Neural Learning Methods TRIZ Cases
- Multitask Neural Network for Material Property Prediction with Scarce Data
A multitask neural network shares a common encoder across individual decoders to predict multiple material properties simultaneously.
- Knowledge Graph Entity Generation via Causality Simulation
Machine learning generates new knowledge graph entities by embedding them in a latent space relative to existing nodes.
- LLM Conformance Assistant for Process Mining Non-Conformance
An AI assistant analyzes event logs against process models to generate human-readable explanations for deviations, reducing manual analysis time.
- Neural Network Attention Map Noise Suppression via Nonlinear Transformation
Nonlinear transformation removes noise components before attention mechanism application, improving accuracy while reducing computational complexity.
- Functional Activation Analysis for Deep Neural Network Explainability
A functional activation-based analysis system records layer output time-series to generate neural network activation data indicating degree of activation.
- Edge-Client Federated Graph Learning with Adaptive Neighbor Generation
A graph imputation generator repairs missing links across subgraphs, resolving insufficient feature propagation while maintaining client privacy.
- Neural Network Filter Redundancy Detection for Model Compression
Multi-criteria filter redundancy detection identifies and removes redundant filters from neural network layers to reduce computational complexity.
- Neural Network MAC Scheduling for Variable Terminal Counts
A neural network model determines scheduling weights for terminal devices across varying user counts without retraining.
- Switch Layer Neural Network Routing for Computational Capacity
A switch neural network layer selects a single expert from multiple networks to process inputs.
- Runway Configuration Prediction Using 3D Tensor CNNs
A deep learning framework processes three-dimensional tensor data to predict hourly runway configurations and airport acceptance rates for multi-airport systems.
- Neural Architecture Search for In-Vehicle Depth Estimation
Neural architecture search and knowledge distillation generate compact neural networks that satisfy latency specifications for in-vehicle object detection.
- Hyperuniform Neural Network Sparse Topology
Hyperuniform neural networks enforce sparse connectivity constraints to suppress density fluctuations and lower loss error.
- Neural Network Design Space Reduction via Dataset Characteristics
A design space reduction apparatus narrows neural network architecture search using dataset characteristics.
- Synthetic Dataset Generation via Inverse Copula and Marginal CDF Networks
Inverse copula and marginal CDF networks transform independent noise into correlated synthetic datasets that accurately capture dependence structures.
- Functional Interpolation for Relative Position Biases in Attention Networks
Functional interpolation computes relative position biases to generalize attention networks across longer sequences without quadratic computational costs.