arXiv subjects

Naofumi Hama

Publications and source records attributed to Naofumi Hama.

8 recordsLinked to original sources

arXiv2023

Notes on Applicability of Explainable AI Methods to Machine Learning Models Using Features Extracted by Persistent Homology

Data analysis that uses the output of topological data analysis as input for machine learning algorithms has been the subject of extensive research. This approach offers a means of capturing the global structure of data. Persistent homology (PH), a common methodology within the field of TDA, has found wide-ranging applications in machine learning. One of the key reasons for the success of the PH-ML pipeline lies in the deterministic nature of feature extraction conducted through PH. The ability to achieve satisfactory levels of accuracy with relatively simple downstream machine learning models, when processing these extracted features, underlines the pipeline's superior interpretability. However, it must be noted that this interpretation has encountered issues. Specifically, it fails to accurately reflect the feasible parameter region in the data generation process, and the physical or chemical constraints that restrict this process. Against this backdrop, we explore the potential application of explainable AI methodologies to this PH-ML pipeline. We apply this approach to the specific problem of predicting gas adsorption in metal-organic frameworks and demonstrate that it can yield suggestive results. The codes to reproduce our results are available at https://github.com/naofumihama/xai_ph_ml

cs.LG↗

arXiv2023

Deletion and Insertion Tests in Regression Models

A basic task in explainable AI (XAI) is to identify the most important features behind a prediction made by a black box function $f$. The insertion and deletion tests of Petsiuk et al. (2018) can be used to judge the quality of algorithms that rank pixels from most to least important for a classification. Motivated by regression problems we establish a formula for their area under the curve (AUC) criteria in terms of certain main effects and interactions in an anchored decomposition of $f$. We find an expression for the expected value of the AUC under a random ordering of inputs to $f$ and propose an alternative area above a straight line for the regression setting. We use this criterion to compare feature importances computed by integrated gradients (IG) to those computed by Kernel SHAP (KS) as well as LIME, DeepLIFT, vanilla gradient and input$\times$gradient methods. KS has the best overall performance in two datasets we consider but it is very expensive to compute. We find that IG is nearly as good as KS while being much faster. Our comparison problems include some binary inputs that pose a challenge to IG because it must use values between the possible variable levels and so we consider ways to handle binary variables in IG. We show that sorting variables by their Shapley value does not necessarily give the optimal ordering for an insertion-deletion test. It will however do that for monotone functions of additive models, such as logistic regression.

cs.LG↗

arXiv2023

Model free variable importance for high dimensional data

A model-agnostic variable importance method can be used with arbitrary prediction functions. Here we present some model-free methods that do not require access to the prediction function. This is useful when that function is proprietary and not available, or just extremely expensive. It is also useful when studying residuals from a model. The cohort Shapley (CS) method is model-free but has exponential cost in the dimension of the input space. A supervised on-manifold Shapley method from Frye et al. (2020) is also model free but requires as input a second black box model that has to be trained for the Shapley value problem. We introduce an integrated gradient (IG) version of cohort Shapley, called IGCS, with cost $\mathcal{O}(nd)$. We show that over the vast majority of the relevant unit cube that the IGCS value function is close to a multilinear function for which IGCS matches CS. Another benefit of IGCS is that is allows IG methods to be used with binary predictors. We use some area between curves (ABC) measures to quantify the performance of IGCS. On a problem from high energy physics we verify that IGCS has nearly the same ABCs as CS does. We also use it on a problem from computational chemistry in 1024 variables. We see there that IGCS attains much higher ABCs than we get from Monte Carlo sampling. The code is publicly available at https://github.com/cohortshapley/cohortintgrad

cs.LG↗

arXiv2014

Supersymmetric Rényi Entropy in Five Dimensions

We introduce supersymmetric Rényi entropies for $\mathcal{N}=1$ supersymmetric gauge theories in five dimensions. The matrix model representation is obtained using the localization method and the large-$N$ behavior is studied. The gravity dual is a supersymmetric charged topological AdS$_6$ black hole in the Romans $F(4)$ supergravity. The variation of the supersymmetric Rényi entropy due to the insertion of a BPS Wilson loop is computed. We find perfect agreements between the large-$N$ and the dual gravity computations both with and without the Wilson loop operator.

hep-th↗

arXiv2013

AGT relation in the light asymptotic limit

It is known that the path integral of correlators in Liouville theory reduces to a finite dimensional integral in the limit of vanishing coupling b. We take the example of four-point functions on sphere and investigate how the simple integral expression is reproduced from the path integral of gauge theory on extremely squashed ellipsoids. The simplified form for correlators suggests there is a 2D gauge theory describing the limit.

hep-th↗

arXiv2012

Seiberg-Witten Theories on Ellipsoids

We present a set of equations for a 4D Killing spinor which guarantees the Seiberg-Witten theories on a curved background to be supersymmetric. The equations involve an SU(2) gauge field and some auxiliary fields in addition to the metric. Four-dimensional ellipsoids with U(1)xU(1) isometry are shown to admit a supersymmetry if these additional fields are chosen appropriately. We compute the partition function of general Seiberg-Witten theories on ellipsoids, and the result suggests a correspondence with 2D Liouville or Toda correlators with general coupling constant b.

hep-th↗

arXiv2011

Notes on SUSY Gauge Theories on Three-Sphere

We extend the formula for partition functions of N=2 superconformal gauge theories on S^3 obtained recently by Kapustin, Willett and Yaakov, to incorporate matter fields with arbitrary R-charge assignments. We use the result to check that the self-mirror property of N=4 SQED with two electron hypermultiplets is preserved under a certain mass deformation which breaks the supersymmetry to N=2.

hep-th↗

arXiv2011

SUSY Gauge Theories on Squashed Three-Spheres

We study Euclidean 3D N=2 supersymmetric gauge theories on squashed three-spheres preserving isometries SU(2) x U(1) or U(1) x U(1). We show that, when a suitable background U(1) gauge field is turned on, these squashed spheres support charged Killing spinors and therefore N=2 supersymmetric gauge theories. We present the Lagrangian and supersymmetry rules for general gauge theories. The partition functions are computed using localization principle, and are expressed as integrals over Coulomb branch. For the squashed sphere with U(1) x U(1) isometry, its measure and integrand are identified with the building blocks of structure constants in Liouville or Toda conformal field theories with b \neq 1.

hep-th↗