Neural Architectures TRIZ Cases
- Neuron Circuit Biristor Voltage Divider Energy Reduction
Biristor neuron circuit reduces energy consumption by merging integration and comparison functions into a single component.
- Auditory Selection Using Dynamic Memory for Speech Separation
A neural network encodes speech signals into vectors using a dynamic memory unit to store speaker data.
- Neural Network Layer Compression for Image Analysis
A compression method identifies and retains only necessary artificial neural layers to form a student network.
- DNN Memory Bandwidth Optimization via Incremental Weight Updates
Incremental output updates reduce memory access frequency, lowering computation time and resource consumption for speech recognition systems.
- Event-Based Signal Saliency Detection Using CNN Classification
An event-based signal detection system converts input signals to frequency domain bins and monitors magnitude changes for saliency classification.
- Informative Looping Signals in Pulsed Neural Networks
A pulsed neural network identifies informative looping signals using spike-timing dependent plasticity inhibitory gating.
- Checksum-Based Fault Tolerant Neural Network Engine
A hardware neural network engine uses matrix checksums for fault detection and correction without triplicating processing blocks.
- Configurable CNN Processor Serial Link Architecture
Serial communication links between processing elements reduce redundant data movement and power consumption in real-time neural network applications.
- Compiler Converts Quasi-Affine Tensor Indexing to Matrix Operations
A compiler converts quasi-affine tensor indexing expressions into matrix multiplications and transposes for efficient execution.
- Deep Learning Stack for Image-Borne ID Detection
A deep learning stack classifies production images to detect sensitive identification documents within cloud security environments.
- Dilated Filter Generation for CNN Feature Data Processing
Sequential coefficient reading simplifies control wiring and reduces transfer volume during dilated convolution processing.
- Multi-Perspective Learned Descriptors for Content Recommendations
A neural network model generates low-dimensional learned descriptors from multiple content perspectives to reconstruct text sequences and identify semantic similarities.