Showing 1–6 of 6 results for author: Tashiro, Y

Self-Supervised Learning of Time Series Representation via Diffusion Process and Imputation-Interpolation-Forecasting Mask

Authors: Zineb Senane, Lele Cao, Valentin Leonhard Buchner, Yusuke Tashiro, Lei You, Pawel Herman, Mats Nordahl, Ruibo Tu, Vilhelm von Ehrenheim

Abstract: Time Series Representation Learning (TSRL) focuses on generating informative representations for various Time Series (TS) modeling tasks. Traditional Self-Supervised Learning (SSL) methods in TSRL fall into four main categories: reconstructive, adversarial, contrastive, and predictive, each with a common challenge of sensitivity to noise and intricate data nuances. Recently, diffusion-based method… ▽ More Time Series Representation Learning (TSRL) focuses on generating informative representations for various Time Series (TS) modeling tasks. Traditional Self-Supervised Learning (SSL) methods in TSRL fall into four main categories: reconstructive, adversarial, contrastive, and predictive, each with a common challenge of sensitivity to noise and intricate data nuances. Recently, diffusion-based methods have shown advanced generative capabilities. However, they primarily target specific application scenarios like imputation and forecasting, leaving a gap in leveraging diffusion models for generic TSRL. Our work, Time Series Diffusion Embedding (TSDE), bridges this gap as the first diffusion-based SSL TSRL approach. TSDE segments TS data into observed and masked parts using an Imputation-Interpolation-Forecasting (IIF) mask. It applies a trainable embedding function, featuring dual-orthogonal Transformer encoders with a crossover mechanism, to the observed part. We train a reverse diffusion process conditioned on the embeddings, designed to predict noise added to the masked part. Extensive experiments demonstrate TSDE's superiority in imputation, interpolation, forecasting, anomaly detection, classification, and clustering. We also conduct an ablation study, present embedding visualizations, and compare inference speed, further substantiating TSDE's efficiency and validity in learning representations of TS data. △ Less

Submitted 17 June, 2024; v1 submitted 9 May, 2024; originally announced May 2024.

Comments: Published as a full paper by KDD 2024 Research Track (12 pages as main paper and 11 pages as appendix). Source code available at https://github.com/llcresearch/TSDE

ACM Class: G.3; I.6.5; I.2.4

A novel sustainable role of compost as a universal protective substitute for fish, chicken, pig, and cattle, and its estimation by structural equation modeling

Authors: Hirokuni Miyamoto, Wataru Suda, Hiroaki Kodama, Hideyuki Takahashi, Yumiko Nakanishi, Shigeharu Moriya, Kana Adachi, Nao Kiriyama, Masaya Wada, Daisuke Sudo, Shunsuke Ito, Shunsuke Ito, Minami Shibata, Shinji Wada, Takako Murano, Hitoshi Taguchi, Chie Shindo, Arisa Tsuboi, Naoko Tsuji, Makiko Matsuura, Chitose Ishii, Teruno Nakaguma, Toshiyuki Ito, Toru Okada, Teruo Matsushita , et al. (18 additional authors not shown)

Abstract: Natural decomposition of organic matter is essential in food systems, and compost is used worldwide as an organic fermented fertilizer. However, as a feature of the ecosystem, its effects on the animals are poorly understood. Here we show that oral administration of compost and/or its derived thermophilic Bacillaceae, i.e., Caldibacillus hisashii and Weizmannia coagulans, can modulate the prophyla… ▽ More Natural decomposition of organic matter is essential in food systems, and compost is used worldwide as an organic fermented fertilizer. However, as a feature of the ecosystem, its effects on the animals are poorly understood. Here we show that oral administration of compost and/or its derived thermophilic Bacillaceae, i.e., Caldibacillus hisashii and Weizmannia coagulans, can modulate the prophylactic activities of various industrial animals. The fecal omics analyses in the modulatory process showed an improving trend dependent upon animal species, environmental conditions, and administration. However, structural equation modeling (SEM) estimated the grouping candidates of bacteria and metabolites as standard key components beyond the animal species. In particular, the SEM model implied a strong relationship among partly digesting fecal amino acids, increasing genus Lactobacillus as inhabitant beneficial bacteria and 2-aminoisobutyric acid involved in lantibiotics. These results highlight the potential role of compost for sustainable protective control in agriculture, fishery, and livestock industries. △ Less

Submitted 27 November, 2022; v1 submitted 26 January, 2022; originally announced January 2022.

CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation

Authors: Yusuke Tashiro, Jiaming Song, Yang Song, Stefano Ermon

Abstract: The imputation of missing values in time series has many applications in healthcare and finance. While autoregressive models are natural candidates for time series imputation, score-based diffusion models have recently outperformed existing counterparts including autoregressive models in many tasks such as image generation and audio synthesis, and would be promising for time series imputation. In… ▽ More The imputation of missing values in time series has many applications in healthcare and finance. While autoregressive models are natural candidates for time series imputation, score-based diffusion models have recently outperformed existing counterparts including autoregressive models in many tasks such as image generation and audio synthesis, and would be promising for time series imputation. In this paper, we propose Conditional Score-based Diffusion models for Imputation (CSDI), a novel time series imputation method that utilizes score-based diffusion models conditioned on observed data. Unlike existing score-based approaches, the conditional diffusion model is explicitly trained for imputation and can exploit correlations between observed values. On healthcare and environmental data, CSDI improves by 40-65% over existing probabilistic imputation methods on popular performance metrics. In addition, deterministic imputation by CSDI reduces the error by 5-20% compared to the state-of-the-art deterministic imputation methods. Furthermore, CSDI can also be applied to time series interpolation and probabilistic forecasting, and is competitive with existing baselines. The code is available at https://github.com/ermongroup/CSDI. △ Less

Submitted 27 October, 2021; v1 submitted 7 July, 2021; originally announced July 2021.

