Search

Your search keyword '"Fujisawa, Masahiro"' showing total 22 results
Start Over Author "Fujisawa, Masahiro"
22 results on '"Fujisawa, Masahiro"'

1. PAC-Bayes Analysis for Recalibration in Classification

Author

Fujisawa, Masahiro and Futami, Futoshi

Subjects
Computer Science - Machine Learning, Statistics - Machine Learning
Abstract

Nonparametric estimation with binning is widely employed in the calibration error evaluation and the recalibration of machine learning models. Recently, theoretical analyses of the bias induced by this estimation approach have been actively pursued; however, the understanding of the generalization of the calibration error to unknown data remains limited. In addition, although many recalibration algorithms have been proposed, their generalization performance lacks theoretical guarantees. To address this problem, we conduct a generalization analysis of the calibration error under the probably approximately correct (PAC) Bayes framework. This approach enables us to derive a first optimizable upper bound for the generalization error in the calibration context. We then propose a generalization-aware recalibration algorithm based on our generalization theory. Numerical experiments show that our algorithm improves the Gaussian-process-based recalibration performance on various benchmark datasets and models., Comment: 27 pages, 3 figures

Published
2024

2. Information-theoretic Generalization Analysis for Expected Calibration Error

Author

Futami, Futoshi and Fujisawa, Masahiro

Subjects
Computer Science - Machine Learning, Mathematics - Statistics Theory, Statistics - Machine Learning
Abstract

While the expected calibration error (ECE), which employs binning, is widely adopted to evaluate the calibration performance of machine learning models, theoretical understanding of its estimation bias is limited. In this paper, we present the first comprehensive analysis of the estimation bias in the two common binning strategies, uniform mass and uniform width binning. Our analysis establishes upper bounds on the bias, achieving an improved convergence rate. Moreover, our bounds reveal, for the first time, the optimal number of bins to minimize the estimation bias. We further extend our bias analysis to generalization error analysis based on the information-theoretic approach, deriving upper bounds that enable the numerical evaluation of how small the ECE is for unknown data. Experiments using deep learning models show that our bounds are nonvacuous thanks to this information-theoretic generalization analysis approach., Comment: 34 pages, 3 figures

Published
2024

5. Time-Independent Information-Theoretic Generalization Bounds for SGLD

Author

Futami, Futoshi and Fujisawa, Masahiro

Subjects
Computer Science - Machine Learning, Statistics - Machine Learning
Abstract

We provide novel information-theoretic generalization bounds for stochastic gradient Langevin dynamics (SGLD) under the assumptions of smoothness and dissipativity, which are widely used in sampling and non-convex optimization studies. Our bounds are time-independent and decay to zero as the sample size increases, regardless of the number of iterations and whether the step size is fixed. Unlike previous studies, we derive the generalization error bounds by focusing on the time evolution of the Kullback--Leibler divergence, which is related to the stability of datasets and is the upper bound of the mutual information between output parameters and an input dataset. Additionally, we establish the first information-theoretic generalization bound when the training and test loss are the same by showing that a loss function of SGLD is sub-exponential. This bound is also time-independent and removes the problematic step size dependence in existing work, leading to an improved excess risk bound by combining our analysis with the existing non-convex optimization error bounds., Comment: Accepted by the Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS2023), 29 pages

Published
2023

6. Understanding the Expressivity and Trainability of Fourier Neural Operator: A Mean-Field Perspective

Author

Koshizuka, Takeshi, Fujisawa, Masahiro, Tanaka, Yusuke, and Sato, Issei

Subjects
Computer Science - Machine Learning
Abstract

In this paper, we explores the expressivity and trainability of the Fourier Neural Operator (FNO). We establish a mean-field theory for the FNO, analyzing the behavior of the random FNO from an edge of chaos perspective. Our investigation into the expressivity of a random FNO involves examining the ordered-chaos phase transition of the network based on the weight distribution. This phase transition demonstrates characteristics unique to the FNO, induced by mode truncation, while also showcasing similarities to those of densely connected networks. Furthermore, we identify a connection between expressivity and trainability: the ordered and chaotic phases correspond to regions of vanishing and exploding gradients, respectively. This finding provides a practical prerequisite for the stable training of the FNO. Our experimental results corroborate our theoretical findings.

