Lattice AI Research

Research focus

Red-Teaming & Adversarial Robustness (2)Multimodal Models (1)Reasoning & Chain-of-Thought (1)RLHF & Preference Learning (1)

Frequent co-authors

Toshiaki Koike-Akino (2)Vanshaj Khattar (1)Md. Rafi Ur Rashid (1)Moumita Choudhury (1)

Papers (2)

Mar 16, 2026

T. Koike-Akino +12w ago

Directional Embedding Smoothing for Robust Vision Language Models

Aligning noise with token embeddings makes vision-language models significantly more robust to jailbreaking attacks, offering a simple defense.

T. Koike-Akino, Toshiaki Koike-Akino

Multimodal Models Red-Teaming & Adversarial Robustness

2w ago

Amplification Effects in Test-Time Reinforcement Learning: Safety and Reasoning Vulnerabilities

Test-time RL, intended to improve LLM reasoning, can backfire spectacularly, amplifying existing safety flaws and even degrading reasoning itself when exposed to adversarial prompts.

Vanshaj Khattar, Md. Rafi Ur Rashid, Moumita Choudhury +3

Reasoning & Chain-of-Thought Red-Teaming & Adversarial Robustness RLHF & Preference Learning

Search

T. Koike-Akino

Publication activitypapers/week, last 8 weeks

Research focus

Frequent co-authors

Papers (2)