Ye Yuan

[5/5] If you’re running heavy workloads and want to save your energy for actual architecture innovation rather than infrastructure maintenance, definitely give it a look. #MachineLearning #LLMs #AIResearch #GenerativeAI #GPU #TechTools

[1/6] Been iterating on some offline BBO and generative methods lately and got tired of fighting cluster queues. Shifted some of my compute to HyperAI (hyper.ai) and it’s been a massive timesaver for fast validation. 🚀

[4/6] 🔗 Built-in Ecosystem: Datasets and recent SOTA papers are pre-integrated, drastically cutting down the time it takes to reproduce baseline results.

[3/6] 💸 Insane Pricing: Pay-as-you-go is perfect for concurrent ablations. H100s for $1.80/h and the new RTX 5090s for $0.35/h. No idle costs.

[2/6] A quick technical review of why it works for my workflow: ⚡ Zero Setup Friction: Out-of-the-box Jupyter environments. Skipped the entire "debugging CUDA versions" phase and went straight to coding.

Excited to announce that our paper “From Noise to Order: Learning to Rank via Denoising Diffusion” has been accepted to #ICTIR2026! 🎉📚 📄Paper: arxiv.org/pdf/2602.11453 💻 Code: github.com/sadjadeb/Diffu…

Thrilled to share that our latest work on Training Diffusion Language Models for Black-Box Optimization has been accepted as a SPOTLIGHT at #ICML2026! 🔦🎉

Excited to share our #ICML2026 paper: SPADE (Support-Proximity Augmented Diffusion Estimation for Offline Black-Box Optimization) We tackle the challenge of offline BBO by leveraging diffusion models to effectively bridge the gap between dataset support and optimization targets

Follow Mila researchers on day 3 of @iclr_conf 2026! Suivez les chercheur·euse·s de Mila au jour 3 de ICLR2026. mila.quebec/en/news/follow…

🎉 ACL 2026! Can't feed sensitive financial tables to LLMs? Meet FACTS: an agentic workflow that sees ONLY table schemas to generate offline SQL+Jinja templates. 100% zero data leakage, zero math hallucinations, and zero latency on reuse! 🔒📊 #DataPrivacy #AgenticWorkflows #RAG

📊 ACL 2026 Findings! Do LLMs have economic biases? We introduce MFMD-Scen, a comprehensive benchmark evaluating scenario-induced behavioral biases in multilingual financial misinformation detection. High-risk AI needs robust fairness! 💰⚖️ #AI #NLP #FinTech

🎉 ACL 2026! Still using basic relevance for RAG personalization? Meet PURPLE: a Contextual Bandit framework that dynamically reranks retrieval records based on their true contribution to the user profile. Boosts generation quality instantly! 📈 #RAG #ACL2026 #MachineLearning

🎉 Accepted to ACL 2026! Stop treating LLM personalization as a black box! We uncover "Preference Heads" via mechanistic interpretability and propose DPS, a training-free framework for highly controllable and transparent generation. 🧠✨ #ACL2026 #LLM #Interpretability

🚀 New survey out! Humans naturally align intentions in dialogue, but AI agents struggle. We review 18 agents communication protocols via a novel "comm-syntax-semantics" taxonomy, exposing hidden technical debts. Your go-to guide! 👇🔗 arxiv.org/abs/2604.02369 #Agent #LLM #AI

Excited to share that our paper "Embedding-Based Context-Aware Reranker (EBCAR)" is accepted to #ICLR2026! 🎉 We introduce a reranker that "sees" the full context across passages but runs at the speed of embedding models. 🚀

🚀 Excited to announce our paper "Offline Model-Based Optimization: Comprehensive Review" is accepted to @TmlrOrg with Survey Certification! We bridge the gap between surrogate modeling and modern generative AI (Diffusion, GFlowNets) for black-box optimization. 🧪🧬

neurips papers by affiliation since 2015 $goog #NeurIPS2025

Mila est à @NeurIPSConf ! Venez discuter avec Ye Yuan (@StevenYuan666), doctorant à McGill/Mila, qui sera au kiosque de Mila dès 13h pour présenter sa recherche sur l'application des modèles génératifs au problème de l'optimisation hors ligne des boîtes noires.

Mila a son kiosque à la conférence @NeurIPSConf ! Venez discuter avec nous si vous souhaitez rejoindre notre institut de recherche ou rencontrer nos chercheur·euse·s au kiosque #625. #neurips2025