Yuhang He (Henry)

I am honored to introduce my new work, XShapeEnc: Training-free Spatially-Grounded Geometry Shape Encoding. This innovative approach is capable of encoding an arbitrary spatially-grounded 2D geometric shape into high-dimensional space without any training. Key features include: - Utilization of the classic Zernike basis for efficient encoding of shape geometry and shape pose, either jointly or separately. - An interpretable and invertible encoding process that is rich in high-frequency details. XShapeEnc offers wide applicability for various downstream tasks that require shape analysis. For more details: paper: arxiv.org/pdf/2604.07522… code: github.com/yuhanghe01/XSh…

Honored to see my work has been featured by this blog. Deep Neural Networks: From Trustworthy Explanations to Robust Autonomous Systems scipapermill.com/index.php/2026…

🚀 [ICLR 2026] Existing text-to-audio generation (TTA) methods mainly focus on semantic correctness, yet they perform very poorly on relation-aware TTA generation. For example, current models achieve <30% audio event presence accuracy and <10% relation accuracy. In our newly accepted ICLR 2026 paper, we introduce Aurelius, a framework that enables relation-aware TTA research at scale. Specifically, we introduce two meticulously curated corpora: 🗂 AudioEventSet — 110 audio events across 7 major classes. 🗂 AudioRelSet — 100 relations across 6 major relation types. Based on the two corpora and the proposed data creation strategy, we can create massive (nearly unlimited) pairs with both • high linguistic diversity. • high acoustic diversity. We release all resources to support the broader community in AI, acoustics, computer vision, and multimodal research. 📄 Paper: openreview.net/pdf?id=LAYCYiI… 🗂 Dataset: huggingface.co/datasets/yuhan… 💻 Code: github.com/yuhanghe01/Aur… 🌐 Project Page: yuhanghe01.github.io/Aurelius-Proj/ Huge thanks to Andrew Markham, He Liang, @_jainyash and @VibhavVineet at Microsoft Research and University of Oxford for their unwavering support. #ICLR2026 #Multimodality

📣📣 Introducing OS-Marathon⏱️ I am proud to share this interesting project I did during my internship @Microsoft @MSFTResearch. 🥳 This work is about benchmarking computer-use agents' ability on a type of long-horizon, repetitive tasks. ✅ We focus on desktop workflows that are long-horizon and also repetitive, e.g. filling the expense system given a set of various travelling receipts. ✅ We create data, build tasks and environments and categorise them into multiple difficulty levels to enable fine-grained evaluation. 📎Project Page: os-marathon.github.io 📄Paper: arxiv.org/abs/2601.20650 Joint work with Daphne Barretto, Yiye Chen, Nicholas Gydé, Yanan Jian, Yuhang He @HenryOxplore, Vibhav Vineet @VibhavVineet

ONE QUESTION Why doesn't this statement explain why this happened?

MSR Vancouver lab is looking for a Canada based Ph.D. intern that can start the internship asap (preferrably before Christmas). The internship is about LLM pre/post train. The candidate should hold a work permit already. DM me if anyone is interested.

Our lab at MSR is also looking for multimodal learning intern in both Canada and US. Please apply via this link: jobs.careers.microsoft.com/global/en/job/…

Microsoft Research Vancouver lab is hiring multiple interns in Canada working at AI-Driven System Design. The successful applicant is expected to start interning as early as possible. Welcome to share to whoever might be interested in. Apply through: jobs.careers.microsoft.com/global/en/job/…

Microsoft’s roundtable: Industrial-academic collaboration #WiML #ICML2025

Next stop arXiv cleaner github.com/google-researc…

You can just do (hard) things..

Existing text-to-audio generation models fail to model audio events relations. We fill in this gap with a new benchmark, evaluation protocol. Title: RiTTA: Modeling Event Relations in Text-to-Audio Generation. Project site: yuhanghe01.github.io/RiTTA-Proj/ Code: github.com/yuhanghe01/RiT…

@leppert Hi Greg, for the job, do you hire postdoc?

I'm hiring for a new AI lab at Harvard (currently in stealth). If you're interested in doing SoTA work on data and models within a high velocity org, I'd love to chat. DMs open. Open roles include: - AI / data engineers - Program Dir - Partnerships Lead - Research Community Lead

On my way to #ICML2024 , look forward to catch up with new friends. I am also in the job market, look for research job in multimodal audio-visual-x learning.

Honored to introduce our new work: SPEAR, receiver-to-receiver neural warping field to predict spatial acoustic effects for one position from another reference position, without requiring source position. Code: github.com/xu-shitong/spe…. Paper: arxiv.org/abs/2406.11006

Happy to share that I have started an internship at Microsoft applied science group in Munich Germany.

Happy to share the news that I have had one paper on spatial audio learning accepted by #ICML2024 !

Build what you need and use what you build. This is a core philosophy of my research. It shifts the focus away from publishing “papers” to what really matters — impact. This thread unpacks why I think this is a successful approach to science. 1/10 Or see: perceiving-systems.blog/en/post/build-…

@tfburns Thanks Tom, make sense to me. Lots of work to do on the author side then.

@HenryOxplore Typically this is because there were, say, only 2 reviews submitted, and ACs/PCs sadly cannot rely even on all emergency reviewers to respond in time. So, they recruit multiple emergency reviewers. Atop that, the original (very late) reviewers might also submit.

I was invited to be an emergency reviewer for #icml . I accepted and tried my best to give the review. After seeing the released reviews, I suddenly found up to six reviews are here for that paper. Am I not an emergency reviewer at all then?🤣