Neural Learning Methods TRIZ Cases
- Neural Network Reduction for Data-Dependent Compiler Bug Isolation
Reference and user compilers partition neural networks and simplify tensors to reproduce data-dependent bugs with less sensitive data.
- FMCW Radar Target Classification with Multifeature AI at Ultra-Short Range
Range, phase, IQ, and spectrogram features feed an AI model to distinguish people, objects, types, and motion at short range.
- Neural Network Depth Compression Using Dynamic Programming
Dynamic programming removes low-value nonlinear layers to accelerate inference.
- Hybrid Variational Neural Network for Reactive Metric Prediction
A linear network and variational auto-encoder jointly predict metrics, reducing lag and error during macroeconomic change.
- Likelihood-Based Dynamic Graph Prediction Using GNNs And MLE
GNN embeddings, feedforward prediction, and likelihood modeling estimate correlated changes and rank future graph states.
- Neural Network Operator Pruning for Efficient Structure Search
Candidate operators are ranked and removed before weight adjustment, narrowing model search while preserving accuracy.
- Spaceborne SAR and Neural Networks for Airport Pavement Antiskid Evaluation
A feedforward neural network links SAR echo intensity to friction coefficients for continuous, large-area airport pavement monitoring.
- Contact Trajectory Signaling for Diversion Vehicle User Targeting
Machine learning predicts diversion-vehicle engagement for targeted signaling.
- Relative Margin Contrastive Learning for Vanishing Gradient Control
A relative margin in contrastive learning preserves gradients for high-separation pairs and improves encoder training with noisy data.
- Synthetic Training Data for Neural Networks Predicting High-Pollution Events
Synthetic pollutant series help neural networks predict rare high-pollution events.
- Generative AI Bias Evaluation Using Masked Statement Reconstruction
A paired-model process masks key terms, reconstructs statements, and compares outputs to evaluate and mitigate generative AI bias.