Kernel Methods TRIZ Cases
- Multi-Domain Tractability Scoring for Protein–Compound Predictions
Tractability scores filter and weight compound–protein predictions, improving accuracy and reducing computational waste in bioactivity programs.
- Handed Activity Recognition Using Mirror-Flipped Training Data
Left-right activity classes can double model complexity; horizontal input flipping and balanced training support recognition of both handednesses.
- Thermomesh Sensor for Heat Source Localization and Peak Temperature
Interwoven thermocouple wires map heat sources and estimate peak temperature.
- Machine Learning Normalization Pipelining for Lower Memory Bandwidth
This case pipelines tensor segments and combines normalization stages to reduce memory accesses, bandwidth use, and processing latency.
- Neural Network Classifier Verification via Interface Region Analysis
Support vectors and interface-region percentages help verify classifiers when overlapping data categories complicate certification.
- Handed Activity Recognition With Horizontal Flipping for Dataset Balance
A single handed-activity class uses horizontal flipping to distinguish left- and right-handed actions while reducing training complexity.
- Sub-Kernel Searching Module for Adaptive Edge Neural Inference
Adaptive sub-kernel selection balances edge inference accuracy and compute.
- Local Hidden-Stressor Detection From Media and Health Signals
A private on-device model combines text, images, audio, and health metrics to identify stressors and notify users at set thresholds.
- Shapelet Learning for Time Series AUC-Optimized Anomaly Detection
Learn how feature waveforms and AUC optimization improve time-series anomaly detection.
- Neural Network Compute Units Using Indexed Non-Zero Activations
Indexed non-zero activations help compute units skip zero multiplications and save energy.
- Dynamic Network Topology Updates for Distributed Machine Learning
An edge server adjusts topology and consensus weights to limit stragglers, reduce latency, and speed distributed model training.
- Multidimensional Gesture-Voice Interaction with Dynamic Weighting
A deconvolutional neural network, unscented Kalman filter, and SVM combine gesture and voice data for adaptive feedback.
- Multi-Source Content Recording with Adaptive Tuner Scheduling
This case combines preference-based recording, tuner-aware scheduling, and streaming checks to avoid duplicates and storage waste.
- ECG Biometric Verification with Machine-Learning Noise Filtering
Baseline removal, wavelet transforms, and classifiers separate true ECG features from noise during biometric authentication.
- Capacity-Based Enterprise Architecture Generation Using Machine Learning
Machine learning adapts enterprise architectures to changing network demands.