Neural Learning Methods TRIZ Cases
- Neural Network Pooling via Class Probability Weighting
A neural network model applies class probability weighting during pooling to preserve feature map information.
- Logistic Output Neurons for High Capacity Classification
Binary logistic coding reduces output neuron count to resolve device complexity while maintaining classification accuracy.
- Prompt Suitability Analysis for LLM Security
A prompt analysis system evaluates user input suitability using verification scores to filter malicious requests before processing.
- Synthetic Heterogeneous Time-Series Data Generation via GANs
Generative adversarial networks produce realistic synthetic records from heterogeneous time-series data.
- Dynamic Exponent Bias for Neural Network Training
A dynamic exponent bias method converts tensor values to lower precision formats for neural network training.
- Dynamic Computational Element Configuration for GAN Training
A controller functional block dynamically configures computational elements to balance execution times between generative and discriminative networks.
- Style-Based Adaptive Pre-Training for Korean Language Models
A three-step training method adapts language models to distinct writing styles using unlabeled datasets.
- GAN-Based Signal Generator for Indistinguishable Communications
A generator machine-learning network produces communications signals indistinguishable from target data through iterative discriminator feedback.