Alfonso Amayuelas

Self-Resource Allocation in Multi-Agent LLM Systems With the development of LLMs as agents, there is a growing interest in connecting multiple agents into multi-agent systems to solve tasks concurrently, focusing on their role in task assignment and co...

Check out the paper on arxiv if you're interested👇

Special appreciation to everyone who contributed to this project 🙏 Jingbo Yang, Saaket Agashe, Ashwin Nagarajan, Antonis Antoniades, @ericxw.bsky.social William Wang

arxiv.org/abs/2504.02051
🧵(3/3)

We evaluate LLMs as orchestrators and planners in MAS, exploring their resource allocation. The planner method achieves better efficiency and utilization for concurrent tasks. Explicit info on worker capabilities enhances strategies, especially with suboptimal workers.
🧵(2/3)

📜🚨 Check out our latest work on "Self-Resource Allocation in Multi-Agent LLM Systems" where we explore how LLMs can be used to optimize task allocation in multi-agent systems 🤖
🧵(1/3)

Today we are releasing Kaleidoscope 🎉

A comprehensive multimodal & multilingual benchmark for VLMs! It contains real questions from exams in different languages.

🌍 20,911 questions and 18 languages
📚 14 subjects (STEM → Humanities)
📸 55% multimodal questions

Check out the paper and website! 👇

📜Paper: arxiv.org/abs/2504.07072
🌐Website: dataset.ckodser.ir
📁Dataset: huggingface.co/datasets/Coh...

Check out the Kaleidoscope benchmark from @cohereforai.bsky.social ✨

A multi-modal multi-language benchmark of authentic exams covering

🗣️ 18 languages (Bengali → Spanish)
📚 14 subjects (Humanities → STEM)
📷 55% requiring image understanding!

📢BSC International Summer Internship Programme

👩‍💻 Are you an outstanding undergraduate student willing to join our research groups?

BSC offers 16 internship positions.👐Join us!

📆 Deadline to apply: 7 April

➡ bsc.es/Zyc

@upc.edu @fib.upc.edu

Alfonso Amayuelas, Joy Sain, Simerjot Kaur, Charese Smiley
Grounding LLM Reasoning with Knowledge Graphs
https://arxiv.org/abs/2502.13247

Grounding LLM Reasoning with Knowledge Graphs Knowledge Graphs (KGs) are valuable tools for representing relationships between entities in a structured format. Traditionally, these knowledge bases are queried to extract specific information. Howe...

This work shows the potential of combining structured knowledge from KGs with LLMs' generative capabilities, enhancing reasoning while adding relevant domain-specific knowledge and explainability to answers anchored to entities in the graph (4/4)

📜🔗 arxiv.org/abs/2502.13247

🔍ToT strategy excels by exploring multiple reasoning paths, leading to up to a 54.74% performance boost. Targeted interactions with KGs, like the agentic method, improve over time with more steps, while automatic graph exploration excels with fewer steps (3/4)

The study shows significant improvements using different reasoning strategies (CoT, ToT, GoT) on the GRBench dataset. We compare the results for an agent solution vs an automatic graph search at every thought step (2/4)

🚨Check out my work with the AI team at JPMorgan-Chase !!! 📜 "Grounding LLM Reasoning with Knowledge Graphs" explores the intersection of Knowledge Graphs and Reasoning strategies with LLMs which helps boosting answers in domain specific questions (1/4)

🧵👇

Hello! I’m now also on Bluesky ! 🦋