Signal Classification Feedback Correction for Confident Wrong Predictions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Machine learning systems, particularly neural networks, often produce incorrect predictions while being confident, leading to inaccurate outputs in classification and regression tasks, especially in robotic applications where sensor data is processed.

Innovation Solution

A computer-implemented method that determines a feature representation of an input signal, predicts the deviation of an intermediate signal from a desired output, and iteratively adapts this signal to improve accuracy, using multiple neural networks to correct predictions before outputting the final signal.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning systems (neural networks) are used for classification and regression analysis, then prediction performance is improved, but wrong predictions are produced with high confidence

Engineering Contradiction:
Improveprediction accuracyVSAvoidprediction reliability
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where the predicted output is fed back into the system through the same neural network to generate a corrected output. This self-feedback loop allows the system to identify and correct confident but wrong predictions by comparing the original prediction with the corrected prediction, thereby improving both accuracy and reliability simultaneously

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent introduces an intermediate correction mechanism that acts as a mediator between the original neural network prediction and the final output. This intermediary correction step processes the original prediction through the neural network again, generating a corrected output that filters out confident wrong predictions while preserving accurate ones

Inventive Principle:
Principle #24Intermediary (Mediator)

2Speed

If standard neural networks are used, then processing speed is maintained, but incorrect predictions are output with high confidence

Engineering Contradiction:
Improveprediction speedVSAvoidprediction accuracy
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by performing a correction pass through the neural network before finalizing the output. Instead of waiting for wrong predictions to be identified after output, the system proactively corrects predictions by feeding them back through the network, ensuring accuracy is improved without significant speed penalty

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4113380A1Device and method for classifying a signal and/or for performing regression analysis on a signal
Publication Date: 2023.01.04 ROBERT BOSCH GMBH
  • EP4113380A1 patent drawingFigure 1
  • EP4113380A1 patent drawingFigure 2
  • EP4113380A1 patent drawingFigure 3

AI summary

Computer-implemented method for determining an output signal characterizing a classification and/or a regression result of an input signal comprising the steps of: a. Determining a feature representation characterizing the input signal; b. Determining an intermediate signal characterizing a classification and/or regression result of the feature representation; c. Predicting, based on the feature representation and the intermediate signal, a deviation of the intermediate signal from a desired output signal of the input signal; d. Adapting the intermediate signal according to the determined deviation thereby determining an adapted signal; e. Providing the adapted signal as output signal.