-
Statistical Test for Generated Hypotheses by Diffusion Models
Authors:
Teruyuki Katsuoka,
Tomohiro Shiraishi,
Daiki Miwa,
Vo Nguyen Le Duy,
Ichiro Takeuchi
Abstract:
The enhanced performance of AI has accelerated its integration into scientific research. In particular, the use of generative AI to create scientific hypotheses is promising and is increasingly being applied across various fields. However, when employing AI-generated hypotheses for critical decisions, such as medical diagnoses, verifying their reliability is crucial. In this study, we consider a m…
▽ More
The enhanced performance of AI has accelerated its integration into scientific research. In particular, the use of generative AI to create scientific hypotheses is promising and is increasingly being applied across various fields. However, when employing AI-generated hypotheses for critical decisions, such as medical diagnoses, verifying their reliability is crucial. In this study, we consider a medical diagnostic task using generated images by diffusion models, and propose a statistical test to quantify its reliability. The basic idea behind the proposed statistical test is to employ a selective inference framework, where we consider a statistical test conditional on the fact that the generated images are produced by a trained diffusion model. Using the proposed method, the statistical reliability of medical image diagnostic results can be quantified in the form of a p-value, allowing for decision-making with a controlled error rate. We show the theoretical validity of the proposed statistical test and its effectiveness through numerical experiments on synthetic and brain image datasets.
△ Less
Submitted 18 February, 2024;
originally announced February 2024.
-
Statistical Test for Anomaly Detections by Variational Auto-Encoders
Authors:
Daiki Miwa,
Tomohiro Shiraishi,
Vo Nguyen Le Duy,
Teruyuki Katsuoka,
Ichiro Takeuchi
Abstract:
In this study, we consider the reliability assessment of anomaly detection (AD) using Variational Autoencoder (VAE). Over the last decade, VAE-based AD has been actively studied in various perspective, from method development to applied research. However, when the results of ADs are used in high-stakes decision-making, such as in medical diagnosis, it is necessary to ensure the reliability of the…
▽ More
In this study, we consider the reliability assessment of anomaly detection (AD) using Variational Autoencoder (VAE). Over the last decade, VAE-based AD has been actively studied in various perspective, from method development to applied research. However, when the results of ADs are used in high-stakes decision-making, such as in medical diagnosis, it is necessary to ensure the reliability of the detected anomalies. In this study, we propose the VAE-AD Test as a method for quantifying the statistical reliability of VAE-based AD within the framework of statistical testing. Using the VAE-AD Test, the reliability of the anomaly regions detected by a VAE can be quantified in the form of p-values. This means that if an anomaly is declared when the p-value is below a certain threshold, it is possible to control the probability of false detection to a desired level. Since the VAE-AD Test is constructed based on a new statistical inference framework called selective inference, its validity is theoretically guaranteed in finite samples. To demonstrate the validity and effectiveness of the proposed VAE-AD Test, numerical experiments on artificial data and applications to brain image analysis are conducted.
△ Less
Submitted 2 June, 2024; v1 submitted 6 February, 2024;
originally announced February 2024.
-
Statistical Test for Attention Map in Vision Transformer
Authors:
Tomohiro Shiraishi,
Daiki Miwa,
Teruyuki Katsuoka,
Vo Nguyen Le Duy,
Kouichi Taji,
Ichiro Takeuchi
Abstract:
The Vision Transformer (ViT) demonstrates exceptional performance in various computer vision tasks. Attention is crucial for ViT to capture complex wide-ranging relationships among image patches, allowing the model to weigh the importance of image patches and aiding our understanding of the decision-making process. However, when utilizing the attention of ViT as evidence in high-stakes decision-ma…
▽ More
The Vision Transformer (ViT) demonstrates exceptional performance in various computer vision tasks. Attention is crucial for ViT to capture complex wide-ranging relationships among image patches, allowing the model to weigh the importance of image patches and aiding our understanding of the decision-making process. However, when utilizing the attention of ViT as evidence in high-stakes decision-making tasks such as medical diagnostics, a challenge arises due to the potential of attention mechanisms erroneously focusing on irrelevant regions. In this study, we propose a statistical test for ViT's attentions, enabling us to use the attentions as reliable quantitative evidence indicators for ViT's decision-making with a rigorously controlled error rate. Using the framework called selective inference, we quantify the statistical significance of attentions in the form of p-values, which enables the theoretically grounded quantification of the false positive detection probability of attentions. We demonstrate the validity and the effectiveness of the proposed method through numerical experiments and applications to brain image diagnoses.
