Search

Despite increasing interest in the field of Interpretable Machine Learning (IML), a significant gap persists between the technical objectives targeted by researchers' methods and the high-level goals of consumers' use cases. In this work, we synthesize foundational work on IML methods and evaluation into an actionable taxonomy. This taxonomy serves as a tool to conceptualize the gap between researchers and consumers, illustrated by the lack of connections between its methods and use cases components. It also provides the foundation from which we describe a three-step workflow to better enable researchers and consumers to work together to discover what types of methods are useful for what use cases. Eventually, by building on the results generated from this workflow, a more complete version of the taxonomy will increasingly allow consumers to find relevant methods for their target use cases and researchers to identify applicable use cases for their proposed methods., Comment: Presented at ICML HILL Workshop 2021

We report a polarized Raman scattering study of the lattice dynamics of $\beta$-Li$_2$IrO$_3$ under hydrostatic pressures up to 7.62 GPa. At ambient pressure, $\beta$-Li$_2$IrO$_3$ exhibits the hyperhoneycomb crystal structure and a magnetically ordered state of spin-orbit entangled Jeff = 1/2 moments that is strongly influenced by bond-directional (Kitaev) exchange interactions. At a critical pressure of ~ 4.1 GPa, the phonon spectrum changes abruptly consistent with the reported structural transition into a monoclinic, dimerized phase. A comparison to the phonon spectra obtained from density functional calculations shows reasonable overall agreement. The calculations also indicate that the high-pressure phase is a nonmagnetic insulator driven by the formation of Ir-Ir dimer bonds. Our results thus indicate a strong sensitivity of the electronic properties of $\beta$-Li$_2$IrO$_3$ to the pressure-induced structural transition., Comment: 16 pages, 7 figures, 5 tables, Supplemental Material included

Finding new materials with antiferromagnetic (AFM) Kitaev interaction is an urgent issue to broaden and enrich the quantum magnetism research significantly. By carrying out inelastic neutron scattering experiments and subsequent analysis, we conclude that Na$_3$Co$_2$SbO$_6$ and Na$_2$Co$_2$TeO$_6$ are new honeycomb cobalt-based AFM Kitaev systems. The spin-orbit excitons at 20-28~meV in both compounds strongly supports the idea that Co$^{2+}$ ions of both compounds have a spin-orbital entangled $J_\rm{eff}=1/2$ state. Furthermore, we found that a generalized Kitaev-Heisenberg Hamiltonian can well describe the spin-wave excitations of both compounds with additional 3rd nearest-neighbor interaction. Our best-fit parameters show large AFM Kitaev terms and off-diagonal symmetric anisotropy terms of a similar magnitude in both compounds. We should stress that our parameters' optimized magnetic structures are consistent with the magnetic structures reported from neutron diffraction studies. Moreover, there is also the magnon-damping effect at the higher energy part of the spin waves, as usually observed in other Kitaev magnets. We demonstrate that Na$_3$Co$_2$SbO$_6$ and Na$_2$Co$_2$TeO$_6$ are the first experimental realization of AFM Kitaev magnets based on the systematic studies of the spin waves and analysis., Comment: 28 pages, 10 figures, Journal of Physics: Condensed Matter

The min-entropy is a widely used metric to quantify the randomness of generated random numbers in cryptographic applications; it measures the difficulty of guessing the most likely output. An important min-entropy estimator is the compression estimator of NIST Special Publication (SP) 800-90B, which relies on Maurer's universal test. In this paper, we propose two kinds of min-entropy estimators to improve computational complexity and estimation accuracy by leveraging two variations of Maurer's test: Coron's test (for Shannon entropy) and Kim's test (for Renyi entropy). First, we propose a min-entropy estimator based on Coron's test. It is computationally more efficient than the compression estimator while maintaining the estimation accuracy. The secondly proposed estimator relies on Kim's test that computes the Renyi entropy. This estimator improves estimation accuracy as well as computational complexity. We analytically characterize the bias-variance tradeoff, which depends on the order of Renyi entropy. By taking into account this tradeoff, we observe that the order of two is a proper assignment and focus on the min-entropy estimation based on the collision entropy (i.e., Renyi entropy of order two). The min-entropy estimation from the collision entropy can be described by a closed-form solution, whereas both the compression estimator and the proposed estimator based on Coron's test do not have closed-form solutions. By leveraging the closed-form solution, we also propose a lightweight estimator that processes data samples in an online manner. Numerical evaluations demonstrate that the first proposed estimator achieves the same accuracy as the compression estimator with much less computation. The proposed estimator based on the collision entropy can even improve the accuracy and reduce the computational complexity.

The magnetic excitations in CoPS$_3$, a two-dimensional van der Waals (vdW) antiferromagnet with spin $S=3/2$ on a honeycomb lattice, has been measured using powder inelastic neutron scattering. Clear dispersive spin waves are observed with a large spin gap of ~13 meV. The magnon spectra were fitted using an $XXZ$-type $J_1-J_2-J_3$ Heisenberg Hamiltonian with a single-ion anisotropy assuming no magnetic exchange between the honeycomb layers. The best-fit parameters show ferromagnetic exchange $J_1=-2.08$ meV and $J_2=-0.26$ meV for the nearest and second-nearest neighbors and a sizeable antiferromagnetic exchange $J_3=4.21$ meV for the third-nearest neighbor with the strong easy-axis anisotropy $K=-2.06$ meV. The suitable fitting could only be achieved by the anisotropic $XXZ$-type Hamiltonian, in which the exchange interaction for the out-of-plane component is smaller than that for the in-plane one by a ratio $\alpha=J_z/J_x=0.6$. Moreover, the absence of spin-orbit exciton around 30 meV indicates that Co$^{2+}$ ions in CoPS$_3$ have a $S=3/2$ state rather than a spin-orbital entangled $J_\rm{eff}=1/2$ ground state. Our result directly shows that CoPS$_3$ is an experimental realization of the $XXZ$ model with a honeycomb lattice in 2D vdW magnets., Comment: 17 pages, 5 figures

