Trail-Weighted Influence Scores for Approximate Outcome Prediction in Bayesian Networks

Key points

• A traversal enumerates the active trails from each evidence node to the outcome, scores each trail by the product of its edge strengths, and drops evidence whose score falls below a threshold.1
• We prove this bound, check it exhaustively on all small trees, and show that it fails when dropped trails share a branch.1
• Keeping the top half of the active evidence by trail score gives lower error than random pruning on all seven, but it is not clearly better than a mutual-information ranking, and scoring costs more than the inference it saves on six of the seven, so we show no speedup.1
• Models trained on trail, graph and information features and tested on networks they never saw do not beat the trail score or mutual information, but trail features carry most of what they use and clearly outperform generic graph features.1

Sources (1)

• [1]Trail-Weighted Influence Scores for Approximate Outcome Prediction in Bayesian Networks
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 10:46 PM

  We study trail-weighted pruning of evidence for Bayesian-network prediction.

  A traversal enumerates the active trails from each evidence node to the outcome, scores each trail by the product of its edge strengths, and drops evidence whose score falls below a threshold.

Extractive summary: sentences quoted from the sources.