From the 1 of 16 linked papers with an AI index.
16 papers
Visual Access Boundaries in Vision-Language Model Reasoning
Hiroto Osaka, Shohei Taniguchi, Gouki Minegishi +3
The paper investigates how chain-of-thought prompting works in vision-language models by introducing a visual access sweep that masks attention to image tokens, defining a visual a…
On Advantage Estimates for Max@K Policy Gradients
Shota Takashiro, Soichiro Nishimori, Paavo Parmas +6
Reinforcement learning with verifiable rewards is widely used for post-training reasoning models, but sparse outcome rewards make exploration difficult. A complementary approach is…
Steering at the Source: Style Modulation Heads for Robust Persona Control
Yoshihiro Izawa, Gouki Minegishi, Koshi Eguchi +2
Activation steering offers a computationally efficient mechanism for controlling Large Language Models (LLMs) without fine-tuning. While effectively controlling target traits (e.g.…
Zipping the Thought: When and How Compressed Reasoning Data Works in LLM Post-Training
Kohsei Matsutani, Gouki Minegishi, Takeshi Kojima +2
Large language models (LLMs) can now solve complex problems through long chain-of-thought (CoT) reasoning, but the trade-off between performance and token cost remains a central ch…
Emergent Analogical Reasoning in Transformers
Gouki Minegishi, Jingyuan Feng, Hiroki Furuta +3
Analogy is a central faculty of human intelligence, enabling abstract patterns discovered in one domain to be applied to another. Despite its central role in cognition, the mechani…
RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs
Kohsei Matsutani, Shota Takashiro, Gouki Minegishi +3
Large language models (LLMs) are typically trained by reinforcement learning (RL) with verifiable rewards (RLVR) and supervised fine-tuning (SFT) on reasoning traces to improve the…