Neural Network Device Segmented Forward Backward Data Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network devices face increased processing time and traffic congestion when attempting to execute computational and learning processing in parallel, as they are unable to efficiently manage forward and backward data propagation.
Innovation Solution
A neural network device comprising multiple cores and routers, where each core performs computation and processing of neural network components, and routers transmit data between cores to execute forward and backward data propagation efficiently, reducing traffic congestion by segregating data transmission within distinct regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If forward and backward data propagation are executed in parallel in the same neural network, then computational and learning processing can be performed simultaneously, but traffic congestion occurs and processing time increases
Solution Approach 1:
The patent divides the neural network into separate forward and backward regions, with dedicated routers for each direction. This segmentation allows forward and backward data propagation to occur in parallel without interfering with each other, thus enabling simultaneous computational and learning processing while avoiding traffic congestion that would occur in a shared network.
2Productivity
If forward and backward data propagation are executed in parallel in the same neural network, then computational and learning processing can be performed simultaneously, but traffic congestion occurs and cost increases
Solution Approach 1:
The neural network is segmented into forward and backward regions with dedicated routers for each. This segmentation enables parallel processing while managing complexity through structured separation rather than requiring complex routing logic in a single shared network.
Solution Approach 2:
Different regions of the neural network are assigned different functional qualities - the forward region is optimized for computational data propagation while the backward region is optimized for learning error propagation. This local specialization allows each region to be optimized for its specific purpose, improving overall efficiency while maintaining manageable complexity.
Data Source
AI summary
According to an embodiment, a neural network device includes: a plurality of cores each executing computation and processing of a partial component in a neural network; and a plurality of routers transmitting data output from each core to one of the plurality of cores such that computation and processing are executed according to structure of the neural network. Each of the plurality of cores outputs at least one of a forward data and a backward data propagated through the neural network in a forward direction and a backward direction, respectively. Each of the plurality of routers is included in one of a plurality of partial regions each being a forward region or a backward region. A router included in the forward region and a router included in the backward region transmit the forward data and the backward data to other routers in the same partial regions, respectively.