Summary
Multimodal embeddings help computers understand and relate information that comes in different forms, like text and images. The authors found that previous methods let computers guess explanations for their inputs without clear guidance, which limits performance. They developed a new approach called PILOT that guides the computer to explore better reasoning during training by using extra clues about how information relates. This approach leads to better performance on many tasks involving images and text, especially in retrieving visual documents. Their work shows that teaching models how to reason more effectively during training matters more than just rewarding the outcome afterward.
What this means in practice
- •For search engine engineers: Improve search relevance by training embedding models that explore guided reasoning to better match complex queries with images and documents.
- •For digital asset management teams: Enhance retrieval of visual documents by adopting reasoning-enhanced embeddings to better connect text descriptions and image content.
Authors
Tianyu Yang, Yijun Tian, Lars Benedikt Kaesberg, Jan Philip Wahle, Bela Gipp, Terry Ruas, Daniel Kurzawe
Abstract
Advances in Multimodal Large Language Models (MLLMs) have driven substantial progress in multimodal embeddings. Recent methods exploit generative capabilities by augmenting embedded inputs with self-generated reasoning to produce more informative embeddings. However, conventional retrieval supervision for training embedding models specifies pairwise relevance relations without explicitly guiding the reasoning used for input augmentation. Such reasoning is neither directly observed nor uniquely determined by the input, allowing multiple plausible traces to introduce semantic cues of varying retrieval utility. Existing approaches largely optimize reasoning conditioned on the input alone, limiting the exploration of retrieval-relevant traces. We introduce PILOT, a framework that models reasoning as a latent variable and instantiates its prior and posterior distributions as distinct reasoning modes of a shared MLLM. The prior conditions only on the input, whereas the posterior uses additional training-time relational evidence to guide reasoning exploration. To translate the posterior's broader exploration into inference-time improvements, we derive an importance-weighted variational objective and instantiate it as a prior-posterior co-evolution framework that improves both modes under a shared inference-compatible reward while selectively transferring posterior trajectories to the prior according to their utility and prior reachability. On the 78-task MMEB-V2 benchmark, PILOT outperforms the primary baseline across all 12 task groups at both 2B and 7B parameters, improving overall scores by 3.2 and 3.3 points, respectively, with the largest gains in visual-document retrieval (+6.1 and +6.4 points). Our results show that relational supervision is most valuable not as a post-hoc reward but as a signal for shaping how reasoning is explored in training reasoning-enhanced embedding models.