We have carried out neutron diffraction and small angle neutron scattering measurements on a high quality single crystal of the cubic lacunar spinel multiferroic, GaV$_4$S$_8$, as a function of magnetic field and temperature to determine the magnetic properties for the single electron that is located on the tetrahedrally coordinated V$_4$ molecular unit. Our results are in good agreement with the structural transition at 44 K from cubic to rhombohedral symmetry where the system becomes a robust ferroelectric, while long range magnetic order develops below 13 K in the form of an incommensurate cycloidal magnetic structure, which can transform into a N\'eel-type skyrmion phase in a modest applied magnetic field. Below 5.9(3) K, the crystal enters a ferromagnetic phase, and we find the magnetic order parameter indicates a long range ordered ground state with an ordered moment of 0.23(1) ${\mu}_\mathrm{B}$ per V ion. Both polarized and unpolarized neutron data in the ferroelectric-paramagnetic phase have been measured to determine the magnetic form factor. The data are consistent with a model of the single spin being uniformly distributed across the V$_4$ molecular unit, rather than residing on the single apical V ion, in substantial agreement with the results of first-principles theory. In the magnetically ordered state, polarized neutron measurements are important since both the cycloidal and ferromagnetic order parameters are clearly coupled to the ferroelectricity, causing the structural peaks to be temperature and field dependent. For the ferromagnetic ground state, the spins are locked along the $[1,1,1]$ direction by a surprisingly large anisotropy.

Group fairness, a class of fairness notions that measure how different groups of individuals are treated differently according to their protected attributes, has been shown to conflict with one another, often with a necessary cost in loss of model's predictive performance. We propose a general diagnostic that enables systematic characterization of these trade-offs in group fairness. We observe that the majority of group fairness notions can be expressed via the fairness-confusion tensor, which is the confusion matrix split according to the protected attribute values. We frame several optimization problems that directly optimize both accuracy and fairness objectives over the elements of this tensor, which yield a general perspective for understanding multiple trade-offs including group fairness incompatibilities. It also suggests an alternate post-processing method for designing fair classifiers. On synthetic and real datasets, we demonstrate the use cases of our diagnostic, particularly on understanding the trade-off landscape between accuracy and fairness., Comment: Accepted to International Conference on Machine Learning (ICML 2020)

This article covers the construction of exponential attractors in two different functional space settings; one is in Hilbert's space, and the other is in the Banach space. The former relies on the squeezing properties of solution trajectories, but the latter does not. We present these different methods for constructing exponential attractors using the three-dimensional Rotating Navier-Stokes-$\alpha$ equations.

We propose PLLay, a novel topological layer for general deep learning models based on persistence landscapes, in which we can efficiently exploit the underlying topological features of the input data structure. In this work, we show differentiability with respect to layer inputs, for a general persistent homology with arbitrary filtration. Thus, our proposed layer can be placed anywhere in the network and feed critical information on the topological features of input data into subsequent layers to improve the learnability of the networks toward a given task. A task-optimal structure of PLLay is learned during training via backpropagation, without requiring any input featurization or data preprocessing. We provide a novel adaptation for the DTM function-based filtration, and show that the proposed layer is robust against noise and outliers through a stability analysis. We demonstrate the effectiveness of our approach by classification experiments on various datasets., Comment: 29 pages, 7 figures

We study the thermodynamic and high-magnetic-field properties of the magnetic insulator Ba$_5$CuIr$_3$O$_{12}$, which shows no magnetic order down to 2 K consistent with a spin liquid ground state. While the temperature dependence of the magnetic susceptibility and the specific heat shows only weak antiferromagnetic correlations, we find that the magnetization does not saturate up to a field of 59 Tesla, leading to an apparent contradiction. We demonstrate that the paradox can be resolved, and all of the experimental data can be consistently described within the framework of random singlet states. We demonstrate a generic procedure to derive the exchange coupling distribution $P(J)$ from the magnetization measurements and use it to show that the experimental data is consistent with the power-law form $P(J)\sim J^{-\alpha}$ with $\alpha \approx 0.6 $. Thus, we reveal that high-magnetic-field measurements can be essential to discern quantum spin liquid candidates from disorder dominated states that do not exhibit long-range order.

In practical applications of machine learning, it is necessary to look beyond standard metrics such as test accuracy in order to validate various qualitative properties of a model. Partial dependence plots (PDP), including instance-specific PDPs (i.e., ICE plots), have been widely used as a visual tool to understand or validate a model. Yet, current PDPs suffer from two main drawbacks: (1) a user must manually sort or select interesting plots, and (2) PDPs are usually limited to plots along a single feature. To address these drawbacks, we formalize a method for automating the selection of interesting PDPs and extend PDPs beyond showing single features to show the model response along arbitrary directions, for example in raw feature space or a latent space arising from some generative model. We demonstrate the usefulness of our automated dependence plots (ADP) across multiple use-cases and datasets including model selection, bias detection, understanding out-of-sample behavior, and exploring the latent space of a generative model., Comment: In Uncertainty in Artificial Intelligence (UAI 2020). Camera-ready version. Code is available at https://github.com/davidinouye/adp