Moderated and Drifting Linear Dynamical Systems

Authors: Jinyan Guan, Kyle Simek, Ernesto Brau, Clayton Morrison, Emily Butler, Kobus Barnard

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our approach on a real dataset of self-recalled emotional experience measurements of heterosexual couples engaged in a conversation about a potentially emotional topic, with body mass index (BMI) being considered as a moderator. We evaluate several models on their ability to predict future conversation dynamics (the last 20% of the data for each test couple), with shared parameters being learned using held out data. We validate the hypothesis that BMI affects the conversation dynamic in the experimentally chosen topic.
Researcher Affiliation Academia 1 Department of Computer Science, University of Arizona 2 Department of Computer Science, Boston College 3 School of Information: Science, Technology, and Arts, University of Arizona 4 Norton School of Family and Consumer Sciences, University of Arizona
Pseudocode Yes Algorithm 1 Sample procedure from p(Q, σ, Θ, X1|Y) ... Algorithm 2 Sample procedure from p(Q, σ, s, Θ, X1|Y)
Open Source Code No The paper mentions providing web services: 'Predoehl, Andrew, Guan, Jinyan, Butler, Emily, and Barnard, Kobus. Comp Ties web app, 2015. URL http://www.compties.org/COM.html.' This links to a web application, not to the open-source code for the methodology described in the paper.
Open Datasets Yes The real data is composed of recalled self-rating emotion experience of 38 heterosexual couples with different joint weight status during an emotional conversation in a social experiment lab setting as reported by Reed et al. (2015).
Dataset Splits Yes We develop a multi-stage evaluation procedure to learn the shared model parameters Q, σ and s and evaluate the predictive power of the learned models using 9-fold cross validation. First, we randomly divide the couples into nine groups. ... we use 100 samples from the posterior for the first 80% of time to compute 100 estimates for each time point in the next 20% (held out) of time for each testing couple.
Hardware Specification No The paper does not provide any specific hardware details such as GPU models, CPU models, or memory specifications used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., 'Python 3.8', 'PyTorch 1.9') used for the experiments.
Experiment Setup Yes The number of sampling iterations was 30,000 for parameter learning and 100,000 for parameter fitting. For model comparison, we provide the fitting and prediction errors for three base line models: predicting the last 20% of each couple by 1) a line that fitted to each partner s first 80% observations; 2) the average of the first 80% observations; and 3) a fitted CLO model by maximum likelihood estimation (MLE) of p(Y|Θ) over the first 80% time points. ... We somewhat arbitrarily set 0.5 as the sigma of the observation noise during both parameter learning and fitting.