Friday

Teodoro Laino
Fueling the Digital Chemistry Revolution with Language and Multimodal Foundation Models

A rooted knowledge and understanding of the material and its properties stems from a holistic perspective. Indeed, when discussing the properties of a newly engineered material, it is common to present:
- a text-based description of the sequence of actions through which such material was obtained, listing key variables as scalars.
- a characterization of its structure by means of advanced microscopy (e.g., 2D images, 3D tomographies, 4D spatio-temporal analysis) and spectroscopy (e.g., adsorption spectra, NMR spectra), also with the aid of atomistic and electronic structure simulations.
- a list of key performance indicators, in the form of scalar variables (e.g. the mechanical properties of an alloy or the Seebeck coefficient of a thermoelectric) or a time-series (e.g., activity of a catalyst over time, the capacity of a battery over time).
- a mechanistic discussion of the relationships that link structure-to-property, often through quantities extracted from electronic structure and atomistic scale simulations.

Machine learning methods are revolutionizing the way we approach materials design, making an impact in each, yet, it is rare that they fully exploit information and data from different modalities and sources. The aim of this workshop is thus two-fold:
- Young researchers will have the opportunity to grow solid foundations and a complete overview of cutting-edge machine-learning methods that enable the community to tackle outstanding challenges across diverse domains in materials design and discovery.
- Attendees will have the chance to discuss and identify routes on how to best combine information of different nature towards a unified vision (and solution) of the material design and discovery problem. To this end, during the workshop part, invited and contributed speakers, and panel discussions will take place, with a focus on multi-modal, multi-objective, and multi-fidelity machine learning methods in materials science.

Lei Zhang
Boosting Materials Design: Word Embeddings Meet Experimental Data

A rooted knowledge and understanding of the material and its properties stems from a holistic perspective. Indeed, when discussing the properties of a newly engineered material, it is common to present:
- a text-based description of the sequence of actions through which such material was obtained, listing key variables as scalars.
- a characterization of its structure by means of advanced microscopy (e.g., 2D images, 3D tomographies, 4D spatio-temporal analysis) and spectroscopy (e.g., adsorption spectra, NMR spectra), also with the aid of atomistic and electronic structure simulations.
- a list of key performance indicators, in the form of scalar variables (e.g. the mechanical properties of an alloy or the Seebeck coefficient of a thermoelectric) or a time-series (e.g., activity of a catalyst over time, the capacity of a battery over time).
- a mechanistic discussion of the relationships that link structure-to-property, often through quantities extracted from electronic structure and atomistic scale simulations.

Machine learning methods are revolutionizing the way we approach materials design, making an impact in each, yet, it is rare that they fully exploit information and data from different modalities and sources. The aim of this workshop is thus two-fold:
- Young researchers will have the opportunity to grow solid foundations and a complete overview of cutting-edge machine-learning methods that enable the community to tackle outstanding challenges across diverse domains in materials design and discovery.
- Attendees will have the chance to discuss and identify routes on how to best combine information of different nature towards a unified vision (and solution) of the material design and discovery problem. To this end, during the workshop part, invited and contributed speakers, and panel discussions will take place, with a focus on multi-modal, multi-objective, and multi-fidelity machine learning methods in materials science.

Matej Praprotnik
Developing an Implicit Solvation Machine Learning Model for Molecular Simulations of Ionic Media

A rooted knowledge and understanding of the material and its properties stems from a holistic perspective. Indeed, when discussing the properties of a newly engineered material, it is common to present:
- a text-based description of the sequence of actions through which such material was obtained, listing key variables as scalars.
- a characterization of its structure by means of advanced microscopy (e.g., 2D images, 3D tomographies, 4D spatio-temporal analysis) and spectroscopy (e.g., adsorption spectra, NMR spectra), also with the aid of atomistic and electronic structure simulations.
- a list of key performance indicators, in the form of scalar variables (e.g. the mechanical properties of an alloy or the Seebeck coefficient of a thermoelectric) or a time-series (e.g., activity of a catalyst over time, the capacity of a battery over time).
- a mechanistic discussion of the relationships that link structure-to-property, often through quantities extracted from electronic structure and atomistic scale simulations.

Machine learning methods are revolutionizing the way we approach materials design, making an impact in each, yet, it is rare that they fully exploit information and data from different modalities and sources. The aim of this workshop is thus two-fold:
- Young researchers will have the opportunity to grow solid foundations and a complete overview of cutting-edge machine-learning methods that enable the community to tackle outstanding challenges across diverse domains in materials design and discovery.
- Attendees will have the chance to discuss and identify routes on how to best combine information of different nature towards a unified vision (and solution) of the material design and discovery problem. To this end, during the workshop part, invited and contributed speakers, and panel discussions will take place, with a focus on multi-modal, multi-objective, and multi-fidelity machine learning methods in materials science.

Panel discussion

A rooted knowledge and understanding of the material and its properties stems from a holistic perspective. Indeed, when discussing the properties of a newly engineered material, it is common to present:
- a text-based description of the sequence of actions through which such material was obtained, listing key variables as scalars.
- a characterization of its structure by means of advanced microscopy (e.g., 2D images, 3D tomographies, 4D spatio-temporal analysis) and spectroscopy (e.g., adsorption spectra, NMR spectra), also with the aid of atomistic and electronic structure simulations.
- a list of key performance indicators, in the form of scalar variables (e.g. the mechanical properties of an alloy or the Seebeck coefficient of a thermoelectric) or a time-series (e.g., activity of a catalyst over time, the capacity of a battery over time).
- a mechanistic discussion of the relationships that link structure-to-property, often through quantities extracted from electronic structure and atomistic scale simulations.

