Reproduce “User Modelling for Avoiding Overfitting in Interactive Knowledge Elicitation for Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pedram Daee, Tomi Peltola, Aki Vehtari, Samuel Kaski
Researchers at Aalto University developed a user modeling methodology for interactive machine learning that infers a user's latent knowledge from their observed feedback, explicitly accounting for the machine's influence derived from training data. This approach effectively mitigates overfitting and yielded a statistically significant improvement in predictive performance on a sentiment analysis task, reducing the mean squared error on test data.
Key insights
- User feedback in interactive settings is often not independent but influenced by the machine's presentation of training data-derived information.
- By modeling the user as a rational Bayesian agent, it is possible to statistically de-bias observed feedback and infer the user's independent, latent knowledge.
- Correctly disentangling latent user knowledge from machine influence is crucial for preventing overfitting and improving the generalization ability of models in HILML.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.