Trajectory Inference via Mean-field Langevin in Path Space
Authors: Lénaïc Chizat, Stephen Zhang, Matthieu Heitz, Geoffrey Schiebinger
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In Figure 1(left) we illustrate two extreme cases: N = 1 (very few samples at intermediate timepoints) and N = 64 (uniform sampling over time) respectively, with λ = 0.05, λg WOT = 0.0025. Visually, it seems that the output of MFL is robust to the few-sample regime, with relatively little qualitative difference between the reconstructed trajectories for N = 1, 64. On the other hand, the performance of g WOT degrades visibly once the set of observed points is a poor reflection of the support of the underlying process. To examine this quantitatively, we applied both MFL dynamics and g WOT for various values of N and computed the root-mean-square (RMS) Energy Distance [29] over time to an approximate ground truth (see Appendix G). |
| Researcher Affiliation | Academia | Joint first authors. Institute of mathematics, EPFL, Switzerland University of Melbourne, Victoria, Australia University of British Columbia, Canada |
| Pseudocode | No | The paper does not contain a formally labeled pseudocode or algorithm block. |
| Open Source Code | No | The paper does not provide an explicit statement or link to open-source code for the methodology described. |
| Open Datasets | Yes | We now consider the stem cell reprogramming dataset of [2], in which a population of differentiating cells were profiled over a time course using single-cell RNA sequencing. For the purpose of this example, we restrict our attention to days 2.5-6 inclusive making a total of 8 timepoints. |
| Dataset Splits | No | The paper describes how data was subsampled for evaluation, but it does not specify explicit training or validation splits in the context of model parameter tuning or typical machine learning training processes. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software names with version numbers for dependencies. |
| Experiment Setup | Yes | In Figure 1(left) we illustrate two extreme cases: N = 1 (very few samples at intermediate timepoints) and N = 64 (uniform sampling over time) respectively, with λ = 0.05, λg WOT = 0.0025. ... where ρ > 0 is a parameter... From this, we subsampled timepoints consisting of 100 cells at days 2.5 and 6, and 10 cells at the remaining intermediate timepoints. |