Neural Architectures TRIZ Cases
- Synthetic Image Generation for Neural Network Training
A synthetic image generation system creates training datasets for convolutional neural networks.
- Neural Network Text Classification Trend Regression
A second target neural network processes sample text data alongside regression results representing classification trends to generate predicted classification outcomes.
- Deep Neural Network Skip Connections for Dynamic Layer Selection
Skip connections enable selective layer execution in deep neural networks to reduce memory usage and processing latency.
- Neural Network Quantization Range Adjustment via Saturation Ratio
A neural network quantization method adjusts tensor ranges by observing saturation ratios to maintain target levels.
- Neural Network Algorithm Replacement for Memory Reduction
Real-time algorithm replacement lowers memory consumption while maintaining processing speed in artificial neural networks.
- Neural Network Convolution Kernel Grouping and Scaling
Grouping similar convolution kernels into a base kernel with scaling factors reduces computational complexity while maintaining feature extraction accuracy.
- Neural Network Learning Device Adaptive Noise Injection
Adaptive noise injection suppresses overfitting by adjusting noise frequency and size according to inference result contributions.
- Generative Adversarial Networks for Conformance Score Prediction
Attention-weighted generative adversarial networks determine conformance scores without prior models, resolving complexity trade-offs in event trace analysis.