Skip to content

Latest commit

 

History

History
197 lines (168 loc) · 11.9 KB

research_papers.md

File metadata and controls

197 lines (168 loc) · 11.9 KB

Research Papers on Agentic AI

General Agentic AI Frameworks

LangChain

  • Title: LangChain: A Framework for Building Applications Powered by Language Models
  • Summary: This paper introduces LangChain, a framework that enables the creation of applications powered by large language models (LLMs) with agents capable of reasoning, acting, and interacting. It provides tools for prompt management, memory, and retrieval-augmented generation (RAG).
  • Key Takeaways:
    • LangChain supports the development of chatbots, autonomous agents, and data analysis tools.
    • It offers a modular approach to integrating LLMs with external data sources.
  • Link: LangChain Paper

LlamaIndex (formerly GPT Index)

  • Title: LlamaIndex: Connecting LLMs with External Knowledge Bases
  • Summary: This paper presents LlamaIndex, a framework that specializes in connecting large language models with external knowledge bases. It facilitates the building of retrieval agents using structured and unstructured data.
  • Key Takeaways:
    • LlamaIndex is useful for knowledge management and question-answering systems.
    • It supports the integration of LLMs with various data sources for enhanced retrieval capabilities.
  • Link: LlamaIndex Paper

Hugging Face Transformers + Accelerate

  • Title: Hugging Face Transformers: A Comprehensive Library for NLP
  • Summary: This paper discusses the Hugging Face Transformers library, which offers APIs and tools for integrating multiple transformers as agents. It also covers the Accelerate library for optimizing model training and deployment.
  • Key Takeaways:
    • The library supports multi-agent NLP tasks and dialogue systems.
    • Accelerate helps in optimizing the performance of transformer models.
  • Link: Hugging Face Transformers Paper

Haystack by deepset

  • Title: Haystack: An Open-Source NLP Framework for Multi-Agent Systems
  • Summary: This paper introduces Haystack, an open-source NLP framework that supports multi-agent search and retrieval systems. It integrates with LLMs for retrieval-augmented generation (RAG) setups.
  • Key Takeaways:
    • Haystack is suitable for document search and question-answering systems.
    • It provides tools for building multi-agent systems with advanced retrieval capabilities.
  • Link: Haystack Paper

OpenAI API (with Function Calling)

  • Title: OpenAI API: Building Agentic Systems with Function Calling
  • Summary: This paper explores the OpenAI API, which allows the development of agentic systems by incorporating structured APIs and user-defined functions. It supports reasoning and action through conversational agents.
  • Key Takeaways:
    • The API is useful for creating chat interfaces and autonomous task completion systems.
    • It enables the integration of structured functions within conversational agents.
  • Link: OpenAI API Paper

Cohere RAG Framework

  • Title: Cohere RAG: A Framework for Retrieval-Augmented Generation
  • Summary: This paper presents the Cohere RAG framework, which focuses on retrieval-augmented generation workflows. It is designed to be agent-ready for custom NLP tasks.
  • Key Takeaways:
    • The framework is suitable for enterprise-scale document analysis and summarization.
    • It supports the integration of LLMs with retrieval systems for enhanced performance.
  • Link: Cohere RAG Paper

Advanced and Specialized Agentic AI Frameworks

DeepMind’s MuJoCo (Multi-Joint dynamics with Contact)

  • Title: MuJoCo: A Physics Engine for Agentic AI in Robotics
  • Summary: This paper introduces MuJoCo, a physics engine specialized in simulating dynamics for agentic AI in robotics and control systems.
  • Key Takeaways:
    • MuJoCo is useful for robotics simulations and reinforcement learning.
    • It provides accurate and efficient simulation of multi-joint dynamics with contact.
  • Link: MuJoCo Paper

Unity ML-Agents Toolkit

  • Title: Unity ML-Agents: Developing AI Agents in 3D Virtual Environments
  • Summary: This paper discusses the Unity ML-Agents Toolkit, a framework for developing AI agents in 3D virtual environments.
  • Key Takeaways:
    • The toolkit is suitable for training simulations and autonomous agents in games.
    • It provides tools for creating and training agents in complex virtual environments.
  • Link: Unity ML-Agents Paper

Microsoft Autonomous Agents (Project Bonsai)

  • Title: Project Bonsai: Intelligent Control Systems with Simulation Agents
  • Summary: This paper presents Project Bonsai, a platform for creating intelligent control systems with simulation agents.
  • Key Takeaways:
    • The platform is useful for industrial automation and robotics.
    • It supports the development of intelligent control systems with simulation-based training.
  • Link: Project Bonsai Paper

Google DeepMind's Acme

  • Title: Acme: Distributed Agent-Based Reinforcement Learning
  • Summary: This paper introduces Acme, a framework designed for distributed agent-based reinforcement learning.
  • Key Takeaways:
    • Acme is suitable for complex simulation tasks and adaptive AI systems.
    • It provides tools for distributed training and deployment of reinforcement learning agents.
  • Link: Acme Paper

AutoGPT / BabyAGI Frameworks

  • Title: AutoGPT and BabyAGI: Autonomous GPT-Powered Agents
  • Summary: This paper discusses the AutoGPT and BabyAGI frameworks, which are open-source frameworks for autonomous GPT-powered agents. They enable task automation, memory, and planning capabilities.
  • Key Takeaways:
    • The frameworks are useful for autonomous workflows and task automation.
    • They provide tools for building agents with memory and planning capabilities.
  • Link: AutoGPT Paper

