AI Base Caller Bus Network with Spatial-Temporal Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deploying deep convolution neural networks (CNNs) on portable and embedded systems is challenging due to large data volume, intensive computation, varying algorithm structures, and frequent memory accesses, which affects efficiency and performance, particularly in convolution acceleration schemes.
Innovation Solution
A specialized architecture for neural network-based base calling that uses segregated spatial and temporal convolution layers to process sequencing images, combined with a compression logic to store and reuse intermediate results across sliding windows, reducing redundant computations and resource waste.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep convolution neural networks are deployed on portable and embedded systems, then base calling accuracy is improved, but computational resource consumption increases significantly
Solution Approach 1:
The patent segments the convolutional neural network into multiple processing stages with selective feature map retention. Only certain feature maps are preserved in memory while others are discarded, dividing the computational workload to reduce resource consumption while maintaining accuracy.
Solution Approach 2:
The patent extracts and retains only the most important intermediate feature maps from the convolutional processing pipeline. By identifying and keeping only critical feature representations, the system reduces memory usage and computational overhead while preserving the essential information needed for accurate base calling.
2Productivity
If intermediate results are stored and reused across sliding windows, then redundant computations are reduced, but memory usage increases
Solution Approach 1:
The patent applies local quality by retaining feature maps with different preservation strategies at different locations in the processing pipeline. Certain spatial regions and feature map types are retained with higher priority, while others are discarded, creating a non-uniform retention policy that optimizes the balance between memory usage and compute efficiency.
Solution Approach 2:
The patent implements a selective discarding strategy where intermediate feature maps are discarded after their useful lifetime in the processing pipeline. By recovering and retaining only those feature maps that provide the most value across sliding windows, the system minimizes memory usage while maintaining compute efficiency.
3Loss of energy
If compression logic is implemented to store intermediate results, then resource waste is reduced, but device complexity increases
Solution Approach 1:
The patent implements preliminary action by pre-identifying which intermediate feature maps should be retained before processing begins. By determining the retention policy in advance based on the specific convolutional layer and feature map importance, the system avoids complex runtime decisions while reducing resource waste.
Solution Approach 2:
The patent uses copying by creating selective duplicates of important intermediate feature maps for reuse across sliding windows. Instead of full compression logic, the system makes targeted copies of critical feature representations, reducing resource waste with minimal added complexity.
Data Source
AI summary
The technology disclosed relates to a system that comprises a spatial convolution network and a temporal convolution network. The spatial convolution network is configured to process a window of per-cycle sequencing image sets and generate respective per-cycle spatial feature map sets. Trained coefficients of spatial convolution filters in spatial convolution filter banks of respective sequences of spatial convolution filter banks vary between sequences of spatial convolution layers in respective sequences of spatial convolution layers. The temporal convolution network is configured to process the per-cycle spatial feature map sets on a groupwise basis and generate respective per-group temporal feature map sets. Trained coefficients of temporal convolution filters in respective temporal convolution filter banks vary between temporal convolution filter banks in respective temporal convolution filter banks.