Sobol Bit-Stream Processing for Faster Stochastic Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deterministic and stochastic computing methods for stochastic systems are inefficient due to slow convergence and limited scalability, especially in applications that can tolerate some inaccuracy such as image processing and neural networks, where the number of cycles and cost of generating bit-streams increase significantly with increasing inputs.
Innovation Solution
The use of low-discrepancy Sobol sequences to generate bit-streams for stochastic computational systems, which allows for faster convergence and reduced latency by down-sampling bit-streams, thereby improving processing time and energy consumption, and is more tolerant to errors such as bit flips.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional deterministic or stochastic computing methods are used to generate bit-streams, then the system can perform stochastic computations, but the convergence is slow and processing time is long
Solution Approach 1:
The patent changes the fundamental parameter of bit-stream generation from conventional random or deterministic sequences to Sobol low-discrepancy sequences. This parameter change in the sequence generation method enables faster convergence while maintaining computational accuracy, directly resolving the contradiction between computation accuracy and processing time
Solution Approach 2:
The patent substitutes conventional random number generation mechanisms with Sobol sequence generation mechanisms. This substitution replaces the traditional approach with a mathematically optimized sequence that provides better distribution properties, achieving faster convergence without sacrificing accuracy
2Productivity
If the number of inputs increases in stochastic computing systems, then the system can process more data, but the number of cycles and cost of generating bit-streams increase significantly
Solution Approach 1:
The patent implements a universal Sobol sequence generation mechanism that can handle multiple inputs efficiently. The same sequence generation apparatus serves multiple computational purposes, reducing the marginal cost of processing additional inputs and improving scalability
Solution Approach 2:
The patent segments the bit-stream generation process into independent Sobol sequence sources that can be efficiently managed and scaled. This segmentation allows the system to handle increased data processing demands without proportionally increasing system complexity
3Ease of operation
If conventional bit-stream generation methods are used, then the system can operate, but the hardware footprint is large and energy consumption is high
Solution Approach 1:
The patent merges the bit-stream generation function into the existing computational fabric using Sobol sequence sources that can be integrated with minimal additional hardware. This merging approach reduces the overall hardware footprint while maintaining full operability of the stochastic computing system
Data Source
AI summary
Example devices are described that include a computational unit configured to process first set of data bits encoding a first numerical value and a second set of data bits encoding a second numerical value. The computational unit includes a bit-stream generator configured to generate bit combinations representing first and second bit sequences that encode the first and second numerical values, respectively, based on a proportion of the data bits in the sequence that are high relative to the total data bits. The first bit sequence is generated using a first Sobol sequence source, and the second bit sequence is generated using a second Sobol sequence source different from the first Sobol sequence source. The device also includes computation logic configured to perform a computational operation on the bit combinations and produce an output bit-stream having a set of data bits indicating a result of the computational operation.