Sample efficiency of positional task conditioning relative to rationale-based distillation

Determine whether Positional Task Conditioning improves sample efficiency relative to Rationale-Based Distillation when training data size is varied for multi-task product-family defect detection.

Background

The paper compares PTC with Rationale-Based Distillation (RBD) using identical training data and hyperparameters, showing that PTC generally achieves higher F1 scores across the evaluated LLMs. However, the study does not conduct ablations that vary the amount of training data.

Consequently, the reported results establish comparative performance at one dataset scale but do not resolve whether PTC reaches a given performance level with fewer examples than RBD. A training-size analysis is needed to characterize the relative data efficiency of the two distillation approaches.

References

Consequently, whether PTC enhances sample efficiency relative to RBD is a question we leave to future research.

— Positional task conditioning for scalable defect detection across product families in large product catalogs  (2609.09567 - Satyadharma et al., 9 Sep 2026) in Section 6, Limitations, third limitation paragraph