DeepGEM: Generalized Expectation-Maximization for Blind Inversion
Authors: Angela Gao, Jorge Castellanos, Yisong Yue, Zachary Ross, Katherine Bouman
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We showcase the effectiveness of our Deep GEM approach by achieving strong performance on the challenging problem of blind seismic tomography, where we significantly outperform the standard method used in seismology. We also demonstrate the generality of Deep GEM by applying it to a simple case of blind deconvolution. |
| Researcher Affiliation | Academia | Angela F. Gao1 Jorge C. Castellanos2 Yisong Yue1 Zachary E. Ross2 Katherine L. Bouman1 1Computing and Mathematical Sciences, California Institute of Technology 2Caltech Seismological Laboratory, California Institute of Technology {afgao, jcastellanos, yyue, zross, klbouman} @ caltech.edu |
| Pseudocode | No | The paper describes the steps of the E-step and M-step textually but does not provide pseudocode or a clearly labeled algorithm block. |
| Open Source Code | No | The paper does not provide an explicit statement or a link to open-source code for the described methodology. |
| Open Datasets | Yes | In Fig. 7 we show results from Deep GEM on three different Fashion-MNIST [34] images. |
| Dataset Splits | No | The paper mentions using Fashion-MNIST and simulated data but does not specify explicit training/validation/test dataset splits or percentages. |
| Hardware Specification | Yes | Each Deep GEM model takes 6 hours on a NVIDIA Quatro RTX 5000. [...] takes 1 hour on a NVIDIA Tesla V100. |
| Software Dependencies | No | The paper mentions using Adam as an optimizer and Real-NVP networks, and the eikonalfm package, but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | We use Adam as the optimizer [36] with a batch size of 32 and an E-step learning rate of 1e-3 and M-step learning rate of 5e-5. Hyperparameters (λT , λV , λθ) were empirically chosen by inspecting the loss on a grid search over hyperparameters. Results presented have been run with 10 EM iterations, each with 800 E -step epochs and 2000 M-step epochs. [...] We use Adam as the optimizer [36] with a batch size of 64 and an E-step learning rate of 5e-4 and M-step learning rate of 1e-4. Hyperparameters, weights used for sparsity and TV priors, were empirically chosen by a grid search over hyperparameters. Results presented have been run with 10 EM iterations, each with 400 E -step epochs and 400 M-step epochs |