Published
2023

10. $\gamma$-ABC: Outlier-Robust Approximate Bayesian Computation Based on a Robust Divergence Estimator

Author

Fujisawa, Masahiro, Teshima, Takeshi, Sato, Issei, and Sugiyama, Masashi

Subjects
Statistics - Machine Learning, Computer Science - Machine Learning, Statistics - Computation, Statistics - Methodology
Abstract

Approximate Bayesian computation (ABC) is a likelihood-free inference method that has been employed in various applications. However, ABC can be sensitive to outliers if a data discrepancy measure is chosen inappropriately. In this paper, we propose to use a nearest-neighbor-based $\gamma$-divergence estimator as a data discrepancy measure. We show that our estimator possesses a suitable theoretical robustness property called the redescending property. In addition, our estimator enjoys various desirable properties such as high flexibility, asymptotic unbiasedness, almost sure convergence, and linear-time computational complexity. Through experiments, we demonstrate that our method achieves significantly higher robustness than existing discrepancy measures., Comment: The 24th International Conference on Artificial Intelligence and Statistics (AISTATS 2021); 48 pages, 22 figures

Published
2020

11. Multilevel Monte Carlo Variational Inference

Author

Fujisawa, Masahiro and Sato, Issei

Subjects
Statistics - Machine Learning, Computer Science - Machine Learning
Abstract

We propose a variance reduction framework for variational inference using the Multilevel Monte Carlo (MLMC) method. Our framework is built on reparameterized gradient estimators and "recycles" parameters obtained from past update history in optimization. In addition, our framework provides a new optimization algorithm based on stochastic gradient descent (SGD) that adaptively estimates the sample size used for gradient estimation according to the ratio of the gradient variance. We theoretically show that, with our method, the variance of the gradient estimator decreases as optimization proceeds and that a learning rate scheduler function helps improve the convergence. We also show that, in terms of the \textit{signal-to-noise} ratio, our method can improve the quality of gradient estimation by the learning rate scheduler function without increasing the initial sample size. Finally, we confirm that our method achieves faster convergence and reduces the variance of the gradient estimator compared with other methods through experimental comparisons with baseline methods using several benchmark datasets., Comment: 44pages, 10 figures; Journal of Machine Learning Research (JMLR)

Published
2019

13. Initialization Bias of Fourier Neural Operator: Revisiting the Edge of Chaos

Author

Koshizuka, Takeshi, Fujisawa, Masahiro, Tanaka, Yusuke, Sato, Issei, Koshizuka, Takeshi, Fujisawa, Masahiro, Tanaka, Yusuke, and Sato, Issei

Abstract

This paper investigates the initialization bias of the Fourier neural operator (FNO). A mean-field theory for FNO is established, analyzing the behavior of the random FNO from an \emph{edge of chaos} perspective. We uncover that the forward and backward propagation behaviors exhibit characteristics unique to FNO, induced by mode truncation, while also showcasing similarities to those of densely connected networks. Building upon this observation, we also propose an edge of chaos initialization scheme for FNO to mitigate the negative initialization bias leading to training instability. Experimental results show the effectiveness of our initialization scheme, enabling stable training of deep FNO without skip-connection.

