Identification of Nonlinear Latent Hierarchical Models
Authors: Lingjing Kong, Biwei Huang, Feng Xie, Eric Xing, Yuejie Chi, Kun Zhang
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We accompany our theory with an estimation method that can asymptotically identify the causal structure and latent variables for nonlinear latent hierarchical models and validate it on multiple synthetic and real-world datasets. In this section, we present experiments to corroborate our theoretical results in Section 3. We start with the problem of recovering the basis model in Section 4.2, which is the foundation of the overall identifiability. In Section 4.3, we present experiments for hierarchical models on a synthetic dataset and two real-world datasets. |
| Researcher Affiliation | Academia | 1 Carnegie Mellon University 2University of California San Diego 3Beijing Technology and Business University 4Mohamed bin Zayed University of Artificial Intelligence |
| Pseudocode | Yes | Algorithm 1 Identification of Latent Hierarchical Models. |
| Open Source Code | No | The paper describes the implementation details of the estimation model and refers to Appendix E for training configurations, but it does not include an explicit statement about making the source code available or provide a link to a repository. |
| Open Datasets | Yes | The personality dataset was curated through an interactive online personality test [Project, 2019]. For the digit dataset, we construct a multi-view digit dataset from MNIST [Deng, 2012]. |
| Dataset Splits | No | The paper mentions using 'training samples' for experiments and evaluating R2 scores over '8192 samples for each estimated variable pair', but it does not specify explicit training, validation, and test dataset splits with percentages, sample counts, or clear methodologies for partitioning. |
| Hardware Specification | No | The paper discusses deep learning training and model architectures but does not provide any specific details about the hardware used, such as GPU or CPU models, memory, or specific computing clusters. |
| Software Dependencies | No | The paper mentions using Adam for training and parameterizing models with MLPs and Leaky-ReLU activations. It also refers to a pretrained ResNet-44 model. However, it does not provide specific version numbers for any software dependencies, such as deep learning frameworks or programming languages. |
| Experiment Setup | Yes | Training configurations can be found in Appendix E. We apply Adam to train each model for 20,000 steps with a learning rate of 1e-3. Estimation model configuration for each experiment. We repeat each experiment over at least 3 random seeds. We evaluate the R2 score with kernel regression with Gaussian kernel. |