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..
Meta-D2AG: Causal Graph Learning with Interventional Dynamic Data
Authors: Tian Gao, Songtao Lu, Junkyu Lee, Elliot Nelson, Debarun Bhattacharjya, Yue Yu, Miao Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Empirical Evaluation We evaluate our proposed meta dynamic DAG learning algorithm against some of the existing DAG learning algorithms, designed for dynamic time series datasets and/or with multi-domain capabilities. We test the algorithms on synthetic datasets as well as a simulated reinforcement learning (RL) environment, Sprites World, with intervention datasets. |
| Researcher Affiliation | Collaboration | Tian Gao IBM Research Songtao Lu The Chinese University of Hong Kong Junkyu Lee IBM Research Elliot Nelson Independent Debarun Bhattacharjya IBM Research Yue Yu Lehigh University Miao Liu IBM Research |
| Pseudocode | Yes | Algorithm 1 Online Meta-D2AG Algorithm Require: Data Xm, m {1, . . . , M}, w, η, γ 1: Output: Learned shared weighted adjacency matrix Ws and private Wp,m, m 2: initialize Ws and Wp,0 3: for each domain m do 4: Initialize Φp,m := Φp,m 1 5: while equation 11 holds do 6: g i (Φs) = minΦp,i gi(Φs, Φp,i) {Minimize inner loop} 7: Compute Si,w with equation 10 8: Φ proxηf(Φ η Φ b Si,w(Φ)) 9: end while 10: end for |
| Open Source Code | No | Answer: [Yes] Justification: Will release code upon acceptance of the paper. |
| Open Datasets | Yes | We test the algorithms on synthetic datasets as well as a simulated reinforcement learning (RL) environment, Sprites World, with intervention datasets. ... [69] N. Watters, L. Matthey, M. Bosnjak, C. P. Burgess, and A. Lerchner. Cobra: Data-efficient model-based rl through unsupervised object discovery and curiosity-driven exploration. ar Xiv preprint ar Xiv:1905.09275, 2019. |
| Dataset Splits | Yes | For the online setting where we use a sliding window size of 3 previous time domains to meta-train and the current domain as the meta-test dataset. |
| Hardware Specification | Yes | The experiments are run on a machine with a 3.2 GHz CPU and 64 GB of memory. |
| Software Dependencies | No | Sprites world [69] is a Python-based RL environment in which objects of various shapes interact with each other in 2-dimensional space. |
| Experiment Setup | Yes | We search for the best value of each of 5 parameters sequentially, including two L1 penalty coefficients for W a and W b, the threshold to obtain final W a and W b, and the hidden neuron size. For λa and λw, we search over a value range of {10 5, 10 4, 10 3, 10 2, 10 1}. Graph threshold search range is set to be {0.001, 0.01, 0.05, 0.1, 0.2, 0.3}, and neuron size range is searched over {8, 16, 32, 64}. We use a two-layer MLP architecture in the experiments: Xi = W1σ((Wp,i(Ws) Ws)Xi)), where W1 is the MLP layer weight and σ is the Re LU activation function. |