Identification of Intermittent Temporal Latent Process
Authors: Yuke Li, Yujia Zheng, Guangyi Chen, Kun Zhang, Heng Huang
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experiments on both synthetic and real-world datasets verify our theoretical claims. 5 EXPERIMENTS Experimental Setup To evaluate Inter Latent ability to learn causal processes and identify latent variables in non-invertible scenarios, we conduct simulation experiments using random causal structures with specified sample and variable sizes. Results Figure 3 summarizes our main results on our simulations. Ablation Study and Discussions To elucidate the key assumptions of our data generating process in Eq. 1, we further conduct ablation study focusing on the impact of sparse support. 5.2 REAL-WORLD EXPERIMENTS Task setup To evaluate our proposed identification theories in complex real-world scenarios, we apply them to the task of Group Activity Recognition (GAR) using the Volleyball dataset Ibrahim et al. (2016). |
| Researcher Affiliation | Academia | 1 University of Maryland College Park, College Park, MD, USA 2 Carnegie Mellon University, Pittsburgh PA, USA |
| Pseudocode | No | The paper describes the approach and network design in text, but it does not include any explicitly labeled pseudocode or algorithm blocks. Figure 2 illustrates the overall framework but is not a pseudocode representation. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its own source code, nor does it include a link to a code repository for the Inter Latent methodology. It mentions using third-party tools like MMDetection toolbox and Res Net-18, but this is not their own implementation code. |
| Open Datasets | Yes | We evaluate our approach on synthetic and real-world datasets, demonstrating its effectiveness in uncovering complex hidden temporal processes, as well as validating the proposed identifiability theory. Task setup To evaluate our proposed identification theories in complex real-world scenarios, we apply them to the task of Group Activity Recognition (GAR) using the Volleyball dataset Ibrahim et al. (2016). The Volleyball dataset Ibrahim et al. (2016) contains 55 video recordings of volleyball games and is split into 3493 training clips and 1337 testing clips. |
| Dataset Splits | Yes | The Volleyball dataset Ibrahim et al. (2016) contains 55 video recordings of volleyball games and is split into 3493 training clips and 1337 testing clips. |
| Hardware Specification | Yes | All experiments were conducted on a single NVIDIA Ge Force RTX 2080 Ti GPU with 11GB meory. Training is conducted for 80 epochs on a multi-GPU setup consisting of four NVIDIA Ge Force RTX 2080 Ti GPUs, providing a total of 44GB of meory. |
| Software Dependencies | Yes | We implemented our models using Py Torch 1.11.0. |
| Experiment Setup | Yes | The hyperparameters were set as follows: learning rate of 1e-3 and minibatch size of 64. The loss function balances reconstruction error and KL-divergence, with the latter weighted by β = 0.02. We employ the Adam W optimizer with cosine annealing for training our network. The initial learning rate is set to 2e-3, with a weight decay of 1e-2 to mitigate overfitting. For all video sequences in Volleyball dataset, we uniformly sample T = 10 frames as input. The ELBO loss is computed with a β value of 0.02. We utilize a batch size of 128, which we found to provide a good trade-off between computational efficiency and optimization stability. Training is conducted for 80 epochs. |