Kernel Methods TRIZ Cases
- Batch Normalization Fusion and Quantization for Neural Network Inference
Merging batch normalization parameters with convolutional layers enables in-place computation using 8-bit integer operations.
- Far-Field Acoustic Model Training Using Diffuse Reverberation
Simulating diffuse room reverberation without early reflections trains acoustic models for far-field speech recognition.
- Value Realization Analytics System for Dynamic KPI Forecasting
A value realization analytics system dynamically forecasts Key Performance Indicator trajectories using trained data models.
- On-Vehicle Network Detection Device Using Predictive Verification
A detection device monitors on-vehicle network data to predict expected communication states for security verification.
- Semi-supervised User Classification via Embedding Space Segmentation
A semi-supervised learning system maps user data into an embedding space to predict category membership propensities.
- Machine Learning Model for Predicting Customer Interest
A machine learning model predicts customer interests using location and birthdate data to generate personalized recommendations.
- Wearable Camera Barcode Detection for Contactless Transactions
A wearable multimedia device captures bar codes from a scene using a wide field-of-view camera to establish communication channels.
- Multi-Stage Knowledge Transfer via Intermediate Models
Intermediate models transfer knowledge from complex input to simple output models, maintaining accuracy while reducing computational resource consumption.
- Multi-stage Deep Neural Network for Non-ideal Image Recognition
A multi-stage deep neural network framework processes categorized object data through sequential detection and subcategorization stages.
- Hybrid ADMM-IPM Algorithm for SVM Feature Selection
A hybrid algorithm combining alternating direction and interior-point methods solves elastic-net support vector machines.
- Semi-Supervised Data Quality System Using Clustering and Validation
A semi-supervised system purifies datasets by assigning class labels to unlabeled data points through iterative clustering analysis.
- Adaptation Controller for Cloud Code Performance Optimization
An adaptation controller analyzes execution profiles to identify performance-critical code sections and applies modifications.
- Video Seek System Session Information Extraction
A seek content extraction system identifies session information in video frames to display alongside thumbnails.
- Machine Learning Data Segmentation for Product Design Verification
Learning device classifies product data into predetermined categories to distribute training and verification sets proportionally.
- Autonomous Inkjet Printing Optimization via Active Machine Learning
Active machine learning with model selection reduces calibration experiments by 80% while maintaining high precision for new ink-print head pairings.
- Recursive Feature Elimination Using SVM Feedback Loops
Dependent SVM rounds use weight feedback to eliminate features, improving ranking accuracy and computational efficiency.
- Self-Supervised Generative Model for Aerial Image Anomaly Detection
Generative model reconstructs aerial images to detect anomalies in new assets, eliminating the need for historical reference data or predefined anomaly lists.
- Fusion-Based Classifier for Industrial Defect Detection
A fusion-based classifier selects extremum values from probability vectors to determine input data classes.
- SVM Anomaly Detection for CAN Bus Intrusion
An ECU classifier detects anomalies in Controller Area Network traffic using a Radial Basis Function kernel.
- Physiological Signal Identification Using Machine Learning Models
A system extracts features from physiological signals using machine learning models to generate identification data.
- Distributed Gradient Reduction Algorithms for Mesh Topologies
Distributed reduction algorithms process gradient vectors across mesh topologies without central coordination.
- Influence Score Normalization for Interpretable ML Training Data
A normalizing function processes influence indicators to identify relevant training data points for machine learning predictions.
- Virtual Resource Automatic Selection via Supervised Learning
A virtual resource automatic selection system classifies network requests into clusters using supervised learning to allocate capacity.
- Random Time Warping for Time-Series Feature Learning
Random warping series generate reference signals to compute distance-based feature matrices, reducing computational complexity from O(N^2L^2) to O(NRLD).
- Time2Vec Embedding Layer for Machine Learning Temporal Data
Time2Vec decomposes time into a learnable vector representation, resolving exploding gradients and eliminating hand-crafted feature engineering.
- Biosignal Analysis Framework Using Segmented Waveform and DSP Models
A biosignal analysis framework selects segmented waveform and digital signal processing models to extract comprehensive features from input signals.
- Federated Learning Client Parameter Verification
Clients verify server parameters against local values to prevent Byzantine attacks from corrupting training results and reduce server detection load.
- SVM Flight Delay Prediction Model Using Prior Knowledge
A support-vector-machine prediction model incorporates historical flight data as prior knowledge to construct time-specific delay forecasts.
- Attention Matrix Rank Reduction for Neural Translation Memory
Rank reduction decomposes attention matrices to handle long sequences without exceeding memory limits, preserving translation accuracy.
- Diagnosis Support Apparatus Dynamic Weight Integration
A diagnosis support apparatus integrates inference results using dynamic weights to maintain accuracy across varying data conditions.
- Saliency Visualization via Regional Segmentation and Gain Iteration
Regional segmentation generates attribution scores to produce saliency masks, avoiding slow perturbation queries that degrade computational efficiency.
- Automating Digital Forensic Evidence Collection in Cloud Environments
Convolutional neural networks classify cloud images to automate evidence collection, reducing manual investigation time.
- Vulnerability Prioritization via Machine Learning Exploitability
A vulnerability management system uses a machine learning model to determine exploitability levels from internal and external data.