Schoenegger et al. (2024): AI Forecasting

paper

Authors

Wang, Junyao·Faruque, Mohammad Abdullah Al

Credibility Rating

3/5

Good(3)

Good quality. Reputable source with community review or editorial standards, but less rigorous than peer-reviewed venues.

Rating inherited from publication venue: arXiv

A technical paper on SMORE, a domain adaptation algorithm using hyperdimensional computing for efficient model customization; relevant to AI safety through advances in resource-efficient and adaptable AI systems.

Paper Details

Citations

0 influential

Year

2024

arXiv:2402.13233 DOI:10.1145/3649329.3658477 Semantic Scholar

Metadata

arxiv preprintprimary source

Summary

SMORE is a resource-efficient domain adaptation algorithm using hyperdimensional computing to dynamically customize test-time models. It achieves higher accuracy and faster performance compared to existing deep learning approaches.

Key Points

•First HDC-based domain adaptation algorithm for multi-sensor time series classification
•Dynamically customizes test-time models with explicit domain context consideration
•Achieves higher accuracy and significantly faster performance compared to existing methods

Review

This paper addresses a critical challenge in machine learning: distribution shift in multi-sensor time series data. The authors propose SMORE, an innovative domain adaptation algorithm leveraging hyperdimensional computing (HDC) to handle out-of-distribution samples more efficiently than traditional deep learning methods. By dynamically constructing test-time models that consider domain context, SMORE provides a lightweight and adaptable solution for edge computing platforms. The methodology is particularly noteworthy for its unique approach to encoding multi-sensor time series data and constructing domain-specific models. By using HDC's parallel and efficient operations, SMORE achieves significant improvements in both accuracy and computational efficiency. Experimental results demonstrate an average 1.98% higher accuracy than state-of-the-art domain adaptation algorithms, with 18.81x faster training and 4.63x faster inference. The approach is especially promising for resource-constrained edge devices, where traditional deep learning models struggle with computational limitations.

Cited by 1 page

Page	Type	Quality
AI-Augmented Forecasting	Approach	54.0

Cached Content Preview

HTTP 200Fetched Apr 7, 202658 KB

[2402.13233] SMORE: Similarity-based Hyperdimensional Domain Adaptation for Multi-Sensor Time Series Classification 
 
 
 
 
 
 
 
 
 
 
 

 
 

 
 
 
 
 
 
 SMORE: S imilarity-based Hyperdi m ensional D o main Adaptation for Multi-Senso r Tim e Series Classification

 
 
 Junyao Wang, Mohammad Abdullah Al Faruque
 
 junyaow4, alfaruqu@uci.edu 
 
 Department of Computer Science, University of California, Irvine, USA 
 
 
 (2018) 

 
 Abstract.

 Many real-world applications of the Internet of Things (IoT) employ machine learning (ML) algorithms to analyze time series information collected by interconnected sensors.
However, distribution shift , a fundamental challenge in data-driven ML, arises when a model is deployed on a data distribution different from the training data and can substantially degrade model performance.
Additionally, increasingly sophisticated deep neural networks (DNNs) are required to capture intricate spatial and temporal dependencies in multi-sensor time series data, often exceeding the capabilities of today’s edge devices.
In this paper, we propose 𝖲𝖬𝖮𝖱𝖤 𝖲𝖬𝖮𝖱𝖤 \mathsf{SMORE} , a novel resource-efficient domain adaptation (DA) algorithm for multi-sensor time series classification, leveraging the efficient and parallel operations of hyperdimensional computing.
 𝖲𝖬𝖮𝖱𝖤 𝖲𝖬𝖮𝖱𝖤 \mathsf{SMORE} dynamically customizes test-time models with explicit consideration of the domain context of each sample to mitigate the negative impacts of domain shifts. Our evaluation on a variety of multi-sensor time series classification tasks shows that 𝖲𝖬𝖮𝖱𝖤 𝖲𝖬𝖮𝖱𝖤 \mathsf{SMORE} achieves on average 1.98 % percent 1.98 1.98\% higher accuracy than state-of-the-art (SOTA) DNN-based
DA algorithms with 18.81 18.81 18.81 × faster training and 4.63 4.63 4.63 × faster inference.

 
 † † copyright: acmcopyright † † journalyear: 2018 † † doi: XXXXXXX.XXXXXXX † † conference: Make sure to enter the correct
conference title from your rights confirmation emai; June 23–27,
2024; San Francisco, CA † † price: 15.00 † † isbn: 978-1-4503-XXXX-X/18/06 
 
 
 1. Introduction

 
 Figure 1. Motivation of Our Proposed 𝖲𝖬𝖮𝖱𝖤 𝖲𝖬𝖮𝖱𝖤 \mathsf{SMORE} 
 
 
 With the emergence of the Internet of Things (IoT), many real-world applications
utilize heterogeneously connected sensors to collect information over the course of time, constituting multi-sensor time series data  ( qiao2018time, ) .
Machine learning (ML) algorithms, including deep neural networks (DNNs), are often employed to analyze the collected data and perform various learning tasks.
However, distribution shift (DS), a fundamental challenge in data-driven ML, can substantially degrade model performance.
In particular, the excellent performance of these ML algorithms heavily relies on the critical assumption that the training and inference data come from the same distribution, while this assumption can be easily violated as out-of-distribution (OOD) scenarios are inevitable in real-wor

... (truncated, 58 KB total)

Resource ID: 45ecdd052d700154 | Stable ID: OTg2YjQ1Yj