https://joeliang0520.github.io/tag/research/researchHugo Blox Builder (https://hugoblox.com)en-usSat, 20 Apr 2024 00:00:00 +0000https://joeliang0520.github.io/media/icon_hu00eb1932855fc2e3835b26f5de7e6bcd_204974_512x512_fill_lanczos_center_3.pnghttps://joeliang0520.github.io/tag/research/https://joeliang0520.github.io/project/mts_transformer/Sat, 20 Apr 2024 00:00:00 +0000https://joeliang0520.github.io/project/mts_transformer/<h2 id="abstract">Abstract</h2> <p>Multivariate Time Series (MTS) Clustering, entailing the grouping of samples exhibiting similarity across multiple temporal variables, holds significant promise for various applications. However, prevailing clustering algorithms often encounter challenges such as noise within raw time series data, diminished feature correla- tion, and necessitated human preprocessing. In response, this paper introduces a novel framework for MTS clustering utilizing Transformer architecture, leveraging its multi-head attention mechanism to potentially capture intricate multivariate relationships. The proposed approach entails learning a cluster-oriented univariate representation of MTS using a Transformer before applying clustering algorithms. The efficacy of this novel framework is substantiated through a series of experi- ments conducted on real-world datasets.</p> <h2 id="keywords">Keywords</h2> <p>Multivariate Time Series, Clustering, Transfomer, Representation Learning, Deep Learning</p> <h2 id="paper">Paper</h2> <p>Please use the link above to access the full paper (Note: this paper is not published and not peer-reviewed)</p>https://joeliang0520.github.io/project/temporal/Sat, 01 Jan 2022 00:00:00 +0000https://joeliang0520.github.io/project/temporal/<p>Clustering, a widely employed unsupervised machine-learning tool, has found applications spanning diverse disciplines. Temporal clustering, particularly the task of grouping unlabelled multivariate time-series data, has attracted considerable attention from researchers.</p> <p>Jolomi Tosanwumi, a current Master of Applied Science student, has recently introduced a generalized framework for temporal clustering that effectively addresses two critical issues inherent in prior algorithms: (1) the incapacity to analyze cluster changes over time and (2) suboptimal outcomes arising from different initializations. In collaboration with Tosanwumi, this project undertook a comparative performance analysis of the newly proposed framework against existing algorithms using synthetic data.</p> <p>Additionally, two case studies were conducted employing real-world census and climate temporal data to validate the efficacy of the proposed temporal clustering framework. The results of the case studies unveiled potential applications through analytical insights, including the discernment of coastal and inland cities using worldwide historical climate data, the revelation of a potential migration pattern in the workforce with occupations in art and culture in the city of Toronto, and the identification of two distinct trends in the rental market of Downtown Toronto.</p>