Comments: NeurIPS 2021

Diversity can be Transferred: Output Diversification for White- and Black-box Attacks

Authors: Yusuke Tashiro, Yang Song, Stefano Ermon

Abstract: Adversarial attacks often involve random perturbations of the inputs drawn from uniform or Gaussian distributions, e.g., to initialize optimization-based white-box attacks or generate update directions in black-box attacks. These simple perturbations, however, could be sub-optimal as they are agnostic to the model being attacked. To improve the efficiency of these attacks, we propose Output Divers… ▽ More Adversarial attacks often involve random perturbations of the inputs drawn from uniform or Gaussian distributions, e.g., to initialize optimization-based white-box attacks or generate update directions in black-box attacks. These simple perturbations, however, could be sub-optimal as they are agnostic to the model being attacked. To improve the efficiency of these attacks, we propose Output Diversified Sampling (ODS), a novel sampling strategy that attempts to maximize diversity in the target model's outputs among the generated samples. While ODS is a gradient-based strategy, the diversity offered by ODS is transferable and can be helpful for both white-box and black-box attacks via surrogate models. Empirically, we demonstrate that ODS significantly improves the performance of existing white-box and black-box attacks. In particular, ODS reduces the number of queries needed for state-of-the-art black-box attacks on ImageNet by a factor of two. △ Less

Submitted 29 October, 2020; v1 submitted 15 March, 2020; originally announced March 2020.

Comments: NeurIPS 2020

Study of Vertical Magnetic Field in Face-on Galaxies using Faraday Tomography

Authors: Shinsuke Ideguchi, Yuichi Tashiro, Takuya Akahori, Keitaro Takahashi, Dongsu Ryu

Abstract: Faraday tomography allows astronomers to probe the distribution of magnetic field along the line of sight (LOS), but that can be achieved only after Faraday spectrum is interpreted. However, the interpretation is not straightforward, mainly because Faraday spectrum is complicated due to turbulent magnetic field; it ruins the one-to-one relation between the Faraday depth and the physical depth, and… ▽ More Faraday tomography allows astronomers to probe the distribution of magnetic field along the line of sight (LOS), but that can be achieved only after Faraday spectrum is interpreted. However, the interpretation is not straightforward, mainly because Faraday spectrum is complicated due to turbulent magnetic field; it ruins the one-to-one relation between the Faraday depth and the physical depth, and appears as many small-scale features in Faraday spectrum. In this paper, employing "simple toy models" for the magnetic field, we describe numerically as well as analytically the characteristic properties of Faraday spectrum. We show that Faraday spectrum along "multiple loss" can be used to extract the global properties of magnetic field. Specifically, considering face-on spiral galaxies and modeling turbulent magnetic field as a random field with single coherence length, we numerically calculate Faraday spectrum along a number of LOSs and its shape-characterizing parameters, that is, the moments. When multiple LOSs cover a region of $\gtrsim (10\ {\rm coherence\ length)^2}$, the shape of Faraday spectrum becomes smooth and the shape-characterizing parameters are well specified. With the Faraday spectrum constructed as a sum of Gaussian functions with different means and variances, we analytically show that the parameters are expressed in terms of the regular and turbulent components of LOS magnetic field and the coherence length. We also consider the turbulent magnetic field modeled with power-law spectrum, and study how the magnetic field is revealed in Faraday spectrum. Our work suggests a way toward obtaining the information of magnetic field from Faraday tomography study. △ Less

Submitted 12 June, 2017; v1 submitted 1 January, 2017; originally announced January 2017.

Comments: To appear in ApJ

Faraday dispersion functions of galaxies

Authors: Shinsuke Ideguchi, Yuichi Tashiro, Takuya Akahori, Keitaro Takahashi, Dongsu Ryu

Abstract: The Faraday dispersion function (FDF), which can be derived from an observed polarization spec- trum by Faraday rotation measure synthesis, is a profile of polarized emissions as a function of Faraday depth. We study intrinsic FDFs along sight lines through face-on, Milky-Way-like galaxies by means of a sophisticated galactic model incorporating 3D MHD turbulence, and investigate how much the FDF… ▽ More The Faraday dispersion function (FDF), which can be derived from an observed polarization spec- trum by Faraday rotation measure synthesis, is a profile of polarized emissions as a function of Faraday depth. We study intrinsic FDFs along sight lines through face-on, Milky-Way-like galaxies by means of a sophisticated galactic model incorporating 3D MHD turbulence, and investigate how much the FDF contains information intrinsically. Since the FDF reflects distributions of thermal and cosmic- ray electrons as well as magnetic fields, it has been expected that the FDF could be a new probe to examine internal structures of galaxies. We, however, find that an intrinsic FDF along a sight line through a galaxy is very complicated, depending significantly on actual configurations of turbulence. We perform 800 realizations of turbulence, and find no universal shape of the FDF even if we fix the global parameters of the model. We calculate the probability distribution functions of the standard deviation, skewness, and kurtosis of FDFs and compare them for models with different global pa- rameters. Our models predict that the presence of vertical magnetic fields and large scale-height of cosmic-ray electrons tend to make the standard deviation relatively large. Contrastingly, differences in skewness and kurtosis are relatively less significant. △ Less

Submitted 1 July, 2014; originally announced July 2014.

Comments: 8 pages, 11 figures, 1 table, to be published in ApJ