In materials informatics, feature selection and model selection are utilized in the knowledge extraction process, and the confidence evaluation of the selection result is crucial for ensuring the reliability of the extracted knowledge. In this study, we propose a novel method to quantitatively evaluate the significance of low-dimensional models obtained from an expanded feature space using the density of states (DoS) for the evaluation metrics. This method allows us to compare the performance of the selected model to those of other models and evaluate its significance. We further propose an evaluation method for feature importance based on marginal posterior probabilities using the Bayesian model averaging (BMA) framework, which considers all models. We demonstrate the effectiveness of our proposed methods through their application to a crystal structure dataset. Our results show that the DoS analysis reveals the presence of a large number of models with a comparable performance with the best model reported in a previous research work. This suggests that knowledge extracted from a single selected model can be unreliable and that considering other models is crucial. We also show that the BMA-based feature importance evaluation provides valuable insights into the importance of features, highlighting both primary features and functional forms. In addition, we demonstrate that the LASSO + L0 method commonly used in existing research deteriorates the search space for model selection. Our findings suggest that our proposed methods provide valuable tools for assessing the significance of low-dimensional models and extracting knowledge from large-scale data in materials informatics.

Neural operators have recently grown in popularity as Partial Differential Equation (PDE) surrogate models. Learning solution functionals, rather than functions, has proven to be a powerful approach to calculate fast, accurate solutions to complex PDEs. While much work has been performed evaluating neural operator performance on a wide variety of surrogate modeling tasks, these works normally evaluate performance on a single equation at a time. In this work, we develop a novel contrastive pretraining framework utilizing generalized contrastive loss that improves neural operator generalization across multiple governing equations simultaneously. Governing equation coefficients are used to measure ground-truth similarity between systems. A combination of physics-informed system evolution and latent-space model output is anchored to input data and used in our distance function. We find that physics-informed contrastive pretraining improves accuracy for the Fourier neural operator in fixed-future and autoregressive rollout tasks for the 1D and 2D heat, Burgers’, and linear advection equations.

In recent years, neural networks have been used to solve phase retrieval problems in imaging with superior accuracy and speed than traditional techniques, especially in the presence of noise. However, in the context of interferometric imaging, phase noise has been largely unaddressed by existing neural network architectures. Such noise arises naturally in an interferometer due to mechanical instabilities or atmospheric turbulence, limiting measurement acquisition times and posing a challenge in scenarios with limited light intensity, such as remote sensing. Here, we introduce a 3D–2D Phase Retrieval U-Net (PRUNe) that takes noisy and randomly phase-shifted interferograms as inputs and outputs a single 2D phase image. A 3D downsampling convolutional encoder captures correlations within and between frames to produce a 2D latent space, which is upsampled by a 2D decoder into a phase image. We test our model against a state-of-the-art singular value decomposition algorithm and find PRUNe reconstructions consistently show more accurate and smooth reconstructions, with a ×2.5–4 lower mean squared error at multiple signal-to-noise ratios for interferograms with low (<1 photon/pixel) and high (∼100 photons/pixel) signal intensity. Our model presents a faster and more accurate approach to perform phase retrieval in extremely low light intensity interferometry in the presence of phase noise and will find application in other multi-frame noisy imaging techniques.

Predicting the thermal conductivity of two-dimensional (2D) heterostructures is challenging and cannot be adequately resolved using conventional computational approaches. To address this challenge, we propose a new and efficient approach that combines first-principles density functional theory (DFT) calculations with a machine-learning interatomic potential (MLIP) methodology to determine the thermal conductivity of a novel 2D van der Waals TiS2/MoS2 heterostructure. We leverage the proposed approach to estimate the thermal conductivities of TiS2/MoS2 heterostructures as well as bilayer-TiS2 and bilayer-MoS2. A unique aspect of this approach is the combined implementation of the moment tensor potential for short-range (intralayer) interactions and the D3-dispersion correction scheme for long-range (interlayer) van der Waals interactions. This approach employs relatively inexpensive computational DFT-based datasets generated from ab initio molecular dynamics simulations to accurately describe the interatomic interactions in the bilayers. The thermal conductivities of the bilayers exhibit the following trend: bilayer-TiS2 > bilayer-MoS2 > the TiS2/MoS2 heterostructure. In addition, this work makes the case that the 2D bilayers exhibit considerably higher thermal conductivities than bulk graphite, a common battery anode material, indicating the potential to utilize 2D heterostructures in thermal management applications and energy storage devices. Furthermore, the MLIP-based methodology provides a reliable approach for estimating the thermal conductivity of bilayers and heterostructures.

