Corpus ID: 237940548

Mixed Integer Neural Inverse Design

@article{Ansari2021MixedIN,
  title={Mixed Integer Neural Inverse Design},
  author={Navid Ansari and Hans-Peter Seidel and Vahid Babaei},
  journal={ArXiv},
  year={2021},
  volume={abs/2109.12888}
}
Fig. 1. Taking advantage of the underlying mathematical properties of neural surrogate models (NSMs), we address challenging problems in neural inverse design. For example, given a piecewise linear NSM (left) that predicts the spectrum of a 3D printed surface as a function of the input ink ratios, we can find the best pair of inks for reproducing a certain target. On the right, we show how this toy problem, when cast as a mixed-integer linear programming (MILP), is solved combinatorially. 

Figures and Tables from this paper

References

SHOWING 1-10 OF 46 REFERENCES
A Unified View of Piecewise Linear Neural Network Verification
TLDR
A unified framework that encompasses previous methods is presented and the identification of new methods that combine the strengths of multiple existing approaches are identified, accomplishing a speedup of two orders of magnitude compared to the previous state of the art. Expand
Deep neural networks and mixed integer linear optimization
Deep Neural Networks (DNNs) are very popular these days, and are the subject of a very intense investigation. A DNN is made up of layers of internal units (or neurons), each of which computes anExpand
Deep learning for accelerated all-dielectric metasurface design.
TLDR
A novel method to solve the inverse modeling problem, termed fast forward dictionary search (FFDS), is developed, which offers tremendous controls to the designer and only requires an accurate forward neural network model. Expand
On the Number of Linear Regions of Deep Neural Networks
We study the complexity of functions computable by deep feedforward neural networks with piecewise linear activations in terms of the symmetries and the number of linear regions that they have. DeepExpand
Evaluating Robustness of Neural Networks with Mixed Integer Programming
TLDR
Verification of piecewise-linear neural networks as a mixed integer program that is able to certify more samples than the state-of-the-art and find more adversarial examples than a strong first-order attack for every network. Expand
Nanophotonic particle simulation and inverse design using artificial neural networks
TLDR
A method to use artificial neural networks to approximate light scattering by multilayer nanoparticles is proposed and it is found that the network needs to be trained on only a small sampling of the data to approximate the simulation to high precision. Expand
High-contrast computational caustic design
TLDR
This work presents a new algorithm that solves for the shape of a transparent object such that the refracted light paints a desired caustic image on a receiver screen and introduces an optimal transport formulation to establish a correspondence between the input geometry and the unknown target shape. Expand
Mixed Integer Linear Programming Formulation Techniques
TLDR
This survey reviews advanced MIP formulation techniques that result in stronger and/or smaller formulations for a wide class of problems. Expand
ON THE SOLUTION OF DISCRETE PROGRAMMING PROBLEMS
Abstract : This paper considers optimization problems in which some or all variables must take on integral values. An ability to solve such problems would be valuable in itself and would also allowExpand
Practical Guidelines for Solving Difficult Mixed Integer Linear
TLDR
Suggestions for appropriate use of state-of-the-art optimizers and guidelines for careful formulation are presented, both of which can vastly improve performance. Expand
...
1
2
3
4
5
...