Bayesian optimization in continuous spaces <i>via</i> virtual process embeddings

Valleti, Mani; Vasudevan, Rama K.; Ziatdinov, Maxim; Kalinin, Sergei V.

doi:10.1039/d2dd00065b

Cited by 8 publications

(11 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Deep learning models such as VAEs have been used in high-dimensional process optimization problems which optimize a scalar target function in the low-dimensional latent space [55][56][57][58]. However, their latent space has limitations when it comes to modeling categorical and continuous target functions.…”

Section: Latent Distributions In Vae and Rotationally Invariant Autoe...mentioning

confidence: 99%

Deep kernel methods learn better: from cards to process optimization

Valleti,

Vasudevan,

Ziatdinov

et al. 2024

Mach. Learn.: Sci. Technol.

Self Cite

View full text Add to dashboard Cite

The ability of deep learning methods to perform classification and regression tasks relies heavily on their capacity to uncover manifolds in high-dimensional data spaces and project them into low-dimensional representation spaces. In this study, we investigate the structure and character of the manifolds generated by classical variational autoencoder (VAE) approaches and deep kernel learning (DKL). In the former case, the structure of the latent space is determined by the properties of the input data alone, while in the latter, the latent manifold forms as a result of an active learning process that balances the data distribution and target functionalities. We show that DKL with active learning can produce a more compact and smooth latent space which is more conducive to optimization compared to previously reported methods, such as the VAE. We demonstrate this behavior using a simple cards data set and extend it to the optimization of domain-generated trajectories in physical systems. Our findings suggest that latent manifolds constructed through active learning have a more beneficial structure for optimization problems, especially in feature-rich target-poor scenarios that are common in domain sciences, such as materials synthesis, energy storage, and molecular discovery. The jupyter notebooks that encapsulate the complete analysis accompany the article.

show abstract

Section: Latent Distributions In Vae and Rotationally Invariant Autoe...mentioning

confidence: 99%

Deep kernel methods learn better: from cards to process optimization

Valleti,

Vasudevan,

Ziatdinov

et al. 2024

Mach. Learn.: Sci. Technol.

Self Cite

View full text Add to dashboard Cite

show abstract

“…41–44 Additionally, there are hybrid methods such as deep kernel learning (DKL) in which a deep neural network is trained to transform the input data to a GPR. 45,46 This concatenation of models makes the system able to accommodate variable length-scale correlations and discontinuities in the input space.…”

Section: Actions To Be Taken – the Knobsmentioning

confidence: 99%

Driving school for self-driving labs

Snapp,

Brown

2023

Digital Discovery

View full text Add to dashboard Cite

show abstract

“…Another method of reducing data dimensionality is using variational autoencoders (VAE) that perform this reduction by sampling from the latent space (i.e., a distribution). In Valleti et al., [189] a combination of a VAE with Bayesian optimization was used on functional data, the electric field as a function of time, to maximize the total curl of the polarization field of a system.…”

Section: Emerging Opportunitiesmentioning

confidence: 99%

Mechanisms of Biomolecular Self‐Assembly Investigated Through In Situ Observations of Structures and Dynamics

Schmid,

Lachowski,

Chiang

et al. 2023

Angew Chem Int Ed

View full text Add to dashboard Cite

Biomolecular self‐assembly of hierarchical materials is a precise and adaptable bottom‐up approach to synthesizing across scales with considerable energy, health, environment, sustainability, and information technology applications. To achieve desired functions in biomaterials, it is essential to directly observe assembly dynamics and structural evolutions that reflect the underlying energy landscape and the assembly mechanism. This review will summarize the current understanding of biomolecular assembly mechanisms based on in situ characterization and discuss the broader significance and achievements of newly gained insights. In addition, we will also introduce how emerging deep learning/machine learning‐based approaches, multiparametric characterization, and high‐throughput methods can boost the development of biomolecular self‐assembly. The objective of this review is to accelerate the development of in situ characterization approaches for biomolecular self‐assembly and to inspire the next generation of biomimetic materials.

show abstract

Bayesian optimization in continuous spaces via virtual process embeddings

Abstract: Process optimization in the latent space of functions via variational autoencoder (VAE) and Bayesian Optimization (BO). We demonstrate this to optimize the curl of a kinetic ferroelectric model.

Cited by 8 publications

References 51 publications

Deep kernel methods learn better: from cards to process optimization

Deep kernel methods learn better: from cards to process optimization

Driving school for self-driving labs

Mechanisms of Biomolecular Self‐Assembly Investigated Through In Situ Observations of Structures and Dynamics

Contact Info

Product

Resources

About