Yu He

@kfountou @semihcanturk_en @DobrikG @ChendiQian @sirolfg @StefanieJegelka @AspectStalence @chrsmrrs thank you so much for the pointers!!

1. Learning to Execute Graph Algorithms Exactly with Graph Neural Networks arxiv.org/abs/2601.23207 2. Learning to Add, Multiply, and Execute Algorithmic Instructions Exactly with Neural Networks arxiv.org/abs/2502.16763 Also, fyi, the notion of exact learning of algorithms (automata and other machines) is a very well known concept in the computational learning theory community. See the work of Angluin. There is also a ton of work in CLT on approximate learnability of algorithms using statistical queries and also the classical PAC learning framework.

📢 New position paper: “𝗡𝗲𝘂𝗿𝗮𝗹 𝗔𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺𝗶𝗰 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗥𝗲𝗾𝘂𝗶𝗿𝗲𝘀 𝗮 𝗖𝗹𝗲𝗮𝗿 𝗦𝗰𝗼𝗽𝗲, 𝗧𝗵𝗲𝗼𝗿𝗲𝘁𝗶𝗰𝗮𝗹 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀, 𝗮𝗻𝗱 𝗘𝗺𝗽𝗶𝗿𝗶𝗰𝗮𝗹 𝗥𝗶𝗴𝗼𝗿” When and where should we use Neural Algorithmic Reasoning (NAR)? NAR is an exciting emerging area at the intersection of 𝙘𝙡𝙖𝙨𝙨𝙞𝙘𝙖𝙡 𝙖𝙡𝙜𝙤𝙧𝙞𝙩𝙝𝙢 𝙙𝙚𝙨𝙞𝙜𝙣 and 𝙣𝙚𝙪𝙧𝙖𝙡 𝙘𝙤𝙢𝙥𝙪𝙩𝙖𝙩𝙞𝙤𝙣 — but its foundations are still underdeveloped. In this paper, we argue that the field needs: ✨ a sharper 𝙙𝙚𝙛𝙞𝙣𝙞𝙩𝙞𝙤𝙣 of what NAR is (and is not), including clearer distinctions from neighboring paradigms, 📐 stronger 𝙩𝙝𝙚𝙤𝙧𝙮 for expressivity and generalization, 🧪 more rigorous 𝙗𝙚𝙣𝙘𝙝𝙢𝙖𝙧𝙠𝙨 and connection to high-impact 𝙖𝙥𝙥𝙡𝙞𝙘𝙖𝙩𝙞𝙤𝙣𝙨.

Check out our paper here: dransyhe.github.io/assets/pdf/nar…. We’d love to hear the community’s thoughts! This is a joint work by Robert R. Nerem, Timo Stoll, @semihcanturk_en, @DobrikG, @ChendiQian, Solveig Wittig, @sirolfg, @StefanieJegelka, @Vitercik, and Yusu Wang; co-led by @AspectStalence and @chrsmrrs.

@chaitjo Congratulations! 🙌

@crwhite_ml Tagging @vitercik !

[5/n] This is a joint work with my amazing collaborators Yingxi Li, @crwhite_ml, and Ellen Vitercik. Paper: arxiv.org/abs/2505.24069 Code: github.com/dransyhe/DSR-B… Data: huggingface.co/collections/vi…

[0/n] Can LLMs 𝘢𝘤𝘵𝘶𝘢𝘭𝘭𝘺 reason about structure—order, hierarchy, connectivity, and how parts fit together?🧩 We introduce 𝗗𝗦𝗥-𝗕𝗲𝗻𝗰𝗵: a data-structure benchmark designed to test this 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝘁𝗼𝗼𝗹𝘀. Even SOTA LLMs still struggle in the hardest settings.

The thing about a more focused, in-person conference is the spontaneous research discussions and friendships that felt natural and intimate. Grateful to be part of this community and to everyone who made LoG 2025 happen. Phoenix is great and will always hold a place in my heart.

🚨 Keynote lineup for #LoG2025! 🚨 Thrilled to welcome these amazing pioneers! 🚀 - Aditi S. Krishnapriyan @ask1729 (UC Berkeley) - Haggai Maron @HaggaiMaron (Technion / NVIDIA) - Melanie Weber @mweber_PU (Harvard) - Michael Galkin @michael_galkin (Google)

@stjpozg @DobrikG congrats!

Time to pack our knapsacks 🎒 for Phoenix 🔥 — KNARsack has been accepted to LoG ’25! Thanks to Petar — and to @DobrikG and Marin Šilić 🙌

@stjpozg @DobrikG @PetarV_93 Interesting idea that applies NAR on the knapsack problem. Our ICML paper also tried to expand NAR to the NP-hard domain — using primal-dual approx algo that applies to a wide range of NP-hard problems. Might be relevant too! arxiv.org/abs/2505.24067

🚀 Excited to share that our preprint is out! Huge thanks to my amazing collaborators @DobrikG, Marin Šilić, and @PetarV_93 ! KNARsack – a neural algorithmic reasoner for the Knapsack problem. 🎒🧠

@PetarV_93 @EEMLcommunity Interesting! Excited to see NAR applied to more NP-hard problems

NAR on pseudo-polynomial algorithms! congrats Stjepan for your tireless efforts and finding tricks that make such a model tick. 🧙 also many thanks to @EEMLcommunity for connecting us and making this work possible -- as a direct byproduct of discussions at EEML'24 in Novi Sad 😊

[6/n] This is joint work with my advisor Ellen Vitercik. 📍 Don’t miss our poster: Tuesday 11:00–13:30 @ # E-3003 Come chat about NAR, primal-dual reasoning, and how neural networks can think like algorithms 🧠➡️⚙️

[5/n] We evaluate PDNAR and show: ✅ Strong generalization to larger & OOD instances across three NP-hard tasks (vertex cover, set cover, hitting set) ✅ Utility as algorithmically-informed on real-world tasks ✅ Performance gains as warm starts in commercial solvers 🚀🧩

Our ✨spotlight paper✨ "Primal-Dual Neural Algorithmic Reasoning" is coming to #ICML2025! We bring Neural Algorithmic Reasoning (NAR) to the NP-hard frontier 💥 🗓 Poster session: Tuesday 11:00–13:30 📍 East Exhibition Hall A-B, # E-3003 🔗 openreview.net/pdf?id=iBpkzB5… 🧵