△ Less
Submitted 19 January, 2024; v1 submitted 16 January, 2024;
originally announced January 2024.
-
Selective Inference for Changepoint detection by Recurrent Neural Network
Authors:
Tomohiro Shiraishi,
Daiki Miwa,
Vo Nguyen Le Duy,
Ichiro Takeuchi
Abstract:
In this study, we investigate the quantification of the statistical reliability of detected change points (CPs) in time series using a Recurrent Neural Network (RNN). Thanks to its flexibility, RNN holds the potential to effectively identify CPs in time series characterized by complex dynamics. However, there is an increased risk of erroneously detecting random noise fluctuations as CPs. The prima…
▽ More
In this study, we investigate the quantification of the statistical reliability of detected change points (CPs) in time series using a Recurrent Neural Network (RNN). Thanks to its flexibility, RNN holds the potential to effectively identify CPs in time series characterized by complex dynamics. However, there is an increased risk of erroneously detecting random noise fluctuations as CPs. The primary goal of this study is to rigorously control the risk of false detections by providing theoretically valid p-values to the CPs detected by RNN. To achieve this, we introduce a novel method based on the framework of Selective Inference (SI). SI enables valid inferences by conditioning on the event of hypothesis selection, thus mitigating selection bias. In this study, we apply SI framework to RNN-based CP detection, where characterizing the complex process of RNN selecting CPs is our main technical challenge. We demonstrate the validity and effectiveness of the proposed method through artificial and real data experiments.
△ Less
Submitted 25 November, 2023;
originally announced November 2023.
-
Bounded P-values in Parametric Programming-based Selective Inference
Authors:
Tomohiro Shiraishi,
Daiki Miwa,
Vo Nguyen Le Duy,
Ichiro Takeuchi
Abstract:
Selective inference (SI) has been actively studied as a promising framework for statistical hypothesis testing for data-driven hypotheses. The basic idea of SI is to make inferences conditional on an event that a hypothesis is selected. In order to perform SI, this event must be characterized in a traceable form. When selection event is too difficult to characterize, additional conditions are intr…
▽ More
Selective inference (SI) has been actively studied as a promising framework for statistical hypothesis testing for data-driven hypotheses. The basic idea of SI is to make inferences conditional on an event that a hypothesis is selected. In order to perform SI, this event must be characterized in a traceable form. When selection event is too difficult to characterize, additional conditions are introduced for tractability. This additional conditions often causes the loss of power, and this issue is referred to as over-conditioning in [Fithian et al., 2014]. Parametric programming-based SI (PP-based SI) has been proposed as one way to address the over-conditioning issue. The main problem of PP-based SI is its high computational cost due to the need to exhaustively explore the data space. In this study, we introduce a procedure to reduce the computational cost while guaranteeing the desired precision, by proposing a method to compute the lower and upper bounds of p-values. We also proposed three types of search strategies that efficiently improve these bounds. We demonstrate the effectiveness of the proposed method in hypothesis testing problems for feature selection in linear models and attention region identification in deep neural networks.
△ Less
Submitted 28 December, 2023; v1 submitted 21 July, 2023;
originally announced July 2023.
