Polymer-derived ceramics (PDCs) can be obtained upon pyrolysis of suitable inorganic polymers in inert or reactive atmosphere.[1-3] An important characteristic of PDCs is the strong relationship between the molecular structure and chemistry of the polymers, their processing and the nanostructure and properties of the resulting ceramics.[2] They are nanoscopically heterogeneous and composed of nanodomains. This feature is controlled by the nature and organization of the segregated carbon formed during the polymer-to-ceramic conversion.[2,4,5] Due to their specific features such as tunable electrical, dielectrical, optical and thermal properties, PDCs can act as multifunctional materials performing multiple functions in a system.[1-5]

In response to the changing global landscape, energy has become a primary focus of the major world powers and scientific community. There has been great interest in developing and refining more efficient energy storage devices. One such device, the supercapacitor, has matured significantly over the last decade and emerged with the potential to facilitate major advances in energy storage. Supercapacitors, also known as ultracapacitors or electrochemical capacitors, utilize high surface area electrode materials and thin electrolytic dielectrics to achieve capacitances several orders of magnitude larger than conventional capacitors [1]. In doing so, supercapacitors are able to attain greater energy densities while still maintaining the characteristic high power density of conventional capacitors.

Optimizing the performance and parameters of complex systems is a challenging task in modern engineering and scientific research, especially when data acquisition is costly or of variable quality. Multi-fidelity Bayesian optimization (MFBO) is capable of integrating data with different fidelity (i.e., varying accuracy and cost) to improve the efficiency and reduce the total cost of the optimization process.

The aim of this study is to implement a multi-fidelity Bayesian optimization framework that takes into account the cost of data to improve the optimization performance under resource constrained conditions.

In this project, we are going to focus on establishing a methodology to leverage advanced machine learning techniques, particularly deep generative models, to predict novel inorganic material compositions. The model will be trained on a large dataset of computational inorganic materials, learning the underlying distribution and optimizing the relevant physical properties in the properly constructed latent space. Once trained, the model can propose new, valid inorganic compositions that are likely to exhibit desirable properties.

Expertise will be gained in the generative diffusion model as specified in https://arxiv.org/abs/2312.03687, and coding with Python valuable for both future PhD studies and industrial positions.

Polymer-derived ceramics (PDCs) can be obtained upon pyrolysis of suitable inorganic polymers in inert or reactive atmosphere.[1-3] An important characteristic of PDCs is the strong relationship between the molecular structure and chemistry of the polymers, their processing and the nanostructure and properties of the resulting ceramics.[2] They are nanoscopically heterogeneous and composed of nanodomains. This feature is controlled by the nature and organization of the segregated carbon formed during the polymer-to-ceramic conversion.[2,4,5] Due to their specific features such as tunable electrical, dielectrical, optical and thermal properties, PDCs can act as multifunctional materials performing multiple functions in a system.[1-5]

The topic to be addressed will involve the preparation of functional ceramics from tailored polymeric single-source precursors. The research work is closely linked to a scientific cooperation with the company Merck KGaA in Darmstadt.

Optimizing the performance and parameters of complex systems is a challenging task in modern engineering and scientific research, especially when data acquisition is costly or of variable quality. Multi-fidelity Bayesian optimization (MFBO) is capable of integrating data with different fidelity (i.e., varying accuracy and cost) to improve the efficiency and reduce the total cost of the optimization process.

The aim of this study is to implement a multi-fidelity Bayesian optimization framework that takes into account the cost of data to improve the optimization performance under resource constrained conditions.

In this project, we are going to focus on establishing a methodology to leverage advanced machine learning techniques, particularly deep generative models, to predict novel inorganic material compositions. The model will be trained on a large dataset of computational inorganic materials, learning the underlying distribution and optimizing the relevant physical properties in the properly constructed latent space. Once trained, the model can propose new, valid inorganic compositions that are likely to exhibit desirable properties.

Expertise will be gained in the generative diffusion model as specified in https://arxiv.org/abs/2312.03687, and coding with Python valuable for both future PhD studies and industrial positions.

Polymer-derived ceramics (PDCs) can be obtained upon pyrolysis of suitable inorganic polymers in inert or reactive atmosphere.[1-3] An important characteristic of PDCs is the strong relationship between the molecular structure and chemistry of the polymers, their processing and the nanostructure and properties of the resulting ceramics.[2] They are nanoscopically heterogeneous and composed of nanodomains. This feature is controlled by the nature and organization of the segregated carbon formed during the polymer-to-ceramic conversion.[2,4,5] Due to their specific features such as tunable electrical, dielectrical, optical and thermal properties, PDCs can act as multifunctional materials performing multiple functions in a system.[1-5]