@hxyscott

Final Year DPhil in Computer Science, University of Oxford.

 Oxford
Joined October 2023
  • Tweets114
  • Following180
  • Followers367
22Media
Xingyue Huang retweeted
Zecheng Zhang
@zechengzhOct 20
We're releasing AgentVFS, a virtual filesystem interface for AI agents agentvfs.org/ AgentVFS unifies every service into one interface and replaces the need of calling MCP. Stored data becomes one virtual filesystem across services, described by VFS.md, and actions become virtual CLIs, so the agent no longer calls MCP tools at all. AI agents work on live data across every service through the terminal-like interface they know best. They use the Unix tools they learned in training (ls, grep, jq, pipes, loops) and the terminal workflows they practiced in post-training on agentic coding. We're also releasing VFS-Bench 1.0 agentvfs.org/benchmark/vfs-b…, the first systematic study and benchmark to compare a virtual filesystem with MCP on the same tasks. Compared with MCP, AgentVFS uses up to 88% fewer tokens and has about a 5% higher success rate on average.
44
30
11
327
66,965
One model, wildly different graphs and tasks, and some very intriguing generalization capabilities. Excited about the future of this! Fantastic work by @louistichelman with great contributions from the usual suspects @hxyscott and @jw9730!
Louis@louistichelmanOct 15
Can one foundation model reason across different graphs, features, relation types and prediction tasks? Introducing 𝗪𝗮𝗻𝗱𝗲𝗿🚶 Our new preprint explores 𝗴𝗿𝗮𝗽𝗵 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 through a shared prediction objective and a random-walk interface. 🧵 1/7
7
16
733
Xingyue Huang retweeted
Louis@louistichelmanOct 15
Can one foundation model reason across different graphs, features, relation types and prediction tasks? Introducing 𝗪𝗮𝗻𝗱𝗲𝗿🚶 Our new preprint explores 𝗴𝗿𝗮𝗽𝗵 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 through a shared prediction objective and a random-walk interface. 🧵 1/7
1
15
3
28
3,114
Frontier graph foundation models for various tasks and graph types!
Louis@louistichelmanOct 15
Can one foundation model reason across different graphs, features, relation types and prediction tasks? Introducing 𝗪𝗮𝗻𝗱𝗲𝗿🚶 Our new preprint explores 𝗴𝗿𝗮𝗽𝗵 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 through a shared prediction objective and a random-walk interface. 🧵 1/7
3
9
574
That’s a wrap for GFM @ ICML 2026! 🎉 Thank you to all speakers, panelists, presenters, reviewers, ACs, organizers, and everyone who joined us in Seoul. Congrats on our awards: 🏆 GraphPFN 🎖️ MedReason 🎖️ GraphBench Excited to see the GFM community keep growing! #ICML2026
1
2
1
19
2,459
Now at GFM @ ICML 2026, we have the oral talk: GraphBench: Next-generation Graph Learning Benchmarking This work tackles one of the most important questions for graph foundation models: how should we evaluate progress? 📍 Room 308 #ICML2026 #GraphML #GFM
1
3
375
Now at GFM @ ICML 2026, we have the oral talk: GraphPFN: A Prior-Data Fitted Graph Foundation Model This work brings the Prior-Data Fitted Network paradigm to graph learning, aiming to build graph models that can generalize across datasets. 📍 Room 308 #ICML2026 #GraphML #GFM
1
7
494
Xingyue Huang retweeted
Xiaoxin He@he_xiaoxinJul 9
Talk by Andy Huang @shenyangHuang on temporal graph foundation models 📍Room 308, COEX #ICML2026 #GFM
2
13
1,671
Room 308 happening now!
Xiaoxin He@he_xiaoxinJul 9
GFM @ ICML 2026 is happening now! 📍 Room 308, COEX Seoul Join us for invited talks, orals, posters, panel discussion, and awards on graph foundation models. 🔗 sites.google.com/view/gfmicm…… #ICML2026 #GraphML #GFM
2
152
Xingyue Huang retweeted
Xiaoxin He@he_xiaoxinJul 9
GFM @ ICML 2026 is happening now! 📍 Room 308, COEX Seoul Join us for invited talks, orals, posters, panel discussion, and awards on graph foundation models. 🔗 sites.google.com/view/gfmicm…… #ICML2026 #GraphML #GFM
2
1
16
1,984
Starting now at room 308!
Xingyue Huang@hxyscottJul 9
GFM @ ICML 2026 is tomorrow! 🚀 Join us for invited talks, orals, posters, panel discussion, and awards on graph foundation models. 📍 Room 308, COEX Seoul 🗓️ July 10 🔗 sites.google.com/view/gfmicm… #ICML2026 #GraphML #GFM

