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..
A Geometry-Aware Metric for Mode Collapse in Time Series Generative Models
Authors: Yassine ABBAHADDOU, Amine Aboussalah
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on synthetic and real-world datasets using Time GAN, Time VAE, and Diffusion TS show that DMD-GEN aligns with existing metrics while providing the first principled framework for detecting and interpreting mode collapse in time series. |
| Researcher Affiliation | Academia | Yassine Abbahaddou LIX, Ecole Polytechnique Institut Polytechnique de Paris EMAIL Amine Mohamed Aboussalah Department of Finance and Risk Engineering Tandon School of Engineering New York University EMAIL |
| Pseudocode | Yes | Algorithm 1 Computation of DMD-GEN Metric Algorithm 2 Dynamic Mode Decomposition |
| Open Source Code | Yes | Our code is available at: here. Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We released the code in the supplementary material, we also used public datasets. |
| Open Datasets | Yes | Stock price. To test our framework on a complex multimodal dataset, we used Google stocks data from 2004 to 2019, which was used in [60]. Energy. We conducted experiments on UCI s air quality dataset [55] consisting of hourly averaged responses from an array of 5 metal oxide chemical sensors embedded in an Air Quality Chemical Multisensor Device in an Italian city. Electricity Transformer Temperature and humidity (ETTh). The ETTh dataset focuses on temperature and humidity data from electricity transformers [62]. |
| Dataset Splits | Yes | The time series were then cut into sequences with length 24, following the setup in the work done by [60]. Question: Does the paper specify all the training and test details (e.g., data splits, hyperparameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: We provided s details of the experimental setup, including details about 889 the train/val/test used folds and the values of all hyperparameters. |
| Hardware Specification | Yes | The experiments were conducted on an NVIDIA A100 GPU. |
| Software Dependencies | No | We utilized the py DMD package 2 in Python to compute the DMD eigenvalues and eigenvectors. For generating synthetic time series, we used the original settings and the official implementation of Diffusion TS3, Time GAN4 and Time VAE5. Justification: The paper mentions "Python" and "py DMD package" but does not specify a version number for either, nor for Diffusion TS, Time GAN, or Time VAE implementations, which is required for specific ancillary software details. |
| Experiment Setup | Yes | Question: Does the paper specify all the training and test details (e.g., data splits, hyperparameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: We provided s details of the experimental setup, including details about 889 the train/val/test used folds and the values of all hyperparameters. Input: Number of batches B Initialize: d DMD 0 foreach l = 1, . . . , B do Sample a batch of real time series X and generated time series e X. |