A curated, continuously updated collection of academic papers focusing on "latent space" methodologies across various AI domains. It targets researchers, engineers, and practitioners interested in understanding and leveraging latent space representations for enhanced reasoning, perception, and action in Large Language Models (LLMs), Vision-Language Models (VLMs), Vision-Language-Action models (VLAs), and Multi-Agent Systems (MAS). The primary benefit is a centralized, organized resource for cutting-edge research in this specialized area.

How It Works

The project functions as a manually curated bibliography, categorizing research papers into distinct areas: LLM-based, VLM-based, VLA-based, and MAS-based latent space techniques. Papers are sorted chronologically within each category, providing a temporal overview of advancements. For each entry, it lists the paper title, publication date, a brief introduction (often omitted), and a direct link to associated code repositories (primarily GitHub) where available. This structured approach facilitates efficient discovery and access to relevant research.

Quick Start & Requirements

This repository is a curated list of research papers and does not contain executable code or software for installation. It serves as a reference guide for academic literature.

Highlighted Details

• Comprehensive categorization of latent space research into LLM, VLM, VLA, and MAS domains.
• Focus on recent advancements, with papers primarily dated from late 2024 through late 2025.
• Inclusion of direct links to GitHub repositories for many listed papers, enabling access to implementations.

Maintenance & Community

The repository is marked as "continuously updated" and released its initial version on November 30, 2025. Contributions of new resources are welcomed via pull requests, with a WeChat group available for further discussion and issue reporting.

Licensing & Compatibility

The README does not specify a license. Compatibility for commercial use or closed-source linking is undetermined without a defined license.

Limitations & Caveats

As a manually curated list, the completeness and accuracy of the paper selection are subjective. Many entries lack introductory descriptions, and code availability varies, with several papers marked as having no associated code. The focus is solely on academic literature, not on providing a runnable software framework.