research
Hard Negatives Stayed Diverse Instead of Collapsing to a Few
FlowNeg uses a hierarchical generative flow network to sample informative knowledge-graph counterexamples across modes.
Summary
FlowNeg uses a hierarchical generative flow network to sample informative knowledge-graph counterexamples across modes.
Across a five-seed grid of five architectures and five benchmarks, FlowNeg had higher mean reciprocal rank than two comparison methods in 24 of 25 cells. A separate 15-seed control on FB15k-237 with RotatE reported 0.359 versus 0.346 MRR, with fixed diagnostic and compute budgets.
Why it matters
FlowNeg uses a hierarchical generative flow network to sample informative knowledge-graph counterexamples across modes.
Limits and context
No additional limitation was separately recorded.
Key claims
FlowNeg uses a hierarchical generative flow network to sample informative knowledge-graph counterexamples across modes.
Evidence: source-2026-08-26-014
Sources
- arXiv preprint 2608.23849arXiv · primary research
Corrections
No corrections have been recorded for this story.