Computational Storage for Low-Bandwidth Edge Data Triaging

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional systems relying on machine learning algorithms in edge devices face challenges with limited or no connectivity, including difficulty in streaming raw data due to bandwidth and privacy concerns, accumulating new training data, and performing machine learning operations efficiently, which limits the integration of AI/ML technologies in resource-constrained environments.

Innovation Solution

Implementing active triaging and labeling of raw data at the edge using computational storage systems, integrating these labels into the MLOps loop, and utilizing a neural network supervisor to manage and update models autonomously, reducing the need for human curation and centralization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If machine learning algorithms are deployed in edge devices with limited connectivity, then data processing can be performed locally, but bandwidth consumption increases and storage requirements accumulate

Engineering Contradiction:
Improvedata processing capabilityVSAvoiddata volume
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential information from raw sensor data through active triaging and labeling processes. By identifying and extracting only high-value data points that require human review or trigger alerts, the system reduces the quantity of data that needs to be stored and transmitted, while maintaining the core data processing capability for meaningful insights

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system performs preliminary processing at the edge device by pre-labeling and triaging data before it is stored or transmitted. This preliminary action filters out redundant information and prepares only essential data for further processing, reducing the overall data volume while preserving processing effectiveness

Inventive Principle:
Principle #10Preliminary action

2Loss of time

If computational storage systems are used to reduce data transfer, then latency decreases, but processing power and energy consumption increase

Engineering Contradiction:
Improvedata transfer latencyVSAvoidprocessing energy consumption
Core Design Contradiction:
Loss of timeVSUse of energy by moving object

Solution Approach 1:

The patent applies partial processing by performing computational storage operations only on a subset of data that has been triaged as high-value through active labeling. Rather than processing all incoming data uniformly, the system selectively applies processing power to only those data points that require it, reducing overall energy consumption while maintaining low latency for critical operations

Inventive Principle:
Principle #16Partial or excessive action

3Productivity

If AI/ML technologies are integrated into edge devices, then data processing efficiency improves, but device complexity and resource requirements increase

Engineering Contradiction:
Improvedata processing efficiencyVSAvoidsystem complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent segments the AI/ML processing into distinct functional components: active triaging module, labeling module, storage module, and alert generation module. This segmentation allows each component to handle specific tasks independently, making the overall system more manageable and easier to deploy in resource-constrained edge environments while maintaining high processing efficiency

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4583063A1System and method for actively triaging data in a computational storage system deployed in an edge environment
Publication Date: 2025.07.09 UBOTICA TECH LTD
  • EP4583063A1 patent drawingFigure 1
  • EP4583063A1 patent drawingFigure 2
  • EP4583063A1 patent drawingFigure 3A

AI summary

Methods and systems for actively triaging data in a computational storage system deployed in an edge environment include receiving an ensemble of signals aggregated from multiple sensors and processing the signals to generate pre-processed data. The pre-processed data may be applied to one or more neural networks to identify events of interest and generate corresponding labels. Outputs from the neural networks may be combined into a unified detection result that may be used to predict future events. A priority classification may be determined for the detection result or predicted event. High-priority events may trigger low-latency alerts, while non-high-priority data and predicted events may be stored with associated labels in onboard non-volatile storage.