Search arXivSearch

arXiv · 1901.02717

Reliable and Explainable Machine Learning Methods for Accelerated Material Discovery

Abstract

Material scientists are increasingly adopting the use of machine learning (ML) for making potentially important decisions, such as, discovery, development, optimization, synthesis and characterization of materials. However, despite ML's impressive performance in commercial applications, several unique challenges exist when applying ML in materials science applications. In such a context, the contributions of this work are twofold. First, we identify common pitfalls of existing ML techniques when learning from underrepresented/imbalanced material data. Specifically, we show that with imbalanced data, standard methods for assessing quality of ML models break down and lead to misleading conclusions. Furthermore, we found that the model's own confidence score cannot be trusted and model introspection methods (using simpler models) do not help as they result in loss of predictive performance (reliability-explainability trade-off). Second, to overcome these challenges, we propose a general-purpose explainable and reliable machine-learning framework. Specifically, we propose a novel pipeline that employs an ensemble of simpler models to reliably predict material properties. We also propose a transfer learning technique and show that the performance loss due to models' simplicity can be overcome by exploiting correlations among different material properties. A new evaluation metric and a trust score to better quantify the confidence in the predictions are also proposed. To improve the interpretability, we add a rationale generator component to our framework which provides both model-level and decision-level explanations. Finally, we demonstrate the versatility of our technique on two applications: 1) predicting properties of crystalline compounds, and 2) identifying novel potentially stable solar cell materials.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Bhavya Kailkhura, Brian Gallagher, Sookyung Kim, Anna Hiszpanski, T. Yong-Jin Han. 2019-03-08. Reliable and Explainable Machine Learning Methods for Accelerated Material Discovery. https://arxiv.org/abs/1901.02717

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

Reduction of finite-size effects for second-order Møller-Plesset perturbation theory with singularity subtraction

Second-order Moller--Plesset perturbation theory (MP2) for periodic systems suffers from finite-size errors (FSEs) that have inverse volume scaling due to the Coulomb kernel singularity in reciprocal space. This error scaling limits the routine applicability of MP2 and other methods that rely on it to real materials, requiring prohibitively dense k-point meshes for convergence toward the thermodynamic limit (TDL). We introduce MP2 singularity subtraction (MP2SS), a systematic approach that applies the singularity subtraction strategy to reduce MP2 FSEs. The method employs auxiliary functions and fitting procedures that consider both the singularities present at the origin in reciprocal space and also the discontinuities in the MP2 structure factor that arise from finite k-point sampling. We present three possible MP2SS configurations (Gaussian, exponential, and tuned) which use different combinations of decay functions and demonstrate their performance for gapped systems. Across a range of extended solids and molecular crystals at large basis sets, MP2SS reduces the prefactor in MP2 FSEs and subsequently the computational resources required to converge to the TDL. MP2SS with the Gaussian auxiliary function provides the most robust corrections. Our results establish singularity subtraction as a powerful and flexible approach for mitigating finite-size errors in periodic correlation methods and provide a foundation for extending the technique to higher-order perturbation theories and other post-Hartree-Fock (HF) methods.

physics.comp-ph

Mechanical resilience of ultra-low-density racing-shoe foams

High-performance racing shoes rely on ultra-low-density elastomeric foams that undergo large, repeated deformations during running. Yet little is known about how their mechanical properties vary throughout the shoe or change with repeated use. Here, we characterize the midsole foam in an elite-level racing shoe from the heel, midfoot, and toe of a new shoe and a shoe worn for 300 miles. Microscopy reveals a characteristic pore length scale of 128+/-18 um and supports an approximately isotropic continuum description. We then quantify the mechanical response under tension, compression, and shear. Remarkably, despite 300 miles of real-world use, the foam retains its mechanical response across all three loading modes and shoe regions. Energy return remains largely unchanged, with values of 85-93% in tension and compression and 64-71% in shear. At the same time, we observe strong regional variations, with tensile and compressive stiffnesses 38-51% lower in the toe than in the heel. The foam also exhibits a pronounced tension-compression asymmetry in Poisson's ratio. Together, these findings reveal a spatially structured and mode-dependent mechanical response that remains largely preserved after 300 miles of use. This mechanical resilience may extend the functional lifetime of racing shoes, with implications for runners, replacement recommendations, and sustainability.

physics.comp-ph

Fundamental Bounds on the Polarizability of Macroscopic Scatterers

Polarizability predicts how an object responds to an incident electromagnetic field, the interactions between small particles, or the optical forces exerted upon them. Polarizability is responsible for the effective-medium properties of artificial materials or metasurfaces. Despite significant progress in all these areas, it is unclear what the limits of the strength of such interactions are or, more specifically, what the upper bounds on the polarizability of a given spatial region that an unknown and designed particle would occupy are. This work connects the electromagnetic field description via an integral equation with a dual formulation of quadratic programming to derive fundamental bounds on components of all four polarizability tensors or on their specific combinations. In particular, the work establishes an intuitive visualization of what strong polarizability means and how strong it can be. The developed fundamental bound also answers which materials and domains are best for the given demands on polarizability. These findings establish a versatile platform that can accommodate a wide range of demands on polarizable bodies, providing an absolute measure of their performance against which the results of human-powered or automated design procedures can be compared.

physics.comp-ph