📣 Our spicy ICML 2025 position paper: “Graph Learning Will Lose Relevance Due To Poor Benchmarks”. Graph learning is less trendy in the ML world than it was in 2020-2022. We believe the problem is in poor benchmarks that hold the field back - and suggest ways to fix it!.🧵1/10

RT @mmbronstein: Apply for the AITHYRA-CeMM International PhD Program! . 15-20 fully funded PhD fellowships available in Vienna in AI/ML an….

RT @chrsmrrs: This will be great

RT @michael_galkin: 📢 New paper: Distributed computing 🤝 agents in AgentsNet! AgentsNet transforms classical distributed computing problems….

AgentsNet is one of the most fun projects for me in 2025, it opens so many more opportunities to leverage the advancements of graph learning in the LLM era 🙂.We’ll be presenting a poster at the MAS workshop at ICML on Friday 18th 9/9.

We also present a collection of traces obtained from different problem configurations and LLMs so you can actually look into the message passing and how our agents communicate with each other to solve the problem. 8/9

Communication costs are important in large agentic networks - there is a price / performance Pareto frontier which we’d expect to be moving to the top-left corner pretty quickly as more capable and cheaper models become available. 7/9

Besides, AgentsNet is the largest agentic benchmark in the literature - when most existing approaches deal with 2-5 agents, we evaluated setups of up to 100 agents, and the benchmark itself is infinitely scalable in size to catch up with new generations of LLMs. 6/9

We benchmark a range of frontier models (o4-mini, Claude 3.7, Gemini 2.5) across different network topologies and graph sizes. Even frontier LLMs struggle to coordinate as networks scale, revealing fundamental limitations in collective reasoning. 5/9

Agents must collaborate to solve tasks of different theoretical complexity such as:.🎨 Graph Coloring • 👑 Leader Election • 🔗 Matching • ✅ Consensus • 🔍 Vertex Cover.4/9

In AgentsNet, each node is an LLM and an independent agent. In synchronous rounds, agents send and receive natural language messages to and from their neighbors, with no global view and no central controller. 3/9

Right away, all the necessary links.👉 Try the interactive demo: 📄 Read the paper: 💾 Code & data: 2/9.

📢 New paper: Distributed computing 🤝 agents in AgentsNet! AgentsNet transforms classical distributed computing problems into a benchmark for evaluating how LLM agents can coordinate when organized in a network.Led by Florian Grötschla, @luis_pupuis, @jonshoff w/ @phanein .🧵1/9

RT @GoogleResearch: Graph neural networks are becoming increasingly common across a variety of real-world applications. Stop by the #ICML20….

RT @mayabechlerspei: So many ppl came to hear the expo on Graph Foundation Models at #ICML2025 by @michael_galkin @phanein . This makes m….

RT @PetarV_93: i will not be going to @icmlconf #icml2025 this year but my colleagues will be presenting four of our papers throughout the….

RT @mmbronstein: 6. Position: Graph Learning Will Lose Relevance Due To Poor Benchmarks. East Exhibition Hall A-B #E-604.Thu 17 Jul 11 a.m.….

RT @benfinkelshtein: At ICML 🇨🇦 presenting the spicy 🌶️.Position: Graph Learning Will Lose Relevance Due To Poor Benchmarks.📍 East Hall A-B….

RT @mirrokni: Check out two recent blog posts from our team: . 1) Graph Foundation Models, and how they help achieve 3-40x in precision: ht….

We'll be presenting MOTIF with @hxyscott on Wed 4:30pm - Xingyue prepared a great poster talk! If you want to chat with less useful people, I'll be there too 🌚.

RT @ymatias: New advancement on Graph Foundation Models (GFM) for relational data. Similar to leading foundation models, GFMs learn transfe….