-
Valid P-Value for Deep Learning-Driven Salient Region
Authors:
Daiki Miwa,
Vo Nguyen Le Duy,
Ichiro Takeuchi
Abstract:
Various saliency map methods have been proposed to interpret and explain predictions of deep learning models. Saliency maps allow us to interpret which parts of the input signals have a strong influence on the prediction results. However, since a saliency map is obtained by complex computations in deep learning models, it is often difficult to know how reliable the saliency map itself is. In this…
▽ More
Various saliency map methods have been proposed to interpret and explain predictions of deep learning models. Saliency maps allow us to interpret which parts of the input signals have a strong influence on the prediction results. However, since a saliency map is obtained by complex computations in deep learning models, it is often difficult to know how reliable the saliency map itself is. In this study, we propose a method to quantify the reliability of a salient region in the form of p-values. Our idea is to consider a salient region as a selected hypothesis by the trained deep learning model and employ the selective inference framework. The proposed method can provably control the probability of false positive detections of salient regions. We demonstrate the validity of the proposed method through numerical examples in synthetic and real datasets. Furthermore, we develop a Keras-based framework for conducting the proposed selective inference for a wide class of CNNs without additional implementation cost.
△ Less
Submitted 6 January, 2023;
originally announced January 2023.
-
Recoil effects of photoelectrons in a solid
Authors:
Yasutaka Takata,
Yosuke Kayanuma,
Makina Yabashi,
Kenji Tamasaku,
Yoshinori Nishino,
Daigo Miwa,
Yoshihisa Harada,
Koji Horiba,
Shik Shin,
Satoshi Tanaka,
Eiji Ikenaga,
Keisuke Kobayashi,
Yasunori Senba,
Haruhiko Ohashi,
Tetsuya Ishikawa
Abstract:
High energy resolution C 1$s$ photoelectron spectra of graphite were measured at the excitation energy of 340, 870, 5950 and 7940eV using synchrotron radiation. On increasing the excitation energy, i.e., increasing kinetic energy of the photoelectron, the bulk origin C 1$s$ peak position shifts to higher binding energies. This systematic shift is due to the kinetic energy loss of the high-energy…
▽ More
High energy resolution C 1$s$ photoelectron spectra of graphite were measured at the excitation energy of 340, 870, 5950 and 7940eV using synchrotron radiation. On increasing the excitation energy, i.e., increasing kinetic energy of the photoelectron, the bulk origin C 1$s$ peak position shifts to higher binding energies. This systematic shift is due to the kinetic energy loss of the high-energy photoelectron by kicking the atom, and is clear evidence of the recoil effect in photoelectron emission. It is also observed that the asymmetric broadening increases for the higher energy photoelectrons. All these recoil effects can be quantified in the same manner as the Mössbauer effect for $γ$-ray emission from nuclei embedded in crystals.
△ Less
Submitted 25 January, 2007; v1 submitted 11 September, 2006;
originally announced September 2006.
-
Photoemission evidence for a Mott-Hubbard metal-insulator transition in VO$_2$
Authors:
R. Eguchi,
M. Taguchi,
M. Matsunami,
K. Horiba,
K. Yamamoto,
Y. Ishida,
A. Chainani,
Y. Takata,
M. Yabashi,
D. Miwa,
Y. Nishino,
K. Tamasaku,
T. Ishikawa,
Y. Senba,
H. Ohashi,
Y. Muraoka,
Z. Hiroi,
S. Shin
Abstract:
The temperature ($T$) dependent metal-insulator transition (MIT) in VO$_2$ is investigated using bulk sensitive hard x-ray ($\sim$ 8 keV) valence band, core level, and V 2$p-3d$ resonant photoemission spectroscopy (PES). The valence band and core level spectra are compared with full-multiplet cluster model calculations including a coherent screening channel. Across the MIT, V 3$d$ spectral weigh…
▽ More
The temperature ($T$) dependent metal-insulator transition (MIT) in VO$_2$ is investigated using bulk sensitive hard x-ray ($\sim$ 8 keV) valence band, core level, and V 2$p-3d$ resonant photoemission spectroscopy (PES). The valence band and core level spectra are compared with full-multiplet cluster model calculations including a coherent screening channel. Across the MIT, V 3$d$ spectral weight transfer from the coherent ($d^1\underbar{\it {C}}$ final) states at Fermi level to the incoherent ($d^{0}+d^1\underbar{\it {L}}$ final) states, corresponding to the lower Hubbard band, lead to gap-formation. The spectral shape changes in V 1$s$ and V 2$p$ core levels as well as the valence band are nicely reproduced from a cluster model calculations, providing electronic structure parameters. Resonant-PES finds that the $d^1\underbar{\it{L}}$ states resonate across the V 2$p-3d$ threshold in addition to the $d^{0}$ and $d^1\underbar{\it {C}}$ states. The results support a Mott-Hubbard transition picture for the first order MIT in VO$_2$.