Published
2023

14. Takafuta-Tsukatani Mounded Tomb, Jinseki-kōgen Town, Hiroshima Prefecture

Author

NOJIMA, Hisashi, MORIMOTO, Naoto, FUJISAWA, Masahiro, and SHIMOE, Hiroki

Abstract

Takafuta-Tsukatani mounded tomb (kofun) was located in Jinseki-kōgen Town, Hiroshima Prefecture. However, immediately after World War II the site was reclaimed as farmland and the mound disappeared. In 2010, some of the finds unearthed by the landowner of the tomb were donated to Jinseki-kōgen Town. Furthermore, other relics from the Takafuta-Tsukatani mounded tomb were also held at the Hiroshima Prefectural Fuchū High School, but are now in the possession of the Hiroshima Prefectural Museum of History. In this article, all unearthed finds are presented in their entirety. Takafuta-Tsukatani mounded tomb was built during the Late Kofun period (latter half of the 6th century A.D.). Many objects were unearthed, including iron swords, a hand guard of a sword (tsuba), horse bits (kutsuwa), stirrups (abumi), beads of necklaces, Haji ware (hajiki), and Sue ware (sueki). In particular, horse bits with gourd-shaped iron rings (hisagogata kanjō kagami itatsuki kutsuwa) are characteristic for mounded tombs in the Kibi region (ancient province consisting of Okayama Prefecture and the eastern part of Hiroshima Prefecture). In the latter half of the 6th century the first iron manufacture sites appeared in the Bingo region (eastern part of Hiroshima Prefecture). It is highly probable that the person buried in the Takafuta-Tsukatani mounded tomb was involved in horse harness and trappings production, iron manufacture and blacksmithing unique to the Kibi region., 今回の資料紹介は2018・2019年度大学院文学研究科考古学研究室の資料実習授業における遺物観察･実測、写真撮影をもとにし、神石高原町の委託研究費を一部使用した。

Published
2020

15. Topographical Survey of the Chōja Sukumozuka Mounded Tomb No.1 in Saijō, Higashi-Hiroshima City

Author

Shimoe, Hiroki, Nagano, Tomoro, and Fujisawa, Masahiro

Abstract

The Chōja Sukumozuka mounded tomb No.1 is located in Saijō-chō, Higashi-Hiroshima City, and believed to be a scallop-shaped mounded tomb Hiroshima Prefecture. Because the northeastern part of Hiroshima Prefecture has many scallop-shaped mounded tombs, the Chōja Sukumozuka mounded tomb No.1 has been thought to be similar to them. However, according to our recent research, there is a possibility that the shape and background of Chōja Sukumozuka mounded tomb No.1 differ from those of other northeastern scallop-shaped mounds. For this reason, as part of “Field Practice in Archaeology A and B” for the archaeology majors at the Hiroshima University School of Letters, we conducted a topographical survey of this mound with the aim of clarifying its scale, shape, and date of construction in August and September, 2017. The results of this survey indicated that the total length of the Chōja Sukumozuka mounded tomb No.1 is 60.0m; 19.5m for the square front part and 40.5m the round rear portion. In addition, various kinds of haniwa fragments of cylindrical, morning glory-shaped, sunshade-shaped and house-shaped ones were collected from the mound. Features of its cylindrical haniwa indicate that this mound was constructed between the end of the 4th century AD and the beginning of the 5th century AD. Based on the results of this survey, the shape of the Chōja Sukumozuka mounded tomb No.1 is a middle form of a keyhole-shaped and a scallop-shaped mound; therefore, the shape is different from those of the scallop-shaped mounds in the northeastern part of Hiroshima Prefecture and the Kinai Region, which were constructed in the same phase. In conclusion, we can assume that Chōja Sukumozuka mounded tomb No.1 and Maruyama Shrine mounded tomb No.1 have a different background from other areas which have scallop-shaped mound tombs, Therefore, the scallop-shaped mounded tombs in the Saijō basin were adopted as a local, unique mound shape before the Mitsujō mounded tomb No.1 was constructed.

Published
2018

16. 広島市白山第１号古墳出土の短甲について

Author

FUJISAWA, Masahiro

Abstract

Shiroyama mounded tomb (kofun) No.1 is located in Asaminami ward, Hiroshima City. Up to now, several unearthed fragments of chest armor have been reported. While the details and chronological positions of those finds have remained unclear for a long time, an analysis of the fragments, which were kept in the Hiroshima Prefectural History and Folklore Museum, revealed that they belonged to a type of chest armor made of horizontally arranged long metal stripes riveted together (yokohagiita byōdome tankō). It became clear that they could be dated around the middle third of 5th century A. D.. Although there is still room for examination, all chest armor unearthed in the Hiroshima Bay area (including the materials from Shiroyama mounded tomb No.1) is basically considered to have been produced since the middle third of the 5th century A. D..