Machine learning methods are revolutionizing the way we approach materials design, making an impact in each, yet, it is rare that they fully exploit information and data from different modalities and sources. The aim of this workshop is thus two-fold:
- Young researchers will have the opportunity to grow solid foundations and a complete overview of cutting-edge machine-learning methods that enable the community to tackle outstanding challenges across diverse domains in materials design and discovery.
- Attendees will have the chance to discuss and identify routes on how to best combine information of different nature towards a unified vision (and solution) of the material design and discovery problem. To this end, during the workshop part, invited and contributed speakers, and panel discussions will take place, with a focus on multi-modal, multi-objective, and multi-fidelity machine learning methods in materials science.

Matilda Sipilä
Application of the Question Answering method to extract information from materials science literature

A rooted knowledge and understanding of the material and its properties stems from a holistic perspective. Indeed, when discussing the properties of a newly engineered material, it is common to present:
- a text-based description of the sequence of actions through which such material was obtained, listing key variables as scalars.
- a characterization of its structure by means of advanced microscopy (e.g., 2D images, 3D tomographies, 4D spatio-temporal analysis) and spectroscopy (e.g., adsorption spectra, NMR spectra), also with the aid of atomistic and electronic structure simulations.
- a list of key performance indicators, in the form of scalar variables (e.g. the mechanical properties of an alloy or the Seebeck coefficient of a thermoelectric) or a time-series (e.g., activity of a catalyst over time, the capacity of a battery over time).
- a mechanistic discussion of the relationships that link structure-to-property, often through quantities extracted from electronic structure and atomistic scale simulations.

Machine learning methods are revolutionizing the way we approach materials design, making an impact in each, yet, it is rare that they fully exploit information and data from different modalities and sources. The aim of this workshop is thus two-fold:
- Young researchers will have the opportunity to grow solid foundations and a complete overview of cutting-edge machine-learning methods that enable the community to tackle outstanding challenges across diverse domains in materials design and discovery.
- Attendees will have the chance to discuss and identify routes on how to best combine information of different nature towards a unified vision (and solution) of the material design and discovery problem. To this end, during the workshop part, invited and contributed speakers, and panel discussions will take place, with a focus on multi-modal, multi-objective, and multi-fidelity machine learning methods in materials science.

Tian Xie
Accelerating materials design with AI emulators and generators

A rooted knowledge and understanding of the material and its properties stems from a holistic perspective. Indeed, when discussing the properties of a newly engineered material, it is common to present:
- a text-based description of the sequence of actions through which such material was obtained, listing key variables as scalars.
- a characterization of its structure by means of advanced microscopy (e.g., 2D images, 3D tomographies, 4D spatio-temporal analysis) and spectroscopy (e.g., adsorption spectra, NMR spectra), also with the aid of atomistic and electronic structure simulations.
- a list of key performance indicators, in the form of scalar variables (e.g. the mechanical properties of an alloy or the Seebeck coefficient of a thermoelectric) or a time-series (e.g., activity of a catalyst over time, the capacity of a battery over time).
- a mechanistic discussion of the relationships that link structure-to-property, often through quantities extracted from electronic structure and atomistic scale simulations.

Machine learning methods are revolutionizing the way we approach materials design, making an impact in each, yet, it is rare that they fully exploit information and data from different modalities and sources. The aim of this workshop is thus two-fold:
- Young researchers will have the opportunity to grow solid foundations and a complete overview of cutting-edge machine-learning methods that enable the community to tackle outstanding challenges across diverse domains in materials design and discovery.
- Attendees will have the chance to discuss and identify routes on how to best combine information of different nature towards a unified vision (and solution) of the material design and discovery problem. To this end, during the workshop part, invited and contributed speakers, and panel discussions will take place, with a focus on multi-modal, multi-objective, and multi-fidelity machine learning methods in materials science.

ML4MS 2024 Closing

A rooted knowledge and understanding of the material and its properties stems from a holistic perspective. Indeed, when discussing the properties of a newly engineered material, it is common to present:
- a text-based description of the sequence of actions through which such material was obtained, listing key variables as scalars.
- a characterization of its structure by means of advanced microscopy (e.g., 2D images, 3D tomographies, 4D spatio-temporal analysis) and spectroscopy (e.g., adsorption spectra, NMR spectra), also with the aid of atomistic and electronic structure simulations.
- a list of key performance indicators, in the form of scalar variables (e.g. the mechanical properties of an alloy or the Seebeck coefficient of a thermoelectric) or a time-series (e.g., activity of a catalyst over time, the capacity of a battery over time).
- a mechanistic discussion of the relationships that link structure-to-property, often through quantities extracted from electronic structure and atomistic scale simulations.

Machine learning methods are revolutionizing the way we approach materials design, making an impact in each, yet, it is rare that they fully exploit information and data from different modalities and sources. The aim of this workshop is thus two-fold:
- Young researchers will have the opportunity to grow solid foundations and a complete overview of cutting-edge machine-learning methods that enable the community to tackle outstanding challenges across diverse domains in materials design and discovery.
- Attendees will have the chance to discuss and identify routes on how to best combine information of different nature towards a unified vision (and solution) of the material design and discovery problem. To this end, during the workshop part, invited and contributed speakers, and panel discussions will take place, with a focus on multi-modal, multi-objective, and multi-fidelity machine learning methods in materials science.