△ Less
Submitted 30 July, 2008; v1 submitted 27 July, 2006;
originally announced July 2006.
-
Electronic structure of semiconducting CeFe$_4$P$_{12}$: Strong hybridization and relevance of single-impurity Anderson model
Authors:
M. Matsunami,
K. Horiba,
M. Taguchi,
K. Yamamoto,
A. Chainani,
Y. Takata,
Y. Senba,
H. Ohashi,
M. Yabashi,
K. Tamasaku,
Y. Nishino,
D. Miwa,
T. Ishikawa,
E. Ikenaga,
K. Kobayashi,
H. Sugawara,
H. Sato,
H. Harima,
S. Shin
Abstract:
Semiconducting skutterudite CeFe$_4$P$_{12}$ is investigated by synchrotron x-ray photoemission spectroscopy (PES) and x-ray absorption spectroscopy (XAS). Ce 3$d$ core-level PES and 3$d-4f$ XAS, in combination with single impurity Anderson model (SIAM) calculations, confirm features due to $f^0$, $f^1$ and $f^2$ configurations. The Ce 4$f$ density of states (DOS) indicates absence of a Kondo re…
▽ More
Semiconducting skutterudite CeFe$_4$P$_{12}$ is investigated by synchrotron x-ray photoemission spectroscopy (PES) and x-ray absorption spectroscopy (XAS). Ce 3$d$ core-level PES and 3$d-4f$ XAS, in combination with single impurity Anderson model (SIAM) calculations, confirm features due to $f^0$, $f^1$ and $f^2$ configurations. The Ce 4$f$ density of states (DOS) indicates absence of a Kondo resonance at Fermi level, but can still be explained by SIAM with a small gap in non-$f$ DOS. While Ce 4$f$ partial DOS from band structure calculations are also consistent with the main Ce 4$f$ DOS, the importance of SIAM for core and valence spectra indicates Kondo semiconducting mixed valence for CeFe$_4$P$_{12}$, derived from strong hybridization between non-$f$ conduction and Ce 4$f$ DOS.
△ Less
Submitted 6 January, 2006;
originally announced January 2006.
-
Evidence for suppressed metallicity on the surface of La2-xSrxCuO4 and Nd2-xCexCuO4
Authors:
M. Taguchi,
A. Chainani,
K. Horiba,
Y. Takata,
M. Yabashi,
K. Tamasaku,
Y. Nishino,
D. Miwa,
T. Ishikawa,
T. Takeuchi,
K. Yamamoto,
M. Matsunami,
S. Shin,
T. Yokoya,
E. Ikenaga,
K. Kobayashi,
T. Mochiku,
K. Hirata,
J. Hori,
K. Ishii,
F. Nakamura,
T. Suzuki
Abstract:
Hard X-ray Photoemission spectroscopy (PES) of copper core electronic states, with a probing depth of $\sim$60 Å, is used to show that the Zhang-Rice singlet feature is present in La$_2$CuO$_4$ but is absent in Nd$_2$CuO$_4$. Hole- and electron do** in La$_{2-x}$Sr$_x$CuO$_4$ (LSCO) and Nd$_{2-x}$Ce$_x$CuO$_4$ (NCCO) result in new well-screened features which are missing in soft X-ray PES. Imp…
▽ More
Hard X-ray Photoemission spectroscopy (PES) of copper core electronic states, with a probing depth of $\sim$60 Å, is used to show that the Zhang-Rice singlet feature is present in La$_2$CuO$_4$ but is absent in Nd$_2$CuO$_4$. Hole- and electron do** in La$_{2-x}$Sr$_x$CuO$_4$ (LSCO) and Nd$_{2-x}$Ce$_x$CuO$_4$ (NCCO) result in new well-screened features which are missing in soft X-ray PES. Impurity Anderson model calculations establish metallic screening as its origin, which is strongly suppressed within 15 $\textÅ$ of the surface. Complemented with X-ray absorption spectroscopy, the small chemical-potential shift in core levels ($\sim0.2$ eV) are shown to be consistent with modifications of valence and conduction band states spanning the band gap ($\sim1$ eV) upon hole- and electron-do** in LSCO and NCCO.
