AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

ExPO-HM: Learning to Explain-then-Detect for Hateful Meme Detection

Jingbiao Mei, Mingsheng Sun, Jinghong Chen, Pengda Qin, Yuhong Li, Da Chen, Bill Byrne

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Data and hallucination detection

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add policy-manual augmented SFT warmup (SFT-PM) (ablation)GRPO, Qwen2.5-VL-7B · HatefulMemesattack F1: 61.2 -> 70.8Improved
add GRPO curriculum learning (fine-grained before binary) (ablation)ExPO-HM, Qwen2.5-VL-7B · HatefulMemesbinary F1: 75.8 -> 78.4Improved
add CDE reward (ablation)ExPO-HM, Qwen2.5-VL-7B · HatefulMemesbinary F1 / LLM-judge: 81.1 vs 78.4; LLM 5.8 -> 6.2; CDE 0.056 -> 0.026Improved
use Explain-then-Detect post-training (SFT/DPO/GRPO) instead of direct SFTQwen2.5-VL-7B · HatefulMemesbinary F1: 75.0 -> 74.5 (GRPO), 73.6 (DPO)Hurt
add GRPO to Explain-then-Detect SFTQwen2.5-VL-7B · HatefulMemesreasoning LLM-judge: 5.0 -> 5.2No clear effect
use DPO instead of zero-shotQwen2.5-VL-7B · HatefulMemesreasoning LLM-judge: 5.0 -> 4.9Hurt
binary-only SFT warmup vs no warmup before GRPO-CL+CDE (ablation)ExPO-HM · HatefulMemesbinary F1 / attack F1: attack 69.3 -> 66.8Hurt
policy-manual warmup vs fine-grained label warmup (ablation)ExPO-HM · HatefulMemestarget F1: 73.4 -> 77.2Improved
ExPO-HM vs GRPOExPO-HM · HatefulMemesattack F1: 61.2 -> 75.6Improved
ExPO-HM on 3B model vs GRPOExPO-HM · PrideMMbinary F1: 72.1 -> 75.6Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy