Lost in Embeddings: Information Loss in Vision-Language Models
This addresses a bottleneck for researchers and practitioners using VLMs, as it identifies and quantifies a specific source of performance degradation, though it is incremental in focusing on an understudied aspect.
The paper tackled the problem of information loss in vision-language models due to projection steps, finding that connectors distort visual representations by 40-60% in k-nearest neighbor divergence, which correlates with degraded retrieval performance and predicts struggles in question-answering tasks.
Vision--language models (VLMs) often process visual inputs through a pretrained vision encoder, followed by a projection into the language model's embedding space via a connector component. While crucial for modality fusion, the potential information loss induced by this projection step and its direct impact on model capabilities remain understudied. We introduce two complementary approaches to examine and quantify this loss by analyzing the latent representation space. First, we evaluate semantic information preservation by analyzing changes in k-nearest neighbor relationships between image representations, before and after projection. Second, we directly measure information loss by reconstructing visual embeddings from the projected representation, localizing loss at an image patch level. Experiments reveal that connectors substantially distort the local geometry of visual representations, with k-nearest neighbors diverging by 40--60\% post-projection, correlating with degradation in retrieval performance. The patch-level embedding reconstruction provides interpretable insights for model behavior on visually grounded question-answering tasks, finding that areas of high information loss reliably predict instances where models struggle.