ResearchBenchmarksVisual Question Answering on GQA (Relative Accuracy Drop %)For AI assistantsMarkdownContent indexVisual Question Answering on GQA (Relative Accuracy Drop %)Follow0.62Relative Accuracy Drop (%)HONES-0.81528.872418.5628.2476Apr 20, 2026Evaluation resultsMethodMethodLinksRelative Accuracy Drop (%)HONESMasked neuron group=Gen.Masked neuron group=Gen.2026.040.62HONESMasked neuron group=Tr...Masked neuron group=Triple2026.041.2HONESMasked neuron group=Spec.Masked neuron group=Spec.2026.046.9HONESMasked neuron group=Pa...Masked neuron group=Pair (VQA+Retrieval)2026.0436.5