Encoded Tsetlin Machine for Resource-Constrained Sensor Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Deep neural networks (DNNs) and traditional Tsetlin Machines face challenges when deployed on resource-constrained devices due to high memory requirements and computational intensity, making them unsuitable for batteryless sensors that rely on intermittent energy sources.

Innovation Solution

The implementation of encoded Tsetlin Machines with a compressed representation based on a higher number of exclude decisions than include decisions, combined with power-aware adaptive execution techniques, allows for efficient deployment on resource-constrained devices by reducing memory footprint and energy consumption while maintaining low latency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep neural networks are deployed on resource constrained devices, then classification accuracy is improved, but memory consumption and computational capacity requirements increase

Engineering Contradiction:

Improveclassification accuracyVSAvoidmemory consumption

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent transforms the Tsetlin machine model parameters into a compressed encoded format that reduces memory footprint. By changing the representation parameters from full model storage to compressed encoding schemes, the system achieves lower memory consumption while preserving classification functionality on resource-constrained devices

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent extracts only the essential classification logic from the full Tsetlin machine model and encodes it in a compressed format. By separating and storing only the critical decision rules in encoded form, the system reduces memory requirements while maintaining adequate classification accuracy for resource-constrained environments

Inventive Principle:

Principle #2Taking out (Extraction)

2Measurement precision

If deep neural networks are deployed on resource constrained devices, then classification accuracy is improved, but computational capacity requirements increase

Engineering Contradiction:

Improveclassification accuracyVSAvoidcomputational capacity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent uses simple, computationally inexpensive encoded representations that can be quickly executed on resource-constrained devices. The encoded Tsetlin machine uses lightweight logical operations instead of complex neural network computations, enabling deployment on devices with limited computational capacity while maintaining classification functionality

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

3Quantity of substance

If Tsetlin machine is encoded with compressed representation, then memory footprint is reduced, but model complexity increases

Engineering Contradiction:

Improvememory footprintVSAvoidencoding complexity

Core Design Contradiction:

Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent performs the encoding transformation in advance during model deployment, converting the Tsetlin machine into a compressed encoded representation before execution. This preliminary encoding action reduces the memory footprint for runtime execution, while the encoding complexity is paid only once during deployment rather than during each inference

Inventive Principle:

Principle #10Preliminary action

4Use of energy by moving object

If clauses are dropped from Tsetlin machine based on energy availability, then energy consumption is reduced, but classification accuracy may deteriorate

Engineering Contradiction:

Improveenergy consumptionVSAvoidclassification accuracy

Core Design Contradiction:

Use of energy by moving objectVSMeasurement precision

Solution Approach 1:

The patent implements partial execution of the Tsetlin machine by dropping less important clauses when energy is constrained. By selectively executing only the most critical clauses based on energy availability, the system achieves energy-efficient operation while maintaining acceptable classification accuracy through prioritization of essential decision rules

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20240127077A1System for the deployment of fast and memory efficient tsetlin machines models on resource constrained devices

Publication Date: 2024.04.18 NOKIA TECHNOLOGIES OY

AI summary

This specification describes systems, apparatus and methods for deploying Tsetlin machine models on resource-constrained devices. According to a first aspect of this specification, there is described apparatus comprising: one or more sensors; at least one processor; and at least one memory with storing instructions that, when executed by the at least one processor, cause the apparatus at to at least: collect one or more sets of sensor data using the one or more sensors; classify the one or more sets of sensor data using an encoded Tsetlin machine. The encoded Tsetlin machine comprises a compressed representation of a trained Tsetlin machine. The compressed representation is based on a number of exclude decisions of the trained Tsetlin machine being greater than a number of include decisions of the trained Tsetlin machine.