Workshop Papers Accepted at LatinX in AI Research at NeurIPS 2024
Table of Contents
I am thrilled to share that two of our papers have been accepted at the LatinX in AI Research Workshop at NeurIPS 2024! These works push the boundaries in causal inference and generative modeling, contributing new methodologies and insights to the field.
1. Human-Aided Discovery of Ancestral Graphs #
Authors: Tiago Silva, Eliezer de Souza da Silva, António Góis, Dominik Heider, Samuel Kaski, Diego Mesquita, Adele H Ribeiro
Summary:
We introduce Ancestral GFlowNets (AGFNs), a fully probabilistic causal discovery method that samples ancestral graphs proportionally to a belief score, enabling uncertainty quantification. The method also incorporates expert knowledge via importance sampling, significantly improving inference in data-scarce scenarios with latent confounders.
2. On Divergence Measures for Training GFlowNets #
Authors: Tiago Silva, Eliezer de Souza da Silva, Diego Mesquita
Summary:
We analyze why traditional divergence measures fail for training Generative Flow Networks (GFlowNets) and propose variance-reducing techniques for gradient estimation. These methods accelerate training convergence and align GFlowNets more closely with hierarchical variational inference frameworks.
I am excited to present these contributions at NeurIPS 2024 and to engage with the community on these topics. A big thank you to my collaborators for their exceptional work!
Stay tuned for more updates from the conference!