Vehicle ML Model Simplification for Latency and Power Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models used in vehicle systems, such as autonomous driving and advanced driver assistance systems, face challenges in managing resource consumption, leading to high latency and power usage, which can impact performance and thermal management.
Innovation Solution
The system identifies and replaces resource-intensive non-linear operations with linear operations that approximate them, optimizing processing resources while maintaining accuracy thresholds through experimentation and simulation, using an objective function to select the most efficient replacements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If non-linear operations are used in machine learning models to maintain high accuracy, then model accuracy is improved, but processing resources and power consumption increase
Solution Approach 1:
The patent transforms non-linear operations into linear operations by changing the mathematical parameters and functional form of the operations. This parameter transformation maintains acceptable model accuracy while significantly reducing computational complexity and power consumption, directly resolving the contradiction between accuracy and energy use
2Measurement precision
If non-linear operations are used in machine learning models, then model accuracy is improved, but processing time and latency increase
Solution Approach 1:
The patent applies parameter changes by converting non-linear operations to linear operations, which have lower computational complexity. This transformation reduces processing time and latency while maintaining model accuracy within acceptable thresholds, directly addressing the time-accuracy tradeoff
3Productivity
If non-linear operations are replaced with linear operations, then processing resources are reduced, but model accuracy may deteriorate
Solution Approach 1:
The patent applies partial action by selectively replacing only certain non-linear operations with linear approximations, rather than all non-linear operations. This selective approach maintains model accuracy within acceptable thresholds while achieving significant resource reduction, balancing productivity and precision
4Productivity
If multiple linear approximations are tested to find the optimal replacement, then processing efficiency is improved, but the complexity of model optimization increases
Solution Approach 1:
The patent employs disposable test models that are created, evaluated, and discarded during the optimization process. Multiple candidate linear approximations are rapidly tested using simple evaluation metrics, with the optimal approximation selected based on efficiency-accuracy tradeoffs. This approach manages optimization complexity by using simple, replaceable test instances rather than complex persistent structures
Data Source
AI summary
A system can include a data processing system. The data processing system can include memory devices coupled with one or more processors. The data processing system can receive a model trained by machine learning comprising first operations, the model to generate an output to operate a vehicle. The data processing system can search the model to identify a non-linear operation of the first operations. The data processing system can select, from second operations, a second operation that approximates the non-linear operation, the selection based on a level of computing resources consumed by the second operations, an accuracy of the model generated with the second operations, and an accuracy threshold to operate the vehicle. The data processing system can replace the non-linear operation with the second operation in the model to produce a second output.