Neural Network Subset Cycling for Low-Latency Uncertainty Estimation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing neural network systems require significant computing resources and latency for uncertainty estimation, particularly in systems with limited resources, leading to inefficiencies and potential safety issues in applications like autonomous vehicles.

Innovation Solution

A method and system that processes sequential input data using a subset of neural networks to determine uncertainty values, reducing computing requirements and latency by cycling through neural networks to obtain predictive determinations and uncertainty values.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a plurality of neural networks are used to process sequential input data events to obtain predictive determinations, then the accuracy and reliability of the system is improved, but the computing resources and latency increase significantly

Engineering Contradiction:

Improveaccuracy of predictive determinationVSAvoidcomputing resources consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system segments the plurality of neural networks into different groups or subsets, where each subset processes a portion of the sequential input data events. This segmentation allows the system to distribute the computational load across multiple networks without requiring all networks to process every input event, thereby reducing overall computing resource consumption while maintaining reliable predictive determinations through ensemble processing.

Inventive Principle:

Principle #1Segmentation

2Reliability

Engineering Contradiction:

Improveaccuracy of predictive determinationVSAvoidprocessing latency

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by pre-processing or pre-computing certain aspects of the input data events before they reach the full ensemble of neural networks. This preliminary action reduces the complexity of the main processing stage, allowing the plurality of neural networks to produce accurate predictive determinations more quickly, thereby reducing latency while maintaining reliability.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If all neural networks process every sequential input data event, then comprehensive uncertainty estimation is achieved, but the computing power consumption increases

Engineering Contradiction:

Improveuncertainty estimation accuracyVSAvoidcomputing power consumption

Core Design Contradiction:

Measurement precisionVSLoss of energy

Solution Approach 1:

The system applies partial action by having only a subset of the neural networks process each sequential input data event, rather than requiring all networks to process every event. This partial processing is sufficient to achieve comprehensive uncertainty estimation through ensemble methods, while significantly reducing computing power consumption compared to having all networks process every event.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS12493787B2Systems having a plurality of neural networks

Publication Date: 2025.12.09 ARM LTD

AI summary

A method of operating a system having a plurality of neural networks includes receiving sequential input data events and processing each sequential input data event using a corresponding subset of the plurality of neural networks to obtain a plurality of sequential outputs. Each sequential output is indicative of a predictive determination of an aspect of the corresponding input data event. The method includes processing the plurality of sequential outputs to determine an uncertainty value associated with the plurality of sequential outputs, and operating the system based on the determined uncertainty value.