Summary
The UN-BIASED project aims at developing an innovative Scientific Modelling paradigm capable of mitigating potential cognitive biases affecting the modelling process in engineering applications. Nowadays, modelling is mostly a subjective process, strongly driven by the prejudice of the Modeller and anchored to the knowledge of well-determined pre-set physics. In practical applications, this often results into models affected by epistemic uncertainty. Data-driven techniques open the path for the construction of computerized models that are able to learn the physics underlying a complex system from the available data alone, requiring little, if not at all, subjectivity. Interestingly, these tools are generally used to obtain mere predictions and no credit is usually given to the possibility of translating the learned patterns and relations into interpretable theories and hypotheses. I propose to assess the physics learned by data-driven algorithms in terms of compliance with fundamental principles e.g., laws of thermodynamics, and to test them against a priori subjective hypotheses. This will expose differences between the actual experiment and the Modeller’s understanding of it. This allows for inverting the rationale underlying the classical modelling process, from a theory-to-data deductive assessment to a data-to-theory inductive inference. The ultimate goal is to advance the state-of-the-art by crafting a two-way modelling framework combining the hypotheses-driven and the data-driven approaches, to mitigate the consequences of biased modelling choices and improve the knowledge about complex physical systems. The proposed paradigm is not to be intended as a substitution of the classical Scientific Modelling method, but rather as an extension of it. The project is conceived with aerospace applications in mind, but the proposed methodology is straightforwardly applicable to the modelling of any physical problem of interest for the academy or the industry.
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/101059320 |
| Start date: | 04-10-2022 |
| End date: | 03-10-2024 |
| Total budget - Public funding: | - 172 750,00 Euro |
Cordis data
Original description
The UN-BIASED project aims at developing an innovative Scientific Modelling paradigm capable of mitigating potential cognitive biases affecting the modelling process in engineering applications. Nowadays, modelling is mostly a subjective process, strongly driven by the prejudice of the Modeller and anchored to the knowledge of well-determined pre-set physics. In practical applications, this often results into models affected by epistemic uncertainty. Data-driven techniques open the path for the construction of computerized models that are able to learn the physics underlying a complex system from the available data alone, requiring little, if not at all, subjectivity. Interestingly, these tools are generally used to obtain mere predictions and no credit is usually given to the possibility of translating the learned patterns and relations into interpretable theories and hypotheses. I propose to assess the physics learned by data-driven algorithms in terms of compliance with fundamental principles e.g., laws of thermodynamics, and to test them against a priori subjective hypotheses. This will expose differences between the actual experiment and the Modeller’s understanding of it. This allows for inverting the rationale underlying the classical modelling process, from a theory-to-data deductive assessment to a data-to-theory inductive inference. The ultimate goal is to advance the state-of-the-art by crafting a two-way modelling framework combining the hypotheses-driven and the data-driven approaches, to mitigate the consequences of biased modelling choices and improve the knowledge about complex physical systems. The proposed paradigm is not to be intended as a substitution of the classical Scientific Modelling method, but rather as an extension of it. The project is conceived with aerospace applications in mind, but the proposed methodology is straightforwardly applicable to the modelling of any physical problem of interest for the academy or the industry.Status
SIGNEDCall topic
HORIZON-MSCA-2021-PF-01-01Update Date
09-02-2023
Images
No images available.
Geographical location(s)