Electrical waves in the heart form rotating spiral or scroll waves during life-threatening arrhythmias, such as atrial or ventricular fibrillation. The wave dynamics are typically modeled using coupled partial differential equations, which describe reaction-diffusion dynamics in excitable media. More recently, data-driven generative modeling has emerged as an alternative to generate spatio-temporal patterns in physical and biological systems. Here, we explore denoising diffusion probabilistic models for the generative modeling of electrical wave patterns in cardiac tissue. We trained diffusion models with simulated electrical wave patterns to be able to generate such wave patterns in unconditional and conditional generation tasks. For instance, we explored the diffusion-based (i) parameter-specific generation, (ii) evolution, and (iii) inpainting of spiral wave dynamics, including reconstructing three-dimensional scroll wave dynamics from superficial two-dimensional measurements. Furthermore, we generated arbitrarily shaped bi-ventricular geometries and simultaneously initiated scroll wave patterns inside these geometries using diffusion. We characterized and compared the diffusion-generated solutions to solutions obtained with corresponding biophysical models and found that diffusion models learn to replicate spiral and scroll wave dynamics so well that they could be used for data-driven modeling of excitation waves in cardiac tissue. For instance, an ensemble of diffusion-generated spiral wave dynamics exhibits similar self-termination statistics as the corresponding ensemble simulated with a biophysical model. However, we also found that diffusion models produce artifacts if training data are lacking, e.g., during self-termination, and "hallucinate" wave patterns when insufficiently constrained.

We report a mixture of expert strategy to create fine-tuned large language models using a deep layer-wise token-level approach based on low-rank adaptation (LoRA). Starting with a set of pre-trained LoRA adapters, our gating strategy uses the hidden states to dynamically mix adapted layers, allowing the resulting X-LoRA model to draw upon different capabilities and create never-before-used deep layer-wise combinations to solve tasks. The design is inspired by the biological principles of universality and diversity, where neural network building blocks are reused in different hierarchical manifestations. Hence, the X-LoRA model can be easily implemented for any existing large language model without a need for modifications of the underlying structure. We develop a tailored X-LoRA model that offers scientific capabilities, including forward/inverse analysis tasks and enhanced reasoning capability, focused on biomaterial analysis, protein mechanics, and design. The impact of this work includes access to readily expandable and adaptable models with strong domain knowledge and the capability to integrate across areas of knowledge. Featuring experts in biology, mathematics, reasoning, bio-inspired materials, mechanics and materials, chemistry, protein biophysics, mechanics, and quantum-mechanics based molecular properties, we conduct a series of physics-focused case studies. We examine knowledge recall, protein mechanics forward/inverse tasks, protein design, adversarial agentic modeling including ontological knowledge graph construction, and molecular design. The model is capable not only of making quantitative predictions of nanomechanical properties of proteins or quantum mechanical molecular properties but also reasoning over the results and correctly predicting likely mechanisms that explain distinct molecular behaviors.

