Biological Neural Network Models TRIZ Cases
- DNN Speech Recognition Using Heterogeneous Wideband Narrowband Training Data
A deep neural network processes wideband and narrowband audio inputs using adaptive feature vectors for automatic speech recognition.
- Spatial Encoding for Structured Data Extraction
Hierarchical spatial tokens resolve OCR noise and context prediction errors to achieve 95% accuracy in form data extraction.
- Causal Explanation Platform for Neural Network Interpretability
A causal explanation platform learns relationships between neural network inputs and outputs using attention mechanisms to generate structured interpretations.
- Liveness Detection Using Photodiode Depth Data
A liveness test apparatus fuses color images with photodiode depth data to distinguish genuine faces from spoofing attempts.
- Neural Network Obfuscation via Dummy Nodes and Layers
Introduces dummy nodes and layers to mask real parameters, preventing reverse engineering while maintaining computational efficiency.
- Dynamic Target Speaker Update in Voice Recognition
An electronic device updates the target speaker using an AI model to extract voice signals from audio.
- Loitering Detection Server Filtering Video Noise
Digital video analytics server extracts loitering event snapshots from continuous streams for targeted processing.
- Neural Network Interpolation for Digital Holographic Portraits
A neural network interpolates images from multiple cameras to record digital holographic portraits with full parallax.
- Video Analysis Tool for Automated Presentation Feedback
A video analysis software application tool records presentations and analyzes visual features using a neural network model to generate automated feedback.
- Model-Based Metrology Error Estimation for Semiconductor Structures
A control system estimates upper bounds on measurement errors for semiconductor parameters using model-based processing.
- Graph Neural Network for Data Recommendation
A graph neural network framework generates user and attribute embeddings from digital graphs to overcome data sparsity in large repositories.
- Neural Network Optical Waveguide Characterization
A receiving device uses a trained neural network to extract mode information from intermixed light beams in multi-mode optical waveguides.
- Arithmetic Processing Device Parallel Comparison Update Architecture
Parallel comparison update parts process information sets across multiple stages to reduce processing time in arithmetic operation devices.