Local Gains and Fixed-Assignment Set Losses in Shared Set Decoders

2026-08-11

AI Summary

This study addresses the inherent conflict between local editing gains and set utility loss in DETR’s shared set decoder. Through pairwise relation unit analysis and composite contrastive experiments, we investigate how query relation deletion impacts prediction sets. Results reveal that successful local interventions do not guarantee optimal joint decoding, with efficacy heavily dependent on read operators and selection condition sensitivity. Furthermore, we confirm an inverse correlation between local gains and fixed assignment losses. Although no intervention-invariant marginal mechanisms were identified, this work establishes the dominance of global set constraints over local edits, offering novel insights into the internal coupling mechanisms of set-based decoders.

Abstract

A query-relation deletion can improve the edited slot while reducing the utility of the prediction set that contains it. We study this tension in two related ResNet-50 DETR-family checkpoints using recorded, selection-conditional evidence from 710 paired image-relation units per checkpoint. The primary comparison subtracts a matched active control, which deletes the same leader source at a different recorded recipient, from the selected target deletion. It is therefore a composite contrast rather than a same-recipient placebo. The target-minus-control contrast is locally positive and fixed-assignment negative in both checkpoints. The opposite-sign pattern occurs within 302/710 DETR units and 460/710 DINO units. After rematching, the corresponding counts are 285/710 and 433/710. Rematching and native selection absorb enough of the mean loss for DETR intervals to cross zero, whereas DINO intervals remain negative, so persistence across readouts differs by checkpoint. A fixed-map comparison between hard deletion and a mass-preserving edit also differs before rematching. That comparison is conditional on the outcome-blind map and does not establish same-dose transport. Local intervention success therefore does not determine the consequence for a jointly decoded set. The supported conclusion is selection-conditional deletion sensitivity whose persistence depends on the readout and intervention operator. We do not identify an intervention-invariant edge mechanism, detector-level degradation, population prevalence, or the value of a training-time regularizer.

Problem

• Shared Set Decoders
• Query-relation deletion
• Local Gains
• Fixed-Assignment Set Losses
• Selection-conditional sensitivity

Innovation

• Shared Set Decoders
• Selection-Conditional Evidence
• Composite Contrast
• Local Gains vs Fixed-Assignment Losses
• DETR-family Checkpoints