Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

Missing Data Imputation by Reducing Mutual Information with Rectified Flows

Authors: Jiahao Yu, Qizhen Ying, Leyang Wang, Ziyue Jiang, Song Liu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments on synthetic and real-world datasets validate the efficacy of our proposed approach, demonstrating its superior imputation performance.
Researcher Affiliation Academia Jiahao Yu University of Cambridge Qizhen Ying University of Oxford Leyang Wang University College London Ziyue Jiang University of Bristol Song Liu University of Bristol
Pseudocode Yes Algorithm 1 MIRI with Rectified Flow (Single Imputation)
Open Source Code Yes Our implementation is available at https://github.com/yujhml/MIRI-Imputation.
Open Datasets Yes Comprehensive experiments on synthetic and real-world datasets validate the efficacy of our proposed approach, demonstrating its superior imputation performance. Our implementation is available at https://github.com/yujhml/MIRI-Imputation. ... UCI Regression Benchmarks. Table 4 lists the ten UCI datasets used. For each, we report sample size and feature dimensionality. CIFAR-10. We randomly sample 5 000 32 32 RGB images and apply pixel-level MCAR masks at varying rates. All three colour channels of each missing pixel are masked. Celeb A. We randomly sample 5 000 64 64 RGB images and apply channel-level MCAR masks at varying rates. All three colour channels of each missing pixel are masked independently.
Dataset Splits No The paper describes generating missingness masks and using a subset of samples for training/evaluation (e.g., "We train using only 5 000 samples (<10% of the full set) with 60% of pixels randomly removed"). However, it does not explicitly provide traditional train/test/validation dataset splits for the imputation model itself, nor for the downstream classification task mentioned.
Hardware Specification Yes Tabular Experiments. NVIDIA P100 GPU (16 GB), Intel Xeon E5-2680 v4 CPU (8 cores, 2.4 GHz), 24 GB RAM. Image Experiments. NVIDIA RTX 3090 GPU (24 GB), Intel Xeon Gold 6330 CPU (14 cores, 2.0 GHz), 90 GB RAM.
Software Dependencies No The paper mentions a "Py Torch+CUDA setup" for computational time analysis but does not specify version numbers for PyTorch, CUDA, or other key software libraries used for their method.
Experiment Setup No Section G.3 "Hyperparameter Selection" states: "Complete settings and search ranges are provided in our supplemental material. MIRI exhibits low sensitivity to these choices." This indicates that specific experimental setup details, such as hyperparameter values, are not present in the main text of the paper.