverage path-planning algorithm that utilizes a local path cost evaluation model and an adaptive dual-guided perturbation strategy to overcome the limitations of existing heuristic methods, thereby achieving superior computational efficiency and path optimization in complex environments.
This paper proposes a polarization-based anomaly detection method using the Minimum Effort to Consensus (MEC) index on social media data, demonstrating that monitoring shifts in public disagreement outperforms traditional volume-based approaches in identifying significant events, as validated through a case study of the 2024 Colombian pension reform debate on X.
This study proposes a novel hybrid deep learning model combining Variational Mode Decomposition (VMD), Temporal Convolutional Network (TCN), and Bidirectional Gated Recurrent Unit (BiGRU) that significantly outperforms nine other machine learning algorithms in predicting long-term monthly solar irradiation using NASA POWER data from Turkey.