Vehicle Data Compression on Manifolds for Robust Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current systems for classifying objects and situations in vehicles using physical measurement data require extensive labeled training data, which is scarce and expensive, and struggle with noise and dimensionality issues in high-dimensional data spaces.

Innovation Solution

A lossy data compressor using a parametrized mapping network that reduces data dimensionality to a Riemannian manifold, preserving semantic similarity and allowing for unsupervised training with unlabeled data, thereby reducing the need for labeled data and enhancing data quality for classification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If supervised learning is used for classification with high-dimensional physical measurement data, then classification accuracy can be achieved, but extensive labeled training data is required which is scarce and expensive

Engineering Contradiction:
Improveclassification accuracyVSAvoidamount of labeled training data
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies preliminary action by performing unsupervised pre-training of the neural network on unlabeled physical measurement data before supervised fine-tuning. This pre-training phase initializes the network weights and features using the manifold compressor without requiring labeled data, thereby reducing the subsequent need for extensive labeled training data while maintaining classification accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces a manifold compressor as an intermediary component that transforms high-dimensional physical measurement data into compressed representations on a Riemannian manifold. This intermediary structure facilitates more efficient learning by reducing data dimensionality and extracting meaningful features, allowing the classification system to achieve accurate results with fewer labeled examples.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Loss of information

If high-dimensional physical measurement data is used directly for classification, then complete information is available, but noise and dimensionality issues reduce training robustness

Engineering Contradiction:
Improveinformation completenessVSAvoidtraining robustness
Core Design Contradiction:
Loss of informationVSReliability

Solution Approach 1:

The patent extracts essential features from high-dimensional physical measurement data by mapping them onto a lower-dimensional Riemannian manifold using the manifold compressor. This extraction process separates meaningful information from noise by identifying and retaining only the most relevant dimensions, thereby improving training robustness while preserving critical information for classification.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter space by transforming data from the original high-dimensional space into a constrained Riemannian manifold with specific geometric properties. This parameter transformation reduces dimensionality and imposes structural constraints that eliminate noisy directions, thereby enhancing training robustness while maintaining information completeness through the manifold's intrinsic geometry.

Inventive Principle:
Principle #35Parameter changes

3Productivity

If data dimensionality is reduced for efficient processing, then computational efficiency improves, but information loss may occur

Engineering Contradiction:
Improvecomputational efficiencyVSAvoiddata information loss
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The patent employs curvature by utilizing a Riemannian manifold with specific geometric properties to embed the compressed data. The manifold's curved structure preserves topological relationships and semantic similarities among data points during dimensionality reduction, ensuring that computational efficiency is improved while minimal information is lost due to the geometry-aware compression that maintains data structure integrity.

Inventive Principle:
Principle #14Spheroidality (Curvature)

Solution Approach 2:

The patent transitions data from high-dimensional Euclidean space to a lower-dimensional Riemannian manifold, effectively changing the dimensional framework. This dimensionality change achieves computational efficiency by reducing the number of parameters while preserving essential information through the manifold's intrinsic geometry, which encodes relationships among data points in a compact form.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP3575986B1A lossy data compressor for vehicle control systems
Publication Date: 2024.07.10 ROBERT BOSCH GMBH
  • EP3575986B1 patent drawingFigure 1
  • EP3575986B1 patent drawingFigure 2
  • EP3575986B1 patent drawingFigure 3a)~3d)

AI summary

A lossy data compressor (1) for physical measurement data (3), comprising a parametrized mapping network (2) that, when applied to a measurement data point x in a space X, produces a point z in a lower-dimensional manifold Z, and configured to provide a point z on Z as output in response to receiving a data point x as input, wherein the manifold Z is a continuous hypersurface that only admits fully continuous paths between any two points on the hypersurface; and the parameters θ of the mapping network (2) are trainable or trained towards an objective that comprises minimizing, on the manifold Z, a distance between a given prior distribution PZ and a distribution PQ induced on manifold Z by mapping a given set PD of physical measurement data (3) from X onto Z using the mapping network (2), according to a given distance measure.