△ Less
Submitted 26 January, 2005;
originally announced January 2005.
-
Electronic Structure of Strained Manganite Thin Films with Room Temperature Ferromagnetism Investigated by Hard X-ray Photoemission Spectroscopy
Authors:
Hidekazu Tanaka,
Yasutaka Takata,
Koji Horiba,
Munetaka Taguchi,
Ashish Chainani,
Shik Shin,
Daigo Miwa,
Kenji Tamasaku,
Yoshinori Nishino,
Tetsuya Ishikawa,
Mitsuhiro Awaji,
Akihisa Takeuchi,
Tomoji Kawai,
Keisuke Kobayashi
Abstract:
We report the bulk sensitive Hard X-ray (hv=5.95keV) core level photoemission spectroscopy to investigate the intrinsic electronic structure of strained (La0.85Ba0.15)MnO3 thin films. In a 20nm thick well-strained film with strongly enhanced ferromagnetism, a new sharp satellite peak appeared at the low energy site of the Mn 2p3/2 main peak, whereas a broader signal was observed for the unstrain…
▽ More
We report the bulk sensitive Hard X-ray (hv=5.95keV) core level photoemission spectroscopy to investigate the intrinsic electronic structure of strained (La0.85Ba0.15)MnO3 thin films. In a 20nm thick well-strained film with strongly enhanced ferromagnetism, a new sharp satellite peak appeared at the low energy site of the Mn 2p3/2 main peak, whereas a broader signal was observed for the unstrained film with 300nm thickness. Cluster calculations revealed that the intensity corresponded to the density of the state at the Fermi level relating to the magnitude of the ferromagnetic order. The satellite intensity also agreed quantitatively with the square of the magnetization.
△ Less
Submitted 9 October, 2004;
originally announced October 2004.
-
Nature of Well-Screened State in Hard X-ray Mn 2$p$ Core-Level Photoemission of La$_{1-x}$Sr$_x$MnO$_3$ Films
Authors:
K. Horiba,
M. Taguchi,
A. Chainani,
Y. Takata,
E. Ikenaga,
H. Namatame,
M. Taniguchi,
M. Awaji,
A. Takeuchi,
D. Miwa,
Y. Nishino,
K. Tamasaku,
T. Ishikawa,
H. Kumigashira,
M. Oshima,
M. Lippmaa,
M. Kawasaki,
H. Koinuma,
K. Kobayashi,
S. Shin
Abstract:
Using hard x-ray (HX ; $hν$ = 5.95 keV) synchrotron photoemission spectroscopy (PES), we study the intrinsic electronic structure of La$_{1-x}$Sr$_x$MnO$_3$ (LSMO) thin films. Comparison of Mn 2$p$ core-levels with Soft x-ray (SX ; $hν$ $\sim$ 1000 eV) -PES shows a clear additional well-screened feature only in HX-PES. Take-off-angle dependent data indicate its bulk ($\ge$ 20 Å) character. The d…
▽ More
Using hard x-ray (HX ; $hν$ = 5.95 keV) synchrotron photoemission spectroscopy (PES), we study the intrinsic electronic structure of La$_{1-x}$Sr$_x$MnO$_3$ (LSMO) thin films. Comparison of Mn 2$p$ core-levels with Soft x-ray (SX ; $hν$ $\sim$ 1000 eV) -PES shows a clear additional well-screened feature only in HX-PES. Take-off-angle dependent data indicate its bulk ($\ge$ 20 Å) character. The do** and temperature dependence track the ferromagnetism and metallicity of the LSMO series. Cluster model calculations including charge transfer from do** induced states show good agreement, confirming this picture of bulk properties reflected in Mn 2$p$ core-levels using HX-PES.
△ Less
Submitted 21 October, 2004; v1 submitted 19 May, 2004;
originally announced May 2004.
