FT-NCFM: An Influence-Aware Data Distillation Framework for Efficient VLA Models

2025-11-20

AI Summary

Vision-Language-Action (VLA) models suffer from limited generalization due to heavy reliance on large-scale, redundant datasets; centralized optimization paradigms fail to fundamentally alleviate this data bottleneck. Method: We propose a data-centric generative distillation framework featuring (i) a novel fact-tracking engine that integrates causal attribution with programmable contrastive validation to enable interpretable, quantitative assessment of sample intrinsic value; (ii) an adversarial Neural Concept Formation Module (NCFM) for generating model-agnostic, information-dense, high-fidelity synthetic data; and (iii) construction of a minimal yet sufficient core dataset. Results: On mainstream VLA benchmarks, our distilled dataset—comprising only 5% of the original training data—achieves 85–90% of the full-data success rate, reduces training time by over 80%, and significantly enhances data efficiency and deployment feasibility.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The powerful generalization of Vision-Language-Action (VLA) models is bottlenecked by their heavy reliance on massive, redundant, and unevenly valued datasets, hindering their widespread application. Existing model-centric optimization paths, such as model compression (which often leads to performance degradation) or policy distillation (whose products are model-dependent and lack generality), fail to fundamentally address this data-level challenge. To this end, this paper introduces FT-NCFM, a fundamentally different, data-centric generative data distillation framework. Our framework employs a self-contained Fact-Tracing (FT) engine that combines causal attribution with programmatic contrastive verification to assess the intrinsic value of samples. Guided by these assessments, an adversarial NCFM process synthesizes a model-agnostic, information-dense, and reusable data asset. Experimental results on several mainstream VLA benchmarks show that models trained on just 5% of our distilled coreset achieve a success rate of 85-90% compared with training on the full dataset, while reducing training time by over 80%. Our work demonstrates that intelligent data distillation is a highly promising new path for building efficient, high-performance VLA models.

Problem

• Addresses data redundancy and inefficiency in Vision-Language-Action model training
• Overcomes limitations of model compression and policy distillation methods
• Creates compact, reusable datasets to maintain performance with reduced training

Innovation

• Data distillation framework using causal attribution
• Adversarial process synthesizes model-agnostic data asset
• Fact-tracing engine assesses intrinsic sample value