New framework enhances dataset distillation robustness under noisy supervision

Researchers have developed a new framework for dataset distillation (DD) designed to improve robustness under noisy supervision. The proposed method, called Robust Trajectory Distillation, combines two components: Selective Guidance Reweighting (SGR) and Teacher-Inspired Auxiliary Targets (TIAT). SGR progressively reweights samples by fusing global forgetting patterns with local neighborhood consistency, prioritizing cleaner supervision along the teacher's learning trajectory. TIAT injects auxiliary guidance from intermediate teacher model dynamics to reinforce informative signals. This approach aims to produce distilled datasets that are cleaner, richer, and more transferable without requiring relabeling or clean subsets, showing consistent gains over existing DD methods across various noise types. AI

IMPACT This research could lead to more efficient and reliable training of AI models by improving the quality of distilled datasets, especially in scenarios with imperfect data.

RANK_REASON The cluster contains an academic paper detailing a new research framework. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Robust Trajectory Distillation
• Selective Guidance Reweighting
• Teacher-Inspired Auxiliary Targets

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →