Fixed-Point ML Computing Architecture for Faster Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
General purpose processors are inefficient in processing large loads, leading to delays in machine learning tasks such as training artificial intelligence models, due to their limited operating speed.
Innovation Solution
A computation device and method that includes a controller unit, operation unit, and conversion unit, utilizing a primary processing circuit and secondary processing circuits to perform machine learning computations by converting input data into fixed-point data and executing operations to improve processing speed and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If general purpose processors are used to process large loads, then they can handle various types of computations, but their operating speed is limited and they cause delays
Solution Approach 1:
The processor is divided into a control unit and multiple processing units (first processing unit, second processing unit, third processing unit). Each processing unit handles specific types of operations in parallel, enabling the system to maintain versatility while significantly increasing operating speed through concurrent execution of multiple instruction streams
Solution Approach 2:
The patent transitions from single-stream sequential processing to multi-stream parallel processing by introducing multiple independent processing units that can execute different instruction streams simultaneously. This dimensional expansion from 1D to 3D processing space allows the system to handle diverse computational tasks at accelerated speeds
2Adaptability or versatility
If general purpose processors are used for machine learning training operations, then they can perform the computations, but the computation amount is large and it takes extraordinarily long time
Solution Approach 1:
Machine learning training operations are segmented across multiple specialized processing units, each handling specific computational tasks (e.g., matrix multiplications, activation functions, gradient calculations). This segmentation enables parallel execution of training operations, reducing overall training time while maintaining the ability to handle complex machine learning workloads
Solution Approach 2:
The control unit pre-processes and prepares computation instructions by parsing them into multiple parallel instruction streams before execution. This preliminary action organizes the large computation amount into manageable parallel tasks, enabling the processing units to execute training operations more efficiently and reduce training time
Data Source
AI summary
The present disclosure provides a computation device. The computation device is configured to perform a machine learning computation, and includes an operation unit, a controller unit, and a conversion unit. The storage unit is configured to obtain input data and a computation instruction. The controller unit is configured to extract and parse the computation instruction from the storage unit to obtain one or more operation instructions, and to send the one or more operation instructions and the input data to the operation unit. The operation unit is configured to perform operations on the input data according to one or more operation instructions to obtain a computation result of the computation instruction. In the examples of the present disclosure, the input data involved in machine learning computations is represented by fixed-point data, thereby improving the processing speed and efficiency of training operations.