Vision-language model improves itself by verifying its own answers
Trinity: Self-Evolving Vision-Language Models with a Self-Verifier
Artificial Intelligence
Summary
Building smarter vision-language models usually needs lots of labeled training data, which is expensive to create. The authors present Trinity, a model that teaches itself by looking at unlabeled images, asking questions, answering them, and then verifying the answers all on its own. Trinity uses a self-verifier to check if questions are relevant to the images and if answers are correct before learning from them. This approach helps the model get better at complex reasoning tasks in science and math without needing any human-labeled data.
What this means in practice
- •For machine learning engineers: Automatically improve vision-language models using only unlabeled image collections without manual labeling effort.
- •For biotech data teams: Enhance scientific image understanding and reasoning in biology datasets without needing expert annotations.
Authors
Youngwan Lee, Yong-Ju Lee, Sung Ju Hwang
Abstract
Self-evolving vision-language models (VLMs), a form of self-improvement in which a model generates its own training data from unlabeled images, are a promising route toward agents that expand their reasoning capability in an unsupervised manner, without relying on ever-larger annotation budgets. Existing methods pair a Questioner that proposes problems with a Solver that answers them, but reward both roles mainly by agreement among sampled answers. Agreement is a weak proxy for truth: it cannot tell whether a question is grounded in the image, whether the proposed reference answer is right, or whether a confident majority is wrong in the same way. We present Trinity, in which one VLM plays three roles, Questioner, Solver, and Verifier, and the Verifier is a self-verifier: an exponential moving average (EMA) of the policy itself, requiring neither labels nor an external judge. The Verifier screens every generated question for image grounding and answer correctness before it becomes supervision, scores Solver reasoning against the image, and adjudicates disputes between the reference answer and a strong Solver consensus, correcting the reference and penalizing the Questioner when the consensus is right. Trained on images alone, Trinity improves Qwen3-VL-8B on mathematical visual reasoning and on science benchmarks with biology content, for example, +8.6 points on the biology split of SciVQR and +12.8 on MathVerse, and its reward dynamics behave as a healthy self-play curriculum should. These results suggest that a self-evolving multimodal agent can strengthen its scientific reasoning from unlabeled scientific images alone, with the model itself serving as the verifier.