RGL
EPFL Logo

Solving Inverse PDE Problems using Grid-Free Monte Carlo Estimators

To appear in Transactions on Graphics (Proceedings of SIGGRAPH Asia 2024)

We ap­ply our in­verse PDE solv­er to a 2D elec­tric­al im­ped­ance tomo­graphy ex­per­i­ment [Haupt­mann et al . 2017], in which an elec­tric cur­rent flows through a sa­line-filled wa­ter tank con­tain­ing con­duct­ing ob­jects of dif­fer­ent sizes (pho­to­graphs in middle). The mark­er in the middle in­dic­ates the cen­ter. In­ject­ing a cur­rent us­ing 2 of 16 uni­formly spaced elec­trodes at the tank bound­ary gen­er­ates a meas­ur­able voltage dif­fer­ence at the oth­er elec­trodes, and in­ject­ing at vari­ous loc­a­tions pro­duces a mat­rix 𝑒ref of meas­ure­ments. The ob­ject­ive of this in­verse prob­lem is to in­fer the prop­er­ties of the con­duct­or from this data. We per­form a dif­fer­en­ti­able sim­u­la­tion of this setup to op­tim­ize the cen­ter and ra­di­us of a con­duct­ing circle. The frames on the left show the pro­gres­sion of the op­tim­iz­a­tion (columns (a)), while the right­most two columns re­veal how the pre­dicted voltages be­come in­creas­ingly con­sist­ent with the meas­ure­ment (columns (d)).

Abstract

Par­tial dif­fer­en­tial equa­tions can mod­el di­verse phys­ic­al phe­nom­ena in­clud­ing heat dif­fu­sion, in­com­press­ible flows, and elec­tro­stat­ic po­ten­tials. Giv­en a de­scrip­tion of an ob­ject’s bound­ary and in­teri­or, tra­di­tion­al meth­ods solve such PDEs by densely mesh­ing the in­teri­or and then solv­ing a large and sparse lin­ear sys­tem de­rived from this mesh. Re­cent grid-free solv­ers take an al­tern­at­ive ap­proach and avoid this com­plex­ity in ex­change for ran­dom­ness: they com­pute stochast­ic solu­tion es­tim­ates and gen­er­ally bear a strik­ing re­semb­lance to phys­ic­ally-based ren­der­ingΒ al­gorithms.

In this art­icle, we de­vel­op al­gorithms tar­get­ing the in­verse form of this prob­lem: giv­en an already ex­ist­ing solu­tion of a PDE, we in­fer para­met­ers char­ac­ter­iz­ing the bound­ary and in­teri­or. In the grid-free set­ting, there are again sig­ni­fic­ant con­nec­tions to ren­der­ing, and we show how in­sights from both fields can be com­bined to com­pute un­biased de­riv­at­ive es­tim­ates that en­able gradi­ent-based op­tim­iz­a­tion. In this pro­cess, we en­counter new chal­lenges that must be ad­dressed to ob­tain prac­tic­al solu­tions. We in­tro­duceΒ ac­cel­er­a­tion and vari­ance re­duc­tion strategies and show how to dif­fer­en­ti­ate branch­ing ran­dom walks in re­verse mode. We fi­nally demon­strate our ap­proach on both sim­u­lated data and a real-world elec­tric­al im­ped­ance tomo­graphy ex­per­i­ment, where we re­con­struct the po­s­i­tion of a con­duct­ing ob­ject from voltage meas­ure­ments taken in a sa­line-filledΒ tank.

Video

Figures

Text citation

Ekrem Fatih Yilmazer, Delio Vicini, and Wenzel Jakob. 2024. Solving Inverse PDE Problems using Monte Carlo Estimators. In Transactions on Graphics (Proceedings of SIGGRAPH Asia) 43.

BibTeX
@article{Yilmazer2024Solving,
    author = {Ekrem Fatih Yilmazer and Delio Vicini and Wenzel Jakob},
    title = {Solving Inverse PDE Problems using Monte Carlo Estimators},
    journal = {Transactions on Graphics (Proceedings of SIGGRAPH Asia)},
    volume = {43},
    year = {2024},
    month = dec,
    doi = {10.1145/3687990}
}