Search papers, labs, and topics across Lattice.
Department of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran
4
0
8
17
VLMs can flip predictions in nearly half of cases due to simple changes in presentation order, revealing hidden vulnerabilities in clinical reliability.
MDP-GRPO improves constraint satisfaction in reinforcement learning by up to 5.0% while ensuring stable convergence even with small group sizes.
Current audio-language models are culturally tone-deaf: they can't even detect Persian poetry meter, despite crushing English speech tasks.
LLMs aren't always the answer: a lightweight BERT model beats them at Persian punctuation restoration, avoiding over-correction and saving compute.
