R1-SyntheticVL: Is Synthetic Data from Generative Models Ready for Multimodal Large Language Model?

TL;DR

This paper introduces CADS, a novel collective adversarial approach for synthesizing high-quality multimodal data to improve large language models, demonstrating superior benchmark performance.

Contribution

We propose CADS, a new collective adversarial data synthesis method that generates diverse, challenging multimodal data to enhance multimodal large language models.

Findings

CADS produces high-quality, diverse multimodal data.

R1-SyntheticVL outperforms existing models on benchmarks.

The approach effectively improves MLLM performance.

Abstract

In this work, we aim to develop effective data synthesis techniques that autonomously synthesize multimodal training data for enhancing MLLMs in solving complex real-world tasks. To this end, we propose Collective Adversarial Data Synthesis (CADS), a novel and general approach to synthesize high-quality, diverse and challenging multimodal data for MLLMs. The core idea of CADS is to leverage collective intelligence to ensure high-quality and diverse generation, while exploring adversarial learning to synthesize challenging samples for effectively driving model improvement. Specifically, CADS operates with two cyclic phases, i.e., Collective Adversarial Data Generation (CAD-Generate) and Collective Adversarial Data Judgment (CAD-Judge). CAD-Generate leverages collective knowledge to jointly generate new and diverse multimodal data, while CAD-Judge collaboratively assesses the quality of synthesized data. In addition, CADS introduces an Adversarial Context Optimization mechanism to optimize the generation context to encourage challenging and high-value data generation. With CADS, we construct MMSynthetic-20K and train our model R1-SyntheticVL, which demonstrates superior performance on various benchmarks.