Research
MemeBLIP2: A Lightweight Multimodal System to Detect Harmful Memes
A lightweight multimodal system that pairs the BLIP-2 encoder with joint image-text modelling to detect harmful memes (IJCAI 2025 workshop paper).
A lightweight multimodal system that pairs the BLIP-2 encoder with joint image-text modelling to detect harmful memes (IJCAI 2025 workshop paper).
Code and results behind our MLNLP 2025 paper: DistilBERT with plain cross-entropy is a strong default for classifying medical abstracts.