Perception Result Display Using Context Instead of Confidence Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Non-technical human users struggle to interpret confidence values associated with perception results from machine learning models, such as facial recognition, which are essential for determining the reliability of the information provided.
Innovation Solution
A method is implemented to selectively communicate perception results to users by suppressing results with low confidence levels and providing contextual information, like elapsed time since the result was obtained, instead of the confidence value, to enhance user understanding of the reliability of the results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If confidence values are provided to users, then the reliability information is complete and accurate, but non-technical users cannot easily interpret the confidence values to assess result reliability
Solution Approach 1:
The patent introduces contextual information (time elapsed, environmental conditions, sensor quality indicators) as an intermediary between the raw confidence value and the user's understanding. Instead of presenting the abstract confidence percentage directly, the system translates it into concrete contextual factors that users can naturally comprehend and use to assess reliability.
Solution Approach 2:
The patent transforms the confidence value parameter into different representations - specifically converting the abstract probability percentage into tangible contextual parameters like time elapsed since capture, environmental conditions, and sensor performance metrics. This parameter transformation makes the reliability information accessible to non-technical users while preserving the underlying confidence assessment.
2Loss of information
If all perception results are communicated to users, then complete information is provided, but low confidence results may mislead users and reduce decision quality
Solution Approach 1:
The patent applies partial action by selectively communicating only those perception results that meet a minimum confidence threshold. Rather than presenting all results equally, the system filters out low-confidence results that would be more harmful than helpful, while still providing sufficient information for users to make informed decisions about the presented results.
Solution Approach 2:
The patent incorporates confidence level feedback into the communication process. Each presented result includes contextual information about its reliability, enabling users to understand the quality of each piece of information. This feedback mechanism allows users to weigh results appropriately without being overwhelmed by uncertain data.
3Measurement precision
If confidence levels are communicated directly, then statistical precision is maintained, but users cannot intuitively understand what the confidence level means for result reliability
Solution Approach 1:
The patent uses contextual information as a mediator that bridges the gap between precise statistical confidence measurements and user comprehension. By presenting contextual factors (such as how recent the data is, environmental conditions, sensor quality) rather than raw confidence percentages, the system maintains measurement precision internally while improving user understanding externally.
Data Source
AI summary
According to a first aspect, there is provided a computer-implemented method of controlling a user interface to selectively communicate perception results to a user, the method comprising: in response to an update instruction, using a current confidence level of each perception result of a set of perception results to determine whether or not to communicate that perception result at the user interface. The perception results are determined by processing sensor signals from a sensor system using at least one perception algorithm. At least one of the perception results is communicated at the user interface together with at least one piece of contextual information, without communicating the current confidence level that caused the perception result to be outputted, the current confidence level having been at least partially derived from the piece of contextual information.