Minimax Multi-Target Conformal Prediction with Applications to Imaging Inverse Problems
Authors: Jeffrey Wen, Rizwan Ahmad, Philip Schniter
TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We numerically compare our proposed method to several existing multi-target conformal prediction methods on a synthetic-data problem and four accelerated-MRI problems. ... We now numerically evaluate the proposed asymptotically minimax multi-target conformal prediction method from Section 3.2, along with the independence-assumption (IA)-based method from Messoudi et al. (2020), the quantile-normalization (QN)-based method from Sampson & Chan (2024), and the copula-based CPTS method from Sun & Yu (2024), all described in Section 2.3. We compare all four methods using both synthetic data and real-world accelerated-MRI data. |
| Researcher Affiliation | Academia | Jeffrey Wen EMAIL Department of Electrical and Computer Engineering The Ohio State University Rizwan Ahmad EMAIL Department of Biomedical Engineering The Ohio State University Philip Schniter EMAIL Department of Electrical and Computer Engineering The Ohio State University |
| Pseudocode | Yes | Algorithm 1 Minimax-based conformal prediction of test target z0 RK from feature vector u0 U. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing code for the methodology described, nor does it include a direct link to a code repository. |
| Open Datasets | Yes | Data: We follow the experimental setup of Wen et al. (2024), which uses the non-fat-suppressed subset of the multicoil fast MRI knee dataset from Zbontar et al. (2018). ... fast MRI+ knee data (Zhao et al., 2022). |
| Dataset Splits | Yes | Validation: We first construct a tuning set dtune using 656 of the 2188 fast MRI validation samples (i.e., 30%), selected randomly. ... we randomly partition the remaining validation data into a calibration set dcal[t] of size n = 1073 (or 50%) and a test set of size ntest = 459 (or 20%) using indices i Itest[t]. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments. It mentions 'accelerated MRI' but no computational hardware specifications. |
| Software Dependencies | No | For the image-recovery model f( ), we use the popular E2E-Var Net from Sriram et al. (2020), and for the posterior-sampling method g( , ), we use the conditional normalizing flow (CNF) from Wen et al. (2023) with c = 32 posterior samples. ... For the classifier, we use a Res Net-50 (He et al., 2016). While software components are mentioned, specific version numbers for libraries or frameworks (e.g., PyTorch, TensorFlow) are not provided. |
| Experiment Setup | Yes | Then we pretrain the network in a self-supervised fashion using the (unlabeled) non-fat-suppressed fast MRI knee data following the Sim CLR procedure from Chen et al. (2020) with a learning rate of 0.0002, batch size of 128, and 500 epochs. Finally, we perform supervised fine-tuning ... for 150 epochs with a batch size of 128, learning rate of 5e-5, and weight decay of 1e-7. |