🚀 We just released the final test split of #RepLiQA —our dataset for evaluating QA on truly unseen content!. 📚 Dataset: 📝 NeurIPS ’24: Big thanks to my amazing co-authors @ @ServiceNowRSRCH ! 🙌. #RAG #LLMs #NLP #QA.

RT @NewInML: New to ML research? Never published at ICML? Don't miss this!. Check out the New in ML workshop at ICML 2025 — no rejections,….

RT @joanrod_ai: Thanks @_akhaliq for sharing our work! Excited to present our next generation of SVG models, now using Reinforcement Learni….

RT @patricebechard: 🚀 New paper from our team at @ServiceNowRSRCH!⁣.⁣.💫𝐒𝐭𝐚𝐫𝐅𝐥𝐨𝐰: 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐧𝐠 𝐒𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐖𝐨𝐫𝐤𝐟𝐥𝐨𝐰 𝐎𝐮𝐭𝐩𝐮𝐭𝐬 𝐅𝐫𝐨𝐦 𝐒𝐤𝐞𝐭𝐜𝐡 𝐈𝐦𝐚𝐠𝐞𝐬⁣….

Our team has released the UI-Vision benchmark (accepted at #ICML2025) for testing GUI agent visual grounding and action prediction! 🚀🚀🚀. 🤗 Dataset: Special thanks to the students to lead this effort, @PShravannayak and @EdwardJian2 . @ServiceNowRSRCH.

A much needed async RL training implementation!.Check out 🚀PipelineRL🚀 by @DBahdanau, @alexpiche_ and Rafael Pardinas!. Code: @ServiceNowRSRCH.

RT @DBahdanau: I am excited to open-source PipelineRL - a scalable async RL implementation with in-flight weight updates. Why wait until yo….

📣📣📣 We just dropped Test Split 3️⃣ of RepLiQA — our Q&A dataset built to really test LLMs on unseen, made-up content. 🚀Great for RAG, context reasoning & in-context learning 🚀. #ServiceNowResearch

RT @joanrod_ai: @rizalrenaldi @mrjeremyblaze StarVector does a very decent job on this example! Try here: https://t….

🚀 Internship Alert! 🇨🇦. Looking for a student to work on UI Agents & visual grounding! 🎯. 🔹Develop vision-language models for UI navigation.🔹Improve visual grounding and UI understanding. Apply here: #Internship #UIAgents #ML.@ServiceNowRSRCH

RT @PShravannayak: 🚀 Super excited to announce UI-Vision: the largest and most diverse desktop GUI benchmark for evaluating agents in real-….

🖥️UI-Vision is a new GUI benchmark from our group at @ServiceNowRSRCH .Evaluate your UI agents on our large-scale real-world desktop GUI data!. 🌐

RT @joanrod_ai: I’m excited to announce that 💫StarVector has been accepted at CVPR 2025! Over a year in the making, StarVector opens a new….

Read the blog post here:

6️⃣ Huge thanks to my amazing and talented co-authors! 📷@Gupta30Shubham @ZichaoL13759 Tianyi Chen @CemSubakan @sivareddyg @vzantedesc . @mcgillu .@LavalUniversity.@Mila_Quebec.@ServiceNowRSRCH.

5️⃣ Why It Matters:.✅ Efficiency: Fast & memory-friendly retrieval.✅ Transparency: Easier to inspect results.✅ Flexibility: Coarse-to-fine retrieval balances cost vs. accuracy. 🔬 ReTreever outperforms hierarchical methods while matching dense retrieval accuracy.

4️⃣ Probabilistic Relaxation & Tree Propagation:.🔹 We make tree learning differentiable by softening routes—inputs can traverse multiple branches with probabilities. 🔹 Tree propagation maintains valid probability flow, using either simple multiplications or learned refinements.

3️⃣ Training the Tree:.We optimize routing end-to-end using contrastive learning. Query-context pairs are routed to maximize similarity, while unrelated pairs are spread across different tree branches. This leads to semantic grouping without explicit clustering.

2️⃣ The Tree Structure:.ReTreever organizes documents in a perfect binary tree. At each internal node, we learn a routing function that sends queries & documents to similar branches, preserving retrieval accuracy while enabling hierarchical representations.

1️⃣ Motivation:.Standard retrieval encodes documents into high-dimensional embeddings, which can be expensive & opaque. ReTreever improves efficiency & interpretability by providing visibility into how documents are organized, making retrieval more transparent and structured.