Jiayuan Rao

3rd-year PhD Candidate @sjtu1896
AI for Sports

[Jiayuan Rao @Houston_Rao]

MatchMate⚽️

An AI football companion product for live game watching: ask questions, check data, and get personalized commentary and interactions in real time.

Born from our AI sports research, now in beta!

Try here👉https://t.co/Vbjf77dDjI
#FootballAI #SportsTech #WorldCup2026

[Jiayuan Rao @Houston_Rao]

@DorothyDDU No enemy了

[Fabrizio Romano @FabrizioRomano]

🌎🇨🇻 OFFICIAL: Vozinha wins FIFA Man of the Match Award for Spain-Cape Verde. 👏🏼

[Jiayuan Rao @Houston_Rao]

Come and see our work with @WeidiXie !!! As FIFA World Cup is coming, our SoccerMaster meets the CVPR award candidate🎉

[Xleprime @haolinyang233]

⚽️ Introducing the first vision foundation model for this beautiful game.

🌟 CVPR 2026 Oral & Award Candidate (with @WeidiXie )

⏰ Oral: June 6, 14:00 (Four Seasons Ballroom)

📍 Poster: June 6, 16:45 (#11)

📑 SoccerMaster: A Vision Foundation Model for Soccer Understanding

[Jiayuan Rao @Houston_Rao]

Fig. 5 | forecasting events from generated futures
GenTac samples future trajectory rollouts, then maps them into tactical events.

So the output is not a single label, but a distribution over possible tactical outcomes grounded in different generated futures.

[Jiayuan Rao @Houston_Rao]

Football tactics are not with a single future, In GenTac, we treat open-play as a generative multi-agent process, unifying:

⚽ multi-player trajectory forecasting
📍 tactical event recognition

Project Page: https://t.co/VI30Mevtpt
Github: https://t.co/MAcMaYiHId
@WeidiXie

[Jiayuan Rao @Houston_Rao]

Fig. 4 | trajectories carry semantics
On TacBench-Event with 5 event types and 15 subtypes, GenTac can recover tactical semantics directly from multi-player trajectories.

Event grounding reaches:

71.2% top-1 for event type
53.7% top-1 for event subtype
with strong top-3/5 Acc.

[Jiayuan Rao @Houston_Rao]

Fig. 3 | beyond soccer
We also retrain GenTac on basketball, American football, and ice hockey. Across all three, opponent-conditioned forecasting consistently outperforms unconditioned rollout.

[Jiayuan Rao @Houston_Rao]

Fig. 2 | conditioned forecasting
GenTac is also about generating futures under different conditions: team /league style and objectives.
Team conditioning improves collective structure at 5s. Objective guidance also makes the rollout steerable toward offense or defense.

[Jiayuan Rao @Houston_Rao]

Welcome to join our Challenges at CVPR 2026!!! See all the details here and I'm hosting VQA challenges with $1000 prize for No.1 candidate sponsored by KNQ technology.

[SoccerNet @soccernet_org]

The SoccerNet 2026 Challenges are in full swing⚽️🔥
Our presentation video is now live : meet the task leaders and don't miss any details before the April 24, 2026 deadline.
▶️Watch it here: https://t.co/PLY5BcTdOy
Time to climb the leaderboard 🚀
@_CVsports #CVPR2026

[Zhanhui Zhou @asapzzhou]

(1/n) 🚨 BERTs that chat: turn any BERT into a chatbot with diffusion

hi @karpathy, we just trained a few BERTs to chat with diffusion — we are releasing all the model checkpoints, training curves, and recipes! Hopefully this spares you the side quest into training nanochat with diffusion for now 🙂. It’s both a hands-on tutorial for beginners and an example showing how to use our complete toolkit (dLLM) for deeper projects.

Code: https://t.co/Nv7d1t8Qin
Report: https://t.co/sGKgA1Cz0O
Checkpoints: https://t.co/iluTMnHkQO

Motivation: I couldn’t find a good “Hello World” example for training a minimally working yet useful diffusion language models, a class of bidirectional language models capable of parallel token generation in arbitrary order. So I tried finetuning BERTs to make it chat with discrete diffusion—and it turned out more fun than I expected.

TLDR: With a small amount of open-source instruction-following data, a standard BERT can gain conversational ability with diffusion. Specifically, a finetuned ModernBERT-large, with a similar number of parameters, performs close to Qwen1.5-0.5B.

[Jiayuan Rao @Houston_Rao]

“Multi-agent System for Comprehensive Soccer Understanding” done with @WeidiXie will be oral presented #ACMMM2025 in Dublin at GoldSmith3, RDCC from around 14:20 tommorrow (10.30).

Come and chat if you are interested!⚽️🏆Also, join our challenge on CVPR 2026 based on this work!

[Jiayuan Rao @Houston_Rao]

NICE Go!!!

[Wenyue Hua @HuaWenyue31539]

Hi all,

I am hosting a dinner party on 11.5 at EMNLP this year!

We've invited a bunch VCs and startup people, also fantastic panelist to talk about embodied AI and LLM agents.

All are welcome to attend!!

[Jiayuan Rao @Houston_Rao]

Join our challenge @CVPR 2026! ⚽️⚽️🏆🏆 For $1,000 Prize with Soccer VQA task!

[SoccerNet @soccernet_org]

🚀🌟We're excited to kick off the third official task of #SoccerNet Challenges 2026!

⚽️3⃣ VQA Visual Question Answering

Details of this task are available at https://t.co/rEUmLAQKZQ

[Jiayuan Rao @Houston_Rao]

iii) Ablation studies demonstrate that the system achieves better performance when provided with general task descriptions as context. More qualitative results of execution pipeline are shown as follows.

[Jiayuan Rao @Houston_Rao]

In ACM MM 2025, we introduced "Multi-Agent System for Comprehensive Soccer Understanding" with @WeidiXie
⚽️AI with external knowledge to analyze on/off-field dynamics! 🧐

📄: https://t.co/SWY6AIgMGe
🌐: https://t.co/upZTE3mjKB

See details below! #AI4Sports #MultiAgent #ACMMM25

[Jiayuan Rao @Houston_Rao]

ii) Also, we designed a multi-agent system, SoccerAgent, to decompose the the given multi-modal soccer-related questions with tool chains and then execute the tools step by step with the help of a distributed toolbox.

[Jiayuan Rao @Houston_Rao]

i) We proposed a multi-modal soccer understanding QA benchmark with multiple choices based on WikiPedia and other existing datasets. Some representative examples for each task are presented for reference.

[Jiayuan Rao @Houston_Rao]

Lift off 🚀🚀

[Houston Rockets @HoustonRockets]

Let’s work, @KDTrey5 🚀

[Jiayuan Rao @Houston_Rao]

Danke Thomas😭🫡2️⃣5️⃣

[FC Bayern München @FCBayern]

Eine Karriere wie keine andere – Wir verneigen uns vor dir, Thomas! #DankeThomas