Character And Pattern Recognition TRIZ Cases
- Autoencoder Signal Feature Detection Without Prior Knowledge
An autoencoder extracts and clusters characteristic signal features from measurement data without prior knowledge, making rare events easier to spot.
- Pruned CNN Intrusion Detection for Faster Edge Security Inference
Filter pruning removes low-weight CNN filters to cut edge security inference cost while preserving intrusion classification accuracy.
- CNN Filter Pruning by Weight Sums for Lower Inference Cost
Removing low-weight CNN filters cuts matrix multiplications and inference cost, then retraining restores accuracy for constrained hardware.
- Deep Learning ADC Correction for Noise and Channel Mismatch
A trained deep network suppresses ADC noise and corrects channel mismatch and nonlinear distortion to improve wideband signal accuracy.
- Distributed Image Coding With Side-Information Reconstruction
Shifting complexity from encoder to decoder enables efficient image compression with side-information reconstruction for low-power imaging platforms.