Published
2020

18. This title is unavailable for guests, please login to see more information.

Author

Shimoe, Hiroki, Nagano, Tomoro, Fujisawa, Masahiro, Shimoe, Hiroki, Nagano, Tomoro, and Fujisawa, Masahiro

Abstract

The Chōja Sukumozuka mounded tomb No.1 is located in Saijō-chō, Higashi-Hiroshima City, and believed to be a scallop-shaped mounded tomb Hiroshima Prefecture. Because the northeastern part of Hiroshima Prefecture has many scallop-shaped mounded tombs, the Chōja Sukumozuka mounded tomb No.1 has been thought to be similar to them. However, according to our recent research, there is a possibility that the shape and background of Chōja Sukumozuka mounded tomb No.1 differ from those of other northeastern scallop-shaped mounds. For this reason, as part of “Field Practice in Archaeology A and B” for the archaeology majors at the Hiroshima University School of Letters, we conducted a topographical survey of this mound with the aim of clarifying its scale, shape, and date of construction in August and September, 2017. The results of this survey indicated that the total length of the Chōja Sukumozuka mounded tomb No.1 is 60.0m; 19.5m for the square front part and 40.5m the round rear portion. In addition, various kinds of haniwa fragments of cylindrical, morning glory-shaped, sunshade-shaped and house-shaped ones were collected from the mound. Features of its cylindrical haniwa indicate that this mound was constructed between the end of the 4th century AD and the beginning of the 5th century AD. Based on the results of this survey, the shape of the Chōja Sukumozuka mounded tomb No.1 is a middle form of a keyhole-shaped and a scallop-shaped mound; therefore, the shape is different from those of the scallop-shaped mounds in the northeastern part of Hiroshima Prefecture and the Kinai Region, which were constructed in the same phase. In conclusion, we can assume that Chōja Sukumozuka mounded tomb No.1 and Maruyama Shrine mounded tomb No.1 have a different background from other areas which have scallop-shaped mound tombs, Therefore, the scallop-shaped mounded tombs in the Saijō basin were adopted as a local, unique mound shape before the Mitsujō mounded tomb No.1 was constructed.

19. This title is unavailable for guests, please login to see more information.

Author

FUJISAWA, Masahiro and FUJISAWA, Masahiro

Abstract

Shiroyama mounded tomb (kofun) No.1 is located in Asaminami ward, Hiroshima City. Up to now, several unearthed fragments of chest armor have been reported. While the details and chronological positions of those finds have remained unclear for a long time, an analysis of the fragments, which were kept in the Hiroshima Prefectural History and Folklore Museum, revealed that they belonged to a type of chest armor made of horizontally arranged long metal stripes riveted together (yokohagiita byōdome tankō). It became clear that they could be dated around the middle third of 5th century A. D.. Although there is still room for examination, all chest armor unearthed in the Hiroshima Bay area (including the materials from Shiroyama mounded tomb No.1) is basically considered to have been produced since the middle third of the 5th century A. D..

20. This title is unavailable for guests, please login to see more information.

Author

NOJIMA, Hisashi, MORIMOTO, Naoto, FUJISAWA, Masahiro, SHIMOE, Hiroki, NOJIMA, Hisashi, MORIMOTO, Naoto, FUJISAWA, Masahiro, and SHIMOE, Hiroki

