Kenji Kawaguchi

A new approach enhances prompt diversity for safer language models.

A method to reduce the size of large language models while maintaining their performance.

Forward Gradient Unrolling offers efficient solutions for optimizing large machine learning models.

This approach uses self-evaluation to guard against harmful outputs in language models.

Research shows simple input changes can lead to harmful outputs in LLMs.

This paper proposes a method to convert ICL into model weights for improved performance.

New methods enhance solving complex fractional partial differential equations.

Researchers develop methods to solve complex FPL equations more efficiently.

This article examines how format bias affects language model performance and suggests improvement strategies.

Mamba offers a new way to analyze complex systems using machine learning.

This article examines how different layers affect LLM performance.

Typos can greatly confuse advanced language models and affect their responses.

Explore how AnchorAttention improves efficiency in processing long texts with language models.

New method streamlines complex mathematics for faster, simpler calculations.

A new method for efficiently pruning image-generating AI models while preserving quality.