RECORD
Space-Time Scattering Network for Electromagnetic Inverse Design and Tomography
Optics · Submitted 17 Nov 2018
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim We show this property leads to orders of magnitude smaller number of unknown weights, and a much faster convergence, compared with inverse design methods using conventional neural networks.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Maxwell equations generally explain the propagation of light through an arbitrary medium by using wave mechanics. However, scientific evidence since Newton suggest a discrete interpretation of light more generally explains its nature. This interpretation lends itself well to the discrete form of computer simulation. While current simulations attempt to discretize Maxwell equations, we present an inherently discrete physical model of light propagation that naturally forms a causal space-time scattering network (STSN). STSN has the topology of neural networks, inverse design and tomography based on STSN can be readily implemented in a variety of software and hardware that are optimized for deep learning. Also, STSN inherently includes the physics of light propagation, and hence the number of unknown weights in STSN is at a minimum. We show this property leads to orders of magnitude smaller number of unknown weights, and a much faster convergence, compared with inverse design methods using conventional neural networks. In addition, the intrinsic presence of space-time fabric in STSN allows time-dependent inverse design and tomography. We show examples of the fast convergence of STSN in predicting time-dependent index profiles while avoiding approximations typically used.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Maxwell equations generally explain the propagation of light through an arbitrary medium by using wave mechanics. | background 1.00 | 0.04 | 0.02 | asserted 0.91 |
| s1 | However, scientific evidence since Newton suggest a discrete interpretation of light more generally explains its nature. | background 0.97 | 0.04 | 0.04 | asserted 0.83 |
| s2 | This interpretation lends itself well to the discrete form of computer simulation. | background 0.81 | 0.08 | 0.03 | asserted 0.76 |
| s3 | While current simulations attempt to discretize Maxwell equations, we present an inherently discrete physical model of light propagation that naturally forms a causal space-time scattering network (STSN). | contribution 0.98 | 0.83 | 0.25 | asserted 1.00 |
| s4 | STSN has the topology of neural networks, inverse design and tomography based on STSN can be readily implemented in a variety of software and hardware that are optimized for deep learning. | interpretation 0.43 | 0.25 | 0.18 | asserted 0.66 |
| s5 | Also, STSN inherently includes the physics of light propagation, and hence the number of unknown weights in STSN is at a minimum. | contribution 0.37 | 0.40 | 0.12 | asserted 1.00 |
| s6 | We show this property leads to orders of magnitude smaller number of unknown weights, and a much faster convergence, compared with inverse design methods using conventional neural networks. | result 0.99 | 0.93 | 0.39 | asserted 1.00 |
| s7 | In addition, the intrinsic presence of space-time fabric in STSN allows time-dependent inverse design and tomography. | contribution 0.84 | 0.31 | 0.13 | asserted 0.98 |
| s8 | We show examples of the fast convergence of STSN in predicting time-dependent index profiles while avoiding approximations typically used. | result 0.94 | 0.95 | 0.19 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Maxwell equations generally explain the propagation of light through an arbitrary medium by using wave mechanics. However, scientific evidence since Newton suggest a discrete interpretation of light more generally explains its nature. This interpretation lends itself well to the discrete form of computer simulation. While current simulations attempt to discretize Maxwell equations, we present an inherently discrete physical model of light propagation that naturally forms a causal space-time scattering network (STSN). STSN has the topology of neural networks, inverse design and tomography based on STSN can be readily implemented in a variety of software and hardware that are optimized for deep learning. Also, STSN inherently includes the physics of light propagation, and hence the number of unknown weights in STSN is at a minimum. We show this property leads to orders of magnitude smaller number of unknown weights, and a much faster convergence, compared with inverse design methods using conventional neural networks. In addition, the intrinsic presence of space-time fabric in STSN allows time-dependent inverse design and tomography. We show examples of the fast convergence of STSN in predicting time-dependent index profiles while avoiding approximations typically used.
Human review
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.