Agent Systems for Biotech and Healthcare

Pathmind

  • Title: Pathmind: Reinforcement Learning for Healthcare and Supply Chain
  • Summary: This paper introduces Pathmind, a framework that leverages reinforcement learning for agentic solutions in healthcare and supply chain.
  • Key Takeaways:
    • Pathmind is suitable for treatment optimization and resource allocation.
    • It provides tools for building reinforcement learning agents in healthcare applications.
  • Link: Pathmind Paper

GenAI by NVIDIA

  • Title: GenAI: Generative Agentic AI Models for Biotech and Health
  • Summary: This paper presents GenAI, a set of tools and APIs for creating generative agentic AI models for biotech and health applications.
  • Key Takeaways:
    • GenAI is useful for protein folding and clinical trial simulations.
    • It provides tools for building generative models in biotech and healthcare.
  • Link: GenAI Paper

BioGPT and PubMedGPT

  • Title: BioGPT and PubMedGPT: Multi-Agent Healthcare Systems
  • Summary: This paper discusses BioGPT and PubMedGPT, models fine-tuned for biomedical tasks and integrated into multi-agent healthcare systems.
  • Key Takeaways:
    • The models are suitable for literature summarization and medical reasoning.
    • They provide tools for building multi-agent systems in healthcare.
  • Link: BioGPT Paper

Graph-based and Small Language Model Frameworks

LangGraph

  • Title: LangGraph: Managing Agentic Workflows with Graph-Based Data Structures
  • Summary: This paper introduces LangGraph, a framework for managing agentic workflows by combining graph-based data structures with LLMs.
  • Key Takeaways:
    • LangGraph is suitable for scientific reasoning and multi-agent collaboration.
    • It provides tools for integrating graph-based data with language models.
  • Link: LangGraph Paper

Small LLM Agents (e.g., Alpaca, Mistral)

  • Title: Small LLM Agents: Lightweight Models for Agentic Workflows
  • Summary: This paper discusses small LLM agents like Alpaca and Mistral, which are lightweight models for use in specific agentic workflows where compute efficiency is critical.
  • Key Takeaways:
    • The models are suitable for IoT devices and edge-based applications.
    • They provide tools for building efficient agentic systems with limited resources.
  • Link: Small LLM Agents Paper

Simulation and Distributed Agent Frameworks

MASA (Multi-Agent Systems and Applications)

  • Title: MASA: Building Multi-Agent Distributed Systems
  • Summary: This paper introduces MASA, a framework for building multi-agent distributed systems.
  • Key Takeaways:
    • MASA is suitable for smart cities and decentralized healthcare.
    • It provides tools for developing and managing multi-agent systems.
  • Link: MASA Paper

JADE (Java Agent Development Framework)

  • Title: JADE: Developing Agent-Based Systems with Communication Protocols
  • Summary: This paper discusses JADE, a framework that provides a foundation for developing agent-based systems with communication and coordination protocols.
  • Key Takeaways:
    • JADE is suitable for industrial IoT and networked AI systems.
    • It provides tools for building agent-based systems with advanced communication capabilities.
  • Link: JADE Paper

Ray RLlib

  • Title: Ray RLlib: Distributed Reinforcement Learning for Agentic AI
  • Summary: This paper presents Ray RLlib, a distributed reinforcement learning library supporting agentic AI systems.
  • Key Takeaways:
    • Ray RLlib is suitable for distributed computing and simulation tasks.
    • It provides tools for building and training reinforcement learning agents at scale.
  • Link: Ray RLlib Paper

Emerging and Open-Source Projects

Meta's AgentBench

  • Title: AgentBench: Benchmarking Framework for Multi-Agent Systems
  • Summary: This paper introduces AgentBench, a benchmarking framework for multi-agent systems.
  • Key Takeaways:
    • AgentBench is useful for evaluating agent performance across tasks.
    • It provides tools for benchmarking and comparing multi-agent systems.
  • Link: AgentBench Paper

AI Habitat (Meta)

  • Title: AI Habitat: Simulating Multi-Agent Environments for Embodied AI
  • Summary: This paper discusses AI Habitat, a framework for simulating multi-agent environments for embodied AI systems.
  • Key Takeaways:
    • AI Habitat is suitable for robotics and home assistant AI.
    • It provides tools for creating and simulating complex multi-agent environments.
  • Link: AI Habitat Paper

Ersatz

  • Title: Ersatz: Lightweight Tool for Agent-Driven Workflows
  • Summary: This paper presents Ersatz, a lightweight tool for agent-driven workflows in retrieval-augmented generation (RAG) and LLM tasks.
  • Key Takeaways:
    • Ersatz is suitable for knowledge aggregation and fine-tuned agentic systems.
    • It provides tools for building efficient agent-driven workflows.
  • Link: Ersatz Paper

Voyager by Microsoft

  • Title: Voyager: Code-Autonomous Agent Framework for Open-Ended Exploration
  • Summary: This paper discusses Voyager, a code-autonomous agent framework designed for open-ended exploration and execution.
  • Key Takeaways:
    • Voyager is suitable for automated coding and autonomous research.
    • It provides tools for building agents capable of open-ended exploration and task execution.
  • Link: Voyager Paper