Machine learning algorithms can enhance the design and experimental processing of solar cells, resulting in increased conversion efficiency. In this study, we introduce a novel machine learning-based methodology for optimizing the Pareto front of four-terminal (4T) perovskite-copper indium selenide (CIS) tandem solar cells (TSCs). By training a neural network using the Bayesian regularization-backpropagation algorithm via Hammersley sampling, we achieve high prediction accuracy when testing with unseen data through random sampling. This surrogate model not only reduces computational costs but also potentially enhances device performance, increasing from 29.4% to 30.4% while simultaneously reducing material costs for fabrication by 50%. Comparing experimentally fabricated cells with the predicted optimal cells, the latter show a thinner front contact electrode, charge-carrier transport layer, and back contact electrode. Highly efficient perovskite cells identified from the Pareto front have a perovskite layer thickness ranging from 420 to 580 nm. Further analysis reveals the front contact electrode needs to be thin, while the back contact electrode can have a thickness ranging from 100 to 145 nm and still achieve high efficiency. The charge-carrier transport layers play a crucial role in minimizing interface recombination and ensuring unidirectional current flow. The optimal design space suggests thinner electron and hole transport layer thicknesses of 7 nm, down from 23 to 10 nm, respectively. It indicates a balanced charge-carrier extraction is crucial for an optimized perovskite cell. Overall, the presented methodology and optimized design parameters have the potential to enhance the performance of 4T perovskite/CIS TSC while reducing material fabrication costs.

The evaluation of well integrity in double-casing wells is critical for ensuring well stability, preventing oil and gas leaks, avoiding pollution, and ensuring safety throughout well development and production. However, the current predominant method of assessing cementing quality primarily focuses on single-casing wells, with limited work conducted on double-casing wells. This study introduces a novel approach for evaluating the cementing quality using the Inception module of convolutional neural networks. First, the finite-difference method is employed to generate borehole sonic data corresponding to a variety of model configurations, which are used to train a neural network that learns spatial features from the borehole sonic data to reconstruct the slowness model. By adjusting the network architecture and parameters, it is discovered that a neural network with two blocks and 4096 nodes in the fully connected layer demonstrated the best imaging results and exhibited strong anti-noise capabilities. The proposed method is validated using practical wellbore size models, demonstrating excellent results and offering a more effective means of evaluating wellbore integrity in double-casing wells. In addition, dipole acoustic logging data are used to conduct slowness model imaging of the compressional (P-) wave and shear (S-) wave in double-casing wells to verify the feasibility of cementing quality evaluation. The developed method contributes to more accurate evaluations of wellbore integrity for the oil and gas industry, leading to improved safety and environmental outcomes.

Memristor crossbar arrays are expected to achieve highly energy-efficient neuromorphic computing via implementing parallel vector–matrix multiplication (VMM) in situ. The similarities between memristors and neural synapses offer opportunities for realizing hardware-based brain-inspired computing, such as spike neural networks. However, the nonlinear I–V characteristics of the memristors limit the implementation of parallel VMM on passive memristor crossbar arrays. In our work, we propose to utilize differential conductance as a synaptic weight to implement linear VMM operations on a passive memristor array in parallel. We fabricated a TiO2/HfO2 memristor crossbar array, in which differential-conductance-based synaptic weight exhibits plasticity, nonvolatility, multi-states, and tunable ON/OFF ratio. The noise-dependent accuracy performance of VMM operations based on the proposed approach was evaluated, offering an optimization guideline. Furthermore, we demonstrated a spike neural network circuit capable of processing small spiking signals through the differential-conductance-based synapses. The experimental results showcase effective space-coded and time-coded spike pattern recognition. Importantly, our work opens up new possibilities for the development of passive memristor arrays, leading to increased energy and area efficiency in brain-inspired chips.

Perovskite-based optoelectronic devices have emerged as a promising energy source due to their potential for scalable production. This study introduces “perovskene,” a novel class of 2D materials derived from the ABC3-like perovskites, synthesized via a data-driven, high-throughput computational strategy. We harness machine learning and multitarget deep neural networks to systematically investigate the structure–property relations, paving the way for targeted material design and optimization in fields such as renewable energy, electronics, and catalysis. The characterization of over 1500 synthesized structures shows that more than 500 structures are stable, revealing properties such as ultra-low work function and large magnetic moment, underscoring the potential for advanced technological applications.