Nam Q. Le

Alexander New, Michael J. Pekala, Nam Q. Le et al.

Properties of interest for crystals and molecules, such as band gap, elasticity, and solubility, are generally related to each other: they are governed by the same underlying laws of physics. However, when state-of-the-art graph neural networks attempt to predict multiple properties simultaneously (the multi-task learning (MTL) setting), they frequently underperform a suite of single property predictors. This suggests graph networks may not be fully leveraging these underlying similarities. Here we investigate a potential explanation for this phenomenon: the curvature of each property's loss surface significantly varies, leading to inefficient learning. This difference in curvature can be assessed by looking at spectral properties of the Hessians of each property's loss function, which is done in a matrix-free manner via randomized numerical linear algebra. We evaluate our hypothesis on two benchmark datasets (Materials Project (MP) and QM8) and consider how these findings can inform the training of novel multi-task learning models.

Alexander New, Michael Pekala, Elizabeth A. Pogue et al.

Generative machine learning models can use data generated by scientific modeling to create large quantities of novel material structures. Here, we assess how one state-of-the-art generative model, the physics-guided crystal generation model (PGCGM), can be used as part of the inverse design process. We show that the default PGCGM's input space is not smooth with respect to parameter variation, making material optimization difficult and limited. We also demonstrate that most generated structures are predicted to be thermodynamically unstable by a separate property-prediction model, partially due to out-of-domain data challenges. Our findings suggest how generative models might be improved to enable better inverse design.

Edward W. Staley, Tom Arbaugh, Michael Pekala et al.

Modern generative machine learning (ML) models can propose novel inorganic crystalline materials with targeted properties; however, synthesis planning of these materials remains difficult due to the complexity of the associated physical processes and limited availability of computational tools. We introduce a novel hybrid framework to evaluate Large Language Models (LLMs) in inorganic synthesis planning by combining thermodynamic databases with simplified kinetics models to approximate realistic synthesis conditions. As a case study, we focus on the niobium-oxygen system, which features multiple industrially relevant oxide phases with well-characterized data. In computational simulations, we compare LLM-generated synthesis routes with classical path-planning algorithms, showing that the implicit priors in LLMs can yield more viable strategies. In our evaluation setting, classical search methods serve primarily as a foil rather than a direct competitor. This illustrates the relative complexity of the problem and highlights where the LLM's implicit priors add value.

Alexander New, Nam Q. Le, Michael J. Pekala et al.

Accurate prediction of the properties of crystalline materials is crucial for targeted discovery, and this prediction is increasingly done with data-driven models. However, for many properties of interest, the number of materials for which a specific property has been determined is much smaller than the number of known materials. To overcome this disparity, we propose a novel self-supervised learning (SSL) strategy for material property prediction. Our approach, crystal denoising self-supervised learning (CDSSL), pretrains predictive models (e.g., graph networks) with a pretext task based on recovering valid material structures when given perturbed versions of these structures. We demonstrate that CDSSL models out-perform models trained without SSL, across material types, properties, and dataset sizes.

Nam Le, Jean-Marc Odobez

Learning speaker turn embeddings has shown considerable improvement in situations where conventional speaker modeling approaches fail. However, this improvement is relatively limited when compared to the gain observed in face embedding learning, which has been proven very successful for face verification and clustering tasks. Assuming that face and voices from the same identities share some latent properties (like age, gender, ethnicity), we propose three transfer learning approaches to leverage the knowledge from the face domain (learned from thousands of images and identities) for tasks in the speaker domain. These approaches, namely target embedding transfer, relative distance transfer, and clustering structure transfer, utilize the structure of the source face embedding space at different granularities to regularize the target speaker turn embedding space as optimizing terms. Our methods are evaluated on two public broadcast corpora and yield promising advances over competitive baselines in verification and audio clustering tasks, especially when dealing with short speaker utterances. The analysis of the results also gives insight into characteristics of the embedding spaces and shows their potential applications.