Who did it? Identifying the Most Likely Origins of Events

Marcel Gehrke, Ralf Möller, Tanya Braun
Proceedings of The 11th International Conference on Probabilistic Graphical Models, PMLR 186:217-228, 2022.

Abstract

One probabilistic inference task concerns answering queries for conditional marginal distributions, where a set of events is given. In this paper, we investigate the problem of only knowing that events are observed, from a number of sensors or for individuals, but not which sensors or individuals exhibit those events specifically. This situation might occur in multi-agent settings, such as in nanosystems, where single agents can no longer be tracked. However, to be able to perform probabilistic inference, those events need to be mapped to random variables, specifically to those that are most likely to exhibit those events. For the mapping, we show how lifting allows for generating all different possibilities to map those events, as we can do it over sets of indistinguishable random variables, leading to a set of queries. Given the mapping that leads to the most likely answer, we can construct evidence to perform probabilistic inference with. Finally, we compare solving the problem on the propositional level, which cannot be done in reasonable time, to our approach, which returns liftable evidence for tractable inference.

Cite this Paper

BibTeX
@InProceedings{pmlr-v186-gehrke22a, title = {{Who did it? Identifying the Most Likely Origins of Events}}, author = {Gehrke, Marcel and M\"oller, Ralf and Braun, Tanya}, booktitle = {Proceedings of The 11th International Conference on Probabilistic Graphical Models}, pages = {217--228}, year = {2022}, editor = {Salmerón, Antonio and Rumı́, Rafael}, volume = {186}, series = {Proceedings of Machine Learning Research}, month = {05--07 Oct}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v186/gehrke22a/gehrke22a.pdf}, url = {https://proceedings.mlr.press/v186/gehrke22a.html}, abstract = {One probabilistic inference task concerns answering queries for conditional marginal distributions, where a set of events is given. In this paper, we investigate the problem of only knowing that events are observed, from a number of sensors or for individuals, but not which sensors or individuals exhibit those events specifically. This situation might occur in multi-agent settings, such as in nanosystems, where single agents can no longer be tracked. However, to be able to perform probabilistic inference, those events need to be mapped to random variables, specifically to those that are most likely to exhibit those events. For the mapping, we show how lifting allows for generating all different possibilities to map those events, as we can do it over sets of indistinguishable random variables, leading to a set of queries. Given the mapping that leads to the most likely answer, we can construct evidence to perform probabilistic inference with. Finally, we compare solving the problem on the propositional level, which cannot be done in reasonable time, to our approach, which returns liftable evidence for tractable inference.} }
Endnote
%0 Conference Paper %T Who did it? Identifying the Most Likely Origins of Events %A Marcel Gehrke %A Ralf Möller %A Tanya Braun %B Proceedings of The 11th International Conference on Probabilistic Graphical Models %C Proceedings of Machine Learning Research %D 2022 %E Antonio Salmerón %E Rafael Rumı́ %F pmlr-v186-gehrke22a %I PMLR %P 217--228 %U https://proceedings.mlr.press/v186/gehrke22a.html %V 186 %X One probabilistic inference task concerns answering queries for conditional marginal distributions, where a set of events is given. In this paper, we investigate the problem of only knowing that events are observed, from a number of sensors or for individuals, but not which sensors or individuals exhibit those events specifically. This situation might occur in multi-agent settings, such as in nanosystems, where single agents can no longer be tracked. However, to be able to perform probabilistic inference, those events need to be mapped to random variables, specifically to those that are most likely to exhibit those events. For the mapping, we show how lifting allows for generating all different possibilities to map those events, as we can do it over sets of indistinguishable random variables, leading to a set of queries. Given the mapping that leads to the most likely answer, we can construct evidence to perform probabilistic inference with. Finally, we compare solving the problem on the propositional level, which cannot be done in reasonable time, to our approach, which returns liftable evidence for tractable inference.
APA
Gehrke, M., Möller, R. & Braun, T.. (2022). Who did it? Identifying the Most Likely Origins of Events. Proceedings of The 11th International Conference on Probabilistic Graphical Models, in Proceedings of Machine Learning Research 186:217-228 Available from https://proceedings.mlr.press/v186/gehrke22a.html.

Related Material