Fangru Lin
185 posts
Fangru Lin
@FangruLin99
DPhil student @UniofOxford; Clarendon Scholar; Prev @GoogleDeepMind, @MSFTResearch, @Microsoft, @turinginst; Computational Linguist
Oxford, England Katılım Ocak 2024
527 Takip Edilen5.2K Takipçiler
Structure mapping is a super interesting question for me, and I really enjoyed this project❤️! Kudos to my amazing collaborators! @vjhofmann, @wanxingchen_, Weixing Wang , @ZifengDing6, Anthony Cohn, and Janet Pierrehumbert. Check the paper here: arxiv.org/abs/2602.03542 🧵 n/n
English
Last, we show that successful procedure generalization is closely related to structural similarity. This pattern can be interpreted as generative analogy/structure mapping. Although humans can often map structures with minimum exposure, LLMs require extended training. 🧵 4/n
English
New preprint! LLMs are trained and tested heavily on code and graphs, but can they generalize to natural language questions that share isomorphic procedures? We test by training Qwen-2.5, Llama-3, and Olmo-2 on isomorphic data in code, graph, and natural language. 🧵1/n
English
Thanks for featuring me! 🤩 It has been a great pleasure to work at @GoogleDeepMind!
Life at Google@lifeatgoogle
At NeurIPS, Fangru — a research PhD intern at Google DeepMind — gave us a behind-the-scenes look at her work on Gemini Deep Research. Explore Google's open AI/ML roles ➡️ goo.gle/4iyaDXQ
English
Fangru Lin retweetledi
The best poster awards go to:
1. Go-Browse: Training Web Agents with Structured Exploration
Apurva Gandhi, Graham Neubig
2. Scaling Open-Ended Reasoning to Predict the Future
Nikhil Chandak, Shashwat Goel, Ameya Prabhu, Moritz Hardt, Jonas Geiping
🎉Congrats!
English
Fangru Lin retweetledi
"Building environments is fun..."
-- Edward Grefenstette @egrefen
Do you agree? 🤗
SEA Workshop@SEAWorkshop
Invited Talk 1 “Open Problems and Interesting Directions for Open-Ended Agents” from Edward Grefenstette @egrefen (Google DeepMind)
English
Loong is being presented at @LAW2025_NeurIPS!
Guohao Li 🐫@guohao_li
Sir, we built this. A RL environment for learning reasoning at scale. GitHub: github.com/camel-ai/loong HF dataset: huggingface.co/datasets/camel… We extracted seed datasets from sources like textbooks, code libraries like sympy, networkX, Gurobi (math programming lib), rdkit (chemistry), prolog (logic) and so on. We gathered 8,729 questions spanning 12 diverse domains. Moreover, next data can be synthesized by generating question with few shot examples and questions by coding with given libraries. The programmatic approach brings strong verification signals to the synthetic data. The data is difficult enough for the SoTA LLMs. It is a good environment for learning long CoT reasoning by generating countless data points and let the agent practice in it.
English
Fangru Lin retweetledi
Is verifiable reward all we need? @egrefen talking about stepping outside of verifiable reward problems now at @SEAWorkshop! #NeurIPS2025
English
Happening now! Meet us at upper floor room 23! #neurips25
SEA Workshop@SEAWorkshop
Invited Talk 1 “Open Problems and Interesting Directions for Open-Ended Agents” from Edward Grefenstette @egrefen (Google DeepMind)
English
Fangru Lin retweetledi
We need more 𝗼𝗽𝗲𝗻, 𝗿𝗲𝗮𝗹𝗶𝘀𝘁𝗶𝗰 𝗮𝗴𝗲𝗻𝘁 𝗲𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁𝘀 for training and evaluating agents! 💡
🚀But what are the 𝗶𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝘁 𝗲𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁𝘀 𝘁𝗼 𝗯𝘂𝗶𝗹𝗱?
📈What are the 𝗶𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝗯𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸𝘀 for these environments in training and evaluation, and how can we 𝘀𝗰𝗮𝗹𝗲 𝘂𝗽 the number of available environments?
🌟Most importantly, how should we utilize these environments: 𝗥𝗟 𝗼𝗿 𝗯𝗲𝘆𝗼𝗻𝗱?
If you’re interested in discussing together, come join us at our workshop on “𝙎𝙘𝙖𝙡𝙞𝙣𝙜 𝙀𝙣𝙫𝙞𝙧𝙤𝙣𝙢𝙚𝙣𝙩𝙨 𝙛𝙤𝙧 𝘼𝙜𝙚𝙣𝙩𝙨” @NeurIPSConf tmr (7th Dec)! We have an amazing lineup of invited speakers and panelists, including 𝐄𝐝𝐰𝐚𝐫𝐝 𝐆𝐫𝐞𝐟𝐞𝐧𝐬𝐭𝐞𝐭𝐭𝐞 from 𝐆𝐨𝐨𝐠𝐥𝐞 𝐃𝐞𝐞𝐩𝐌𝐢𝐧𝐝 and 𝐒𝐡𝐮𝐲𝐚𝐧 𝐙𝐡𝐨𝐮 from 𝐃𝐮𝐤𝐞.
Also check out our latest 𝐬𝐮𝐫𝐯𝐞𝐲 𝐩𝐚𝐩𝐞𝐫 on the topic led by Yuchen Huang: arxiv.org/abs/2511.09586 🎯
English
Fangru Lin retweetledi
🚀 SEA Workshop is going LIVE TOMORROW!
Join us at NeurIPS 2025 for a full day diving into the Scaling Environments for Agents featuring an incredible lineup of speakers and panelists：
@egrefen @Mike_A_Merrill @mialon_gregoire @deepaknathani11 @jl_marino @syz0x1 @qhwang3 Anthony G. Cohn, Eric Sommerlade, @fredsala
📍 Upper Level Room 23ABC
🕘 08:00–17:00
Huge thanks to our sponsors:
@TheInclusionAI (@AntLingAGI) @SnorkelAI @SonicjobsApp and @VmaxAI 🙌
🔥 Get ready for a day of insights and inspiring conversations!
English
Happening now! Our paper presented by @RyanOthKearns, also covered by @guardian!
Come and say hi to us at poster 107!
Oxford Internet Institute@oiioxford
Tomorrow at #NeurIPS2025 in San Diego, here are three things from OII researchers you won't want to miss: 1) Measuring What Matters: Construct Validity in LLM Benchmarks at 11am – 2pm PST, Exhibit Hall C, D, E #107. Read more about the study here: oii.ox.ac.uk/news-events/st…
English
I will be at Google booth today at 1-2pm! Come and chat to us to learn more about Google DeepMind! #NeurIPS2025
English
I mainly work on planning and reasoning. I have rich experience in building auto-evals and optimising agent systems for complex tasks with first-author publications in top confs and internships in frontier labs. I’m also happy to explore other general LLM-related topics!
English
I will be at @NeurIPSConf during Dec 1-7. I’m looking for a full time Research Scientist/Engineer position so please DM me if you are hiring! Also happy to learn about internship opportunities. My CV and details in the thread 🧵
English