Data Selection Through Iterative Self-Filtering for Vision-Language Settings

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Andrei Liviu Nicolicioiu, Sarvjeet Singh Ghotra, Morgane M. Moss, Aaron Courville
• Venue: arXiv: cs.CV
• Published: 2026-06-22
• Collected: 2026-06-23
• URL: http://arxiv.org/abs/2606.23611v1
• PDF: https://arxiv.org/pdf/2606.23611v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers propose a new method to select data for training neural networks in large scale, noisy datasets. The Self-Filtering method iteratively trains a CLIP model on an evolving dataset, balancing clean samples and diverse samples from the entire distribution.

What the Document Says

• Large amounts of clean data are crucial for training neural networks.
• Manual oversight is impractical at large scales, resulting in noisy datasets.
• Previous attempts to mitigate this issue have involved heuristics, curated reference datasets, and pre-trained models.
• A novel bootstrapped method is proposed that trains a CLIP model on an evolving, self-selected dataset.
• This dataset balances filtered clean samples and diverse samples from the entire distribution.
• The Self-Filtering method iterates between training the model and selecting a subsequently improved data mixture.
• Training on vision-language datasets filtered by the proposed approach improves downstream performance without the need for additional data or pre-trained models.

Key Concepts Mentioned

• Neural networks
• Large scale data
• Noisy data
• Manual oversight
• CLIP model
• Self-Filtering method
• Data selection
• Vision-language datasets
• Downstream performance

Claims Made by the Document

• The proposed Self-Filtering method improves downstream performance.
• The method does not require additional data or pre-trained models.

Evidence Used

• Improved downstream performance is observed through training on vision-language datasets filtered by the proposed approach.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

The availability of large amounts of clean data is paramount to training neural networks. However, at large scales, manual oversight is impractical, resulting in sizeable datasets that can be very noisy. Attempts to mitigate this obstacle to producing performant vision-language models have so far involved heuristics, curated reference datasets, and using pre-trained models. Here we propose a novel, bootstrapped method in which a CLIP model is trained on an evolving, self-selected dataset. This evolving dataset constitutes a balance of filtered, highly probable clean samples as well as diverse samples from the entire distribution. Our proposed Self-Filtering method iterates between training the model and selecting a subsequently improved data mixture. Training on vision-language datasets filtered by the proposed approach improves downstream performance without the need for additional data or pre-trained models.