Probabilistic Modeling of Interpersonal Coordination Processes
Authors: Paulo Soares, Adarsh Pyarelal, Meghavarshini Krishnaswamy, Emily Butler, Kobus Barnard
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our model in the context of three-person teams executing a virtual search and rescue (SAR) mission. We first use synthetic data to confirm that our technical definition of coordination is consistent with expectations and that we can recover generated coordination despite noise. We then show that captured coordination can be predictive of team performance on real data. Here we use speech vocalics and semantics to infer coordination for 36 teams carrying out two successive SAR missions. In two different datasets, we find that coordination is generally predictive of team score for the second mission, but not for the first, where teams are largely learning to play the game. |
| Researcher Affiliation | Academia | 1University of Arizona. Correspondence to: Paulo Soares <paulosoares@arizona.edu>, Adarsh Pyarelal <adarsh@arizona.edu>. |
| Pseudocode | No | The paper does not contain any explicit pseudocode blocks or algorithm sections. |
| Open Source Code | Yes | Our code and preprocessed datasets are available online at https: //github.com/ml4ai/tomcat-coordination. |
| Open Datasets | Yes | We evaluate our model on two distinct datasets based on the same task, but with different participants and experimental procedures. The first is the ASIST Study 3 dataset (Huang et al., 2022a), and the second is the To MCAT dataset (Pyarelal et al., 2023). |
| Dataset Splits | Yes | Here we split the data into training and test splits using leave-one-out cross-validation (LOOCV). |
| Hardware Specification | Yes | All experiments were run on a machine with 128 AMD EPYC 7542 CPU cores and inference runs took 10 minutes per trial on average. |
| Software Dependencies | Yes | We implemented the models using Py MC 5.0.2 (Salvatier et al., 2016)... To identify local maxima in the coordination series, we use the function signal.find peaks from the Sci Py library (Virtanen et al., 2020) with the parameter width set to 5 to ignore sharp peaks of short duration. |
| Experiment Setup | Yes | We implemented the models using Py MC 5.0.2 (Salvatier et al., 2016), with 2000 warm-up iterations, 2000 samples, 4 parallel chains, a target acceptance probability of 0.9 and default values for the remaining parameters. ... In experiments with real data, we set 𝜎2 𝑂2 to 5.0. ... Thus, we set 𝜎𝑢= 0.5 and 𝜎𝑂1 = 0.1 (values to which the model is not particularly sensitive). We employed a Half Normal(1) as a hyper-prior for 𝜎𝐴1, N (0, 5) for 𝜇𝑢0, and N (0, 1) for 𝜇1,𝑝 𝐴0 . |