Paula Ontalvilla

We are in #coling2025!

.@paula_onta presenting the paper "Improving the Efficiency of Visually Augmented Language Models" at #coling2025

Joint work with @aormazabalo @gazkune @IxaGroup @Hitz_zentroa

So, visual grounding doesn't have to mean working with images. With the right representations, we can make models simpler, faster, and stronger—all while staying efficient!

I am happy to announce that our latest work on Visually Augmented Language Models has been accepted in #COLING2025 We present Blind-VaLM, a simple approach removing the need for explicit images to visually augment an LM, while still achieving strong performance on VLU and NLU.