GFM@ICML 2026

Motivation Graph machine learning (GML) applies ML to structured data represented as nodes and edges, and underpins applications such as fraud detection, recommendation, drug discovery, weather...

sites.google.com
1
135
Xingyue Huang retweeted
Xingyue Huang@hxyscottJul 9
GFM @ ICML 2026 is tomorrow! 🚀 Join us for invited talks, orals, posters, panel discussion, and awards on graph foundation models. 📍 Room 308, COEX Seoul 🗓️ July 10 🔗 sites.google.com/view/gfmicm… #ICML2026 #GraphML #GFM

GFM@ICML 2026

Motivation Graph machine learning (GML) applies ML to structured data represented as nodes and edges, and underpins applications such as fraud detection, recommendation, drug discovery, weather...

sites.google.com
1
4
1
11
1,552
GFM @ ICML 2026 is tomorrow! 🚀 Join us for invited talks, orals, posters, panel discussion, and awards on graph foundation models. 📍 Room 308, COEX Seoul 🗓️ July 10 🔗 sites.google.com/view/gfmicm… #ICML2026 #GraphML #GFM

GFM@ICML 2026

Motivation Graph machine learning (GML) applies ML to structured data represented as nodes and edges, and underpins applications such as fraud detection, recommendation, drug discovery, weather...

sites.google.com
1
4
1
11
1,552
Xingyue Huang retweeted
Jinwoo Kim
@jw9730Jul 5
I’m attending ICML to present papers on data transformations for generalization and agentic data science. Let's talk if you’re interested in these topics, or in Flow Language Models! Inverting Data Transformations via Diffusion Sampling (arxiv.org/abs/2602.08267), co-led with @sekoumarkaba Tuesday morning session HALL A #2917 RelAgent: LLM Agents as Data Scientists for Relational Learning (arxiv.org/abs/2605.07840), led by @hxyscott GFM workshop on Friday What are the Right Symmetries for Formal Theorem Proving? (arxiv.org/abs/2605.22257), led by @kolejnyyyy AI4Math workshop on Saturday

Inverting Data Transformations via Diffusion Sampling

We study the problem of transformation inversion on general Lie groups: a datum is transformed by an unknown group element, and the goal is to recover an inverse transformation that maps it back...

arxiv.org
7
39
2,308
Heading to Seoul for #ICML2026 🇰🇷! Excited to be leading the organization of GFM @ ICML 2026: Workshop on Graph Foundation Models. Join us on July 10th in Room 308 for invited talks, orals, posters, and discussions on the future of graph learning. sites.google.com/view/gfmicm…

GFM@ICML 2026

Motivation Graph machine learning (GML) applies ML to structured data represented as nodes and edges, and underpins applications such as fraud detection, recommendation, drug discovery, weather...

sites.google.com
4
25
1,602
Are we outgrowing Softmax attention for long-context LMs? Excited to share our #ACL2026 Oral paper: Threshold Differential Attention for Sink-Free, Ultra-Sparse, and Non-Dispersive Language Modeling *work done during my internship at @Snap 🧵1/7
2
3
20
1,417
more replies
🧵6/7 We also implement a fused Triton kernel for Threshold Rectified Attention. At long contexts, the kernel becomes competitive with FlashAttention-2 and reaches up to 1.29× speedup at 32k tokens.
1
1
1
92
🧵7/7 Come find us at Oral Session D: Language Models 3, Mon. July 6, 09:00–10:30! Paper: aclanthology.org/2026.acl-lo… Code: github.com/snap-research/TDA… Huge thanks to my collaborators: Xueying Ding, Mingxuan Ju, Yozen Liu, Neil Shah, and Tong Zhao!
1
1
97
Load more