Convex Regularization behind Neural Reconstruction

Authors: Arda Sahiner, Morteza Mardani, Batu Ozturkler, Mert Pilanci, John M. Pauly

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental A range of experiments with MNIST and fast MRI datasets confirm the efficacy of the dual network optimization problem.
Researcher Affiliation Academia Department of Electrical Engineering Stanford University {sahiner, morteza, ozt, pilanci, pauly}@stanford.edu
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code No The paper mentions the use of the PyTorch deep learning library but does not provide any explicit statement or link for the open-source code of their own methodology.
Open Datasets Yes We use a subset of the MNIST handwritten digits (Le Cun et al., 1998). ... we use the fast MRI dataset (Zbontar et al., 2018), a benchmark dataset for evaluating deep-learning based MRI reconstruction methods.
Dataset Splits No The paper specifies training and test sets but does not explicitly mention a validation set or specific split percentages for training, validation, and test.
Hardware Specification Yes We train both the primal and the dual network in a distributed fashion on a NVIDIA Ge Force GTX 1080 Ti GPU and NVIDIA Titan X GPU.
Software Dependencies No The paper mentions using the PyTorch deep learning library and the Sig Py python package but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes All networks were trained with an Adam optimizer, with β1 = 0.9, β2 = 0.999, and ϵ = 10 8. ... For the primal network, we use 512 filters, whereas for the dual network, we randomly sample 8,000 sign patterns... For the primal network, we train with a learning rate of µ = 10 1, whereas for the dual network we use a learning rate of µ = 10 3. We use a batch size of 25 for all cases. For the weight-decay parameter we use a value of β = 10 5.