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..
Neural Mutual Information Estimation with Vector Copulas
Authors: Yanzhi Chen, Zijing Ou, Adrian Weller, Michael U. Gutmann
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on state-of-the-art synthetic benchmarks and real-world data with diverse modalities demonstrate the advantages of the proposed estimator. |
| Researcher Affiliation | Academia | 1University of Cambridge, 2Imperial College London, 3Alan Turing Institute, 4University of Edinburgh |
| Pseudocode | Yes | Algorithm 1 Vector copula MI estimate (VCE) Input: data D = {x(i), y(i)}n i=1 Output: estimated ˆI(X; Y ) Parameters: flows f X, f Y , copulas {c1, ..c M} Initialization: D = Dtrain Dval, K = 1, |
| Open Source Code | Yes | Code containing both our method and state-of-the-art neural estimators is available in [github repo]. |
| Open Datasets | Yes | We consider representative cases from this benchmark [58], further extending it by (a) considering varying dependence strengths for each chosen case; (b) employing mixing matrices A, B to couple the dimensions in X and Y respectively. We also include the mixture models in [49] to enrich our tests. [...] We next consider the benchmark [59], which contains correlated images X and Y; [...] It consists of pairs of embeddings from a language model (LM) [60, 61] computed on the IMDB dataset [62] |
| Dataset Splits | No | Algorithm 1 ... Initialization: D = Dtrain Dval. While the paper mentions the use of training and validation sets, it does not specify explicit percentages or sample counts for these splits for any of the datasets used in the experiments. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments. The NeurIPS checklist indicates |
| Software Dependencies | No | The paper mentions using the Adam optimizer [67] but does not specify other key software components with version numbers, such as programming languages (e.g., Python), deep learning frameworks (e.g., PyTorch, TensorFlow), or CUDA versions. |
| Experiment Setup | Yes | Hyperparams. For the vector copula in VCE, we consider mixtures with 1, 4, 8, 16, 32 components. [...] All neural networks in our method and in the baselines use the same architecture unless otherwise specified: a multilayer perceptron (MLP) with 3 hidden layers, each with 512 units and ReLU activation. For flow-based models, we adopt 3 coupling layers of neural spline flows. We use the Adam optimizer [67] with a learning rate of 1e-4 and a batch size of 256 for all experiments. All models are trained for 200 epochs. |