Optimization for Amortized Inverse Problems

Authors: Tianci Liu, Tong Yang, Quan Zhang, Qi Lei

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate the performance of the proposed algorithm on three inverse problem tasks, including denoising, noisy compressed sensing, and inpainting.
Researcher Affiliation Academia 1 Purdue University, United States 2 Peking University, China 3 Michigan State University, United States 4 New York University, United States.
Pseudocode Yes Algorithm 1 AIPO algorithm
Open Source Code No The paper does not provide an explicit statement or link for open-source code availability for the described methodology.
Open Datasets Yes The two models are trained on the Celeb A dataset (Liu et al., 2015)
Dataset Splits No The paper mentions using samples from the Celeb A test set but does not provide explicit training, validation, or test splits used within their experimental setup.
Hardware Specification No The paper does not provide specific hardware details such as CPU or GPU models used for the experiments.
Software Dependencies No The paper mentions software components like 'Adam', 'Real NVP', and 'GLOW' but does not specify version numbers for them or any other software dependencies.
Experiment Setup Yes In all the experiments, we compare the algorithms with a prespecified λ, which is set to be 0.3, 0.5, 1.0, 1.5, 2.0, respectively. Our AIPO and the baseline algorithm with the MLE initialization require a solution to (4) on the NCS and inpainting tasks, where we run 500 iterations of projected gradient descent. Table 4. Hyper-parameters used in Amortized Optimization appeared Algorithm 2. Hyperparameter step size α iteration K target rate r min. step size δ0 min δmin,h δmin,h Value 0.05 40 0.05 Λ/20 4δ0 min 1 δ0 min/4