Neural Network Output Layer Softmax Reconfigurable Fabric
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional processors and analysis techniques are overwhelmed by the vast amounts of big data, making data handling and analysis economically infeasible, especially in applications like machine learning, where scalable and efficient data processing techniques are necessary.
Innovation Solution
A neural network output layer implemented within a reconfigurable fabric, which configures processing elements to perform data flow graphs, using techniques like Softmax and parabolic estimator functions with fixed-point calculations, to efficiently map and classify data, thereby enabling effective machine learning operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional processors are used to analyze big data, then data analysis can be performed, but the processing becomes economically infeasible due to overwhelming computational requirements
Solution Approach 1:
The patent replaces traditional general-purpose processor architectures with a specialized data flow graph computation system. This substitution transforms the mechanical processing approach from sequential von Neumann architecture to a parallel data flow model where computations are represented as graphs with nodes (operations) and edges (data flow), enabling more efficient exploitation of data parallelism and reducing computational overhead
Solution Approach 2:
The patent segments the data analysis task into discrete computational nodes within a data flow graph. Each node represents a specific operation (e.g., matrix multiplication, activation function) that can be independently executed and optimized. This segmentation allows for parallel processing of multiple nodes simultaneously, dramatically improving productivity while distributing computational load across the system
2Productivity
If more computational resources are allocated to handle big data, then data analysis performance improves, but economic feasibility deteriorates
Solution Approach 1:
The patent implements dynamic resource allocation through its data flow graph execution model. The system dynamically schedules and executes computational nodes based on data availability, dependencies, and resource status. This dynamic approach allows the system to optimize resource utilization in real-time, achieving high throughput without requiring excessive static computational resources that would consume unnecessary energy
Solution Approach 2:
The data flow graph architecture enables continuous processing by maintaining persistent data flow through the computational graph. Once the graph is constructed, the system can continuously feed data through the pipeline of operations without repeated setup overhead, achieving sustained high throughput with efficient energy utilization compared to batch processing approaches
Data Source
AI summary
Techniques for a neural network output layer for machine learning are disclosed. A plurality of processing elements within a reconfigurable fabric is configured to implement a data flow graph, where the data flow graph implements a neural network. The data flow graph can include machine learning or deep learning. A layer is implemented, within the neural network, that maps a first vector of real values to a second vector of real values bounded by zero and one, where the second vector sums to a value of one using fixed-point calculations. The layer can include a final layer within the neural network. The layer that maps the first vector includes a Softmax function. Results of the neural network are classified based on a value of the second vector. The classifying can include part of a machine learning or a deep learning process.