Bias Free Multiobjective Active Learning for Materials Design and Discovery

Abstract

<div>The design rules for materials are clear for applications with a single objective. For most applications, however, there are often multiple, sometimes competing objectives where there is no single best material, and the design rules change to finding the set of Pareto optimal materials. </div><div>In this work, we introduce an active learning algorithm that directly uses the Pareto dominance relation to compute the set of Pareto optimal materials with desirable accuracy. <br></div><div>We apply our algorithm to de novo polymer design with a prohibitively large search space.</div><div>Using molecular simulations, we compute key descriptors for dispersant applications and reduce the number of materials that need to be evaluated to reconstruct the Pareto front with a desired confidence by over 98% compared to random search.</div><div>This work showcases how simulation and machine learning techniques can be coupled to discover materials within a design space that would be intractable using conventional screening approaches.</div>

Quantum Mechanics Enables "Freedom of Design" in Molecular Property Space

“…A deeper analysis of each front also revealed a series of small and rational changes to the structure and composition of each Pareto-optimal molecule that were very well-aligned with "chemical intuition". Since these findings emerged without any prior knowledge of the α, E gap -space, we argued that Pareto front analysis is a powerful (and largely underutilized) tool for in silico molecular design [56][57][58][59][60] . A potentially interesting next step would use these Pareto-optimal structures in conjunction with current ML approaches (e.g., active learning) to build reliable multi-objective frameworks for identifying the molecules in CCS (beyond that in QM7-X) missing in each front 63,64 .…”

“…When optimizing multiple objective functions among a large candidate pool, Pareto fronts (or frontiers) represent the so-called Pareto-optimal solutions for which no single objective function can be improved without degrading the others. Pareto fronts have been used in a number of fields (e.g., economics, medicine, materials science, chemical engineering) [56][57][58][59][60] and have given rise to evolutionary multi-objective optimization 61,62 . In this work, we extend our analysis in the previous section by using this approach to identify the most promising small organic molecules in CCS (as enumerated by the QM7-X database) to form polymeric battery materials 54,55 , i.e., the Pareto front of molecules in QM7-X which simultaneously have the largest α and E gap values (see Methods).…”

Deep learning for industrial processes: Forecasting amine emissions from a carbon capture plant

“…Our work shows that we could feed the data into an active learning model to harvest all the knowledge that has been collected during these experiments. Interrogation of this model can help us define the next most informative experiments 27,28 which we expect to greatly reduce the time to operability.…”

Featurization Strategies for Polymer Sequence or Composition Design by Machine Learning

“…72 As an alternative to using feature extraction architectures, Jablonka et al generated a hand-crafted vector of descriptors, which contained descriptions of sequence entropy or enumeration of sub-sequence clusters, to guide the in silico design of coarse-grained (CG) polymer dispersants. 73 While these developments are generally promising, it remains unclear under what circumstances and to what extent any given polymer featurization strategy outperforms another.…”