CaRiNG: Learning Temporal Causal Representation under Non-Invertible Generation Process
Authors: Guangyi Chen, Yifan Shen, Zhenhao Chen, Xiangchen Song, Yuewen Sun, Weiran Yao, Xiao Liu, Kun Zhang
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through experiments conducted on synthetic datasets, we validate that our Ca Ri NG method reliably identifies the causal process, even when the generation process is non-invertible. Moreover, we demonstrate that our approach considerably improves temporal understanding and reasoning in practical applications. Code can be accessed through https: //github.com/sanshuiii/Ca Ri NG. |
| Researcher Affiliation | Collaboration | 1Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE 2Carnegie Mellon University, Pittsburg, US 3Salesforce, San Francisco, US. |
| Pseudocode | No | The paper does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | Yes | Code can be accessed through https: //github.com/sanshuiii/Ca Ri NG. |
| Open Datasets | Yes | We conducted the experiments in two simulated environments, utilizing the available ground truth latent variables to evaluate identifiability. Subsequently, we assessed Ca Ri NG on a real-world Video QA task, SUTD-Traffic QA (Xu et al., 2021), to verify its capability in representing complex and non-invertible traffic events. |
| Dataset Splits | No | For SUTD-Traffic QA, the paper states '56,460 QA pairs are used for training and the rest 6,075 QA pairs are used for testing,' but does not explicitly provide the size or percentage for a validation dataset split. While Figure 4 (c) mentions 'validation MCC curves', the text itself does not detail the validation split. |
| Hardware Specification | No | The paper does not explicitly describe the hardware used for running its experiments, such as specific GPU or CPU models. |
| Software Dependencies | Yes | The models were implemented in Py Torch 1.11.0. |
| Experiment Setup | Yes | An Adam W optimizer is used for training this network. We set the learning rate as 0.001 and the mini-batch size as 64. We train each model under four random seeds (770, 771, 772, 773) and report the overall performance with mean and standard deviation across different random seeds. |