-
Bulk screening in core level photoemission from Mott-Hubbard and Charge-Transfer systems
Authors:
M. Taguchi,
A. Chainani,
N. Kamakura,
K. Horiba,
Y. Takata,
E. Ikenaga,
T. Yokoya,
S. Shin,
K. Kobayashi,
K. Tamasaku,
Y. Nishino,
D. Miwa,
M. Yabashi,
T. Ishikawa,
T. Mochiku,
K. Hirata,
K. Motoya
Abstract:
We report bulk-sensitive hard X-ray ($hν$ = 5.95 KeV) core level photoemission spectroscopy (PES) of single crystal V$_{1.98}$Cr$_{0.02}$O$_{3}$ and the high-$T_c$ cuprate Bi$_2$Sr$_{2}$CaCu$_{2}$O$_{8+δ}$ (Bi2212). V$_{1.98}$Cr$_{0.02}$O$_{3}$ exhibits low binding energy "satellites" to the V $2p$ "main lines" in the metallic phase, which are suppressed in the antiferromagnetic insulator phase.…
▽ More
We report bulk-sensitive hard X-ray ($hν$ = 5.95 KeV) core level photoemission spectroscopy (PES) of single crystal V$_{1.98}$Cr$_{0.02}$O$_{3}$ and the high-$T_c$ cuprate Bi$_2$Sr$_{2}$CaCu$_{2}$O$_{8+δ}$ (Bi2212). V$_{1.98}$Cr$_{0.02}$O$_{3}$ exhibits low binding energy "satellites" to the V $2p$ "main lines" in the metallic phase, which are suppressed in the antiferromagnetic insulator phase. In contrast, the Cu $2p$ spectra of Bi2212 do not show temperature dependent features, but a comparison with soft X-ray PES indicates a large increase in the $2p^5 3d^9$ "satellites" or $3d^9$ weight in the bulk. Cluster model calculations, including full multiplet structure and a screening channel derived from the coherent band at the Fermi energy, give very satisfactory agreement with experiments.
△ Less
Submitted 23 January, 2005; v1 submitted 8 April, 2004;
originally announced April 2004.
-
Bulk Electronic structure of Na$_{0.35}$CoO$_{2}$.1.3H$_{2}$O
Authors:
A. Chainani,
T. Yokoya,
Y. Takata,
K. Tamasaku,
M. Taguchi,
T. Shimojima,
N. Kamakura,
K. Horiba,
S. Tsuda,
S. Shin,
D. Miwa,
Y. Nishino,
T. Ishikawa,
M. Yabashi,
K. Kobayashi,
H. Namatame,
M. Taniguchi,
K. Takada,
T. Sasaki,
H. Sakurai,
E. Takayama-Muromachi
Abstract:
High-energy (h$ν$ = 5.95 keV) synchrotron Photoemission spectroscopy (PES) is used to study bulk electronic structure of Na$_{0.35}$CoO$_{2}$.1.3H$_{2}$O, the layered superconductor. In contrast to 3-dimensional doped Co oxides, Co $\it{2p}$ core level spectra show well-separated Co$^{3+}$ and Co$^{4+}$ ions. Cluster calculations suggest low spin Co$^{3+}$ and Co$^{4+}$ character, and a moderate…
▽ More
High-energy (h$ν$ = 5.95 keV) synchrotron Photoemission spectroscopy (PES) is used to study bulk electronic structure of Na$_{0.35}$CoO$_{2}$.1.3H$_{2}$O, the layered superconductor. In contrast to 3-dimensional doped Co oxides, Co $\it{2p}$ core level spectra show well-separated Co$^{3+}$ and Co$^{4+}$ ions. Cluster calculations suggest low spin Co$^{3+}$ and Co$^{4+}$ character, and a moderate on-site Coulomb correlation energy U$_{dd}\sim$3-5.5 eV. Photon dependent valence band PES identifies Co $\it{3d}$ and O $\it{2p}$ derived states, in near agreement with band structure calculations.
△ Less
Submitted 26 February, 2004; v1 submitted 11 December, 2003;
originally announced December 2003.
