Multi-task Learning of Order-Consistent Causal Graphs

Authors: Xinshi Chen, Haoran Sun, Caleb Ellington, Eric Xing, Le Song

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct a set of synthetic experiments to demonstrates the effectiveness of the algorithm and validates the theoretical results. Furthermore, we apply our algorithm to more realistic single-cell expression RNA sequencing data generated by SERGIO [13] based on real gene regulatory networks.
Researcher Affiliation Collaboration Xinshi Chen Georgia Institute of Technology xinshi.chen@gatech.edu Haoran Sun Georgia Institute of Technology haoransun@gatech.edu Caleb Ellington Carnegie Mellon University cellingt@cs.cmu.edu Eric Xing Carnegie Mellon University MBZUAI eric.xing@mbzuai.ac.ae Le Song Bio Map MBZUAI le.song@mbzuai.ac.ae
Pseudocode Yes Algorithm 1: Joint Estimation Algorithm Hyperparameters :ρ, α, λ, t, δ Initialize G(1:K), T randomly; for itr = 1, , M do for itr = 1, , M do [G(1:K), T] Grad Opt Step f; G(1:K), T, β ; Gradient-based update on f i, j [p], G(1:K) ij G(1:K) ij G(1:K) ij 2 max n 0, G(1:K) ij 2 tλ|Tij| o ; Proximal step i, j [p], Tij sign(Tij) max 0, |Tij| tλ G(1:K) ij 2 ; Proximal step β β + τh(T); Dual ascent α α (1 + δ); Typical rule [23]
Open Source Code No The paper does not provide concrete access to source code or explicitly state its release.
Open Datasets Yes Furthermore, we apply our algorithm to more realistic single-cell expression RNA sequencing data generated by SERGIO [13] based on real gene regulatory networks.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning. It only mentions 'n samples' for each task.
Hardware Specification No The paper mentions using 'Partnership for an Advanced Computing Environment (PACE) at the Georgia Institute of Technology' but does not provide specific hardware details such as GPU/CPU models or memory.
Software Dependencies No The paper mentions using SERGIO for data generation but does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes Algorithm 1: Joint Estimation Algorithm Hyperparameters :ρ, α, λ, t, δ Initialize G(1:K), T randomly; for itr = 1, , M do... We simulate the gene expression from each network using SERGIO with a universal non-cooperative hill coefficient of 0.05, which works well with our linear recovery algorithm.