Abstract

Takafuta-Tsukatani mounded tomb (kofun) was located in Jinseki-kōgen Town, Hiroshima Prefecture. However, immediately after World War II the site was reclaimed as farmland and the mound disappeared. In 2010, some of the finds unearthed by the landowner of the tomb were donated to Jinseki-kōgen Town. Furthermore, other relics from the Takafuta-Tsukatani mounded tomb were also held at the Hiroshima Prefectural Fuchū High School, but are now in the possession of the Hiroshima Prefectural Museum of History. In this article, all unearthed finds are presented in their entirety. Takafuta-Tsukatani mounded tomb was built during the Late Kofun period (latter half of the 6th century A.D.). Many objects were unearthed, including iron swords, a hand guard of a sword (tsuba), horse bits (kutsuwa), stirrups (abumi), beads of necklaces, Haji ware (hajiki), and Sue ware (sueki). In particular, horse bits with gourd-shaped iron rings (hisagogata kanjō kagami itatsuki kutsuwa) are characteristic for mounded tombs in the Kibi region (ancient province consisting of Okayama Prefecture and the eastern part of Hiroshima Prefecture). In the latter half of the 6th century the first iron manufacture sites appeared in the Bingo region (eastern part of Hiroshima Prefecture). It is highly probable that the person buried in the Takafuta-Tsukatani mounded tomb was involved in horse harness and trappings production, iron manufacture and blacksmithing unique to the Kibi region.

21. Feasibility of major hepatectomy following preoperative chemotherapy for advanced perihilar cholangiocarcinoma.

Author

Takahashi A, Yoshioka R, Koike S, Fujisawa M, Oba M, Takeda Y, Fukumura Y, Kotera Y, Mise Y, and Saiura A

Abstract

Background: The safety of major hepatectomy following preoperative chemotherapy for perihilar cholangiocarcinoma (PHCC) is underexplored. This study evaluates the impact of preoperative chemotherapy on surgical outcomes and assesses chemotherapy-induced liver injury in patients with advanced PHCC., Methods: This retrospective study included 62 PHCC patients who underwent surgery between January 2019 and January 2024. Patients were divided into an upfront surgery group (UFS, n = 31) and a preoperative chemotherapy group (POC, n = 31). Preoperative chemotherapy was indicated when R0/R1 resection was unachievable, complex surgery was needed, or future liver reserve was insufficient. Baseline characteristics, surgical procedures, postoperative complications, and pathological findings were compared., Results: Postoperative complications were comparable between groups, with Clavien-Dindo grade ≥3a rates of 30.7 % in the POC group and 24.3 % in the UFS group. Despite longer operative times and hospital stays in the POC group, no significant differences in hepatotoxicity or pathological findings, including Kleiner and Rubbia-Brandt scores, were observed. Notably, a pathological complete response was achieved in 12.9 % of the POC group., Conclusion: Major hepatectomy following preoperative chemotherapy for PHCC is safe and does not increase the risk of postoperative complications or hepatotoxicity. Further studies are warranted to refine resectability criteria and optimize patient selection., (Copyright © 2024 Elsevier Ltd, BASO ~ The Association for Cancer Surgery, and the European Society of Surgical Oncology. All rights reserved.)

Published
2024
Full Text
View/download PDF

22. [Late recurrence of gastric cancer diagnosed 10 years after curative gastrectomy: a review of 30 Japanese case reports].

Author

Fujisawa M, Watanobe I, Machida M, and Sugo H

Subjects
Female, Gastrectomy, Humans, Japan, Middle Aged, Adenocarcinoma surgery, Carcinoma, Signet Ring Cell diagnostic imaging, Carcinoma, Signet Ring Cell surgery, Stomach Neoplasms surgery
Abstract

A woman in her 50s was referred to our hospital with intestinal obstruction. Ten years prior, she had been treated for gastric cancer, pathologically confirmed as stage IIIA poorly differentiated adenocarcinoma with signet-ring cell carcinoma. Intraoperatively, a 4-cm hard white tumor was found in the mesoileum and around the ileum. Pathological examination revealed poorly differentiated adenocarcinoma with signet-ring cell carcinoma and infiltration and fibrosis. Late peritoneal recurrence of gastric carcinoma was diagnosed. Recurrence of gastric carcinoma more than 10 years after curative gastrectomy is extremely rare. A review of 30 cases reported in Japan revealed recurrence was more frequent in females (60%) and the mean age was around 50 years at the time of primary surgery. Poorly differentiated adenocarcinoma and/or signet-ring cell carcinoma was the primary gastric cancer in 82% of cases and bone metastasis was the most frequent site of recurrence.

Published
2022
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results