Our work on automating query planning for ML will appear at #nsdi #nsdi24! Interested in working on similar awesome projects? I'm recruiting PhD students Georgia Tech School of Computer Science! Apply by Dec 15 and mention my name!

ML serving systems face a fundamental tension between throughput & latency. Our new work aims to relieve this: arxiv.org/abs/2312.05385. Led by the awesome co-primary authors Yinwei Dai Rui Pan 潘瑞. Want to work on similar problems? Apply to GeorgiaTech by Dec 15!

Radhika Niranjan Mysore and I are chairing NSDI'25. Applications for the NSDI 2025 PC are online! docs.google.com/forms/d/e/1FAI… We are especially looking for folks from underrepresented groups or geographic regions, so if you know of someone, encourage them to apply. #NSDI25

Georgia Institute of Technology, School of Computer Science, GaTech - College of Computing, Atlanta, Gerogia ift.tt/8huSjBQ

Our work on avoiding GPU interference for achieving resource optimized ML inference has been accepted to appear at #osdi #osdi24! Joint work with awesome folks Sudipta Saha Shubha and HaiyingShen!

Excited to be attending USENIX Association NSDI and meeting people! Please come say hi if you're around! #nsdi #nsdi24

Super excited to be at USENIX Association OSDI! Please come say hi if you're around! #osdi #osdi24

In an hour, Sudipta Saha Shubha will present our work on resource optimized ML inference by avoiding GPU interference at USENIX Association #osdi #osdi24. If you're around, please attend the talk and ask him about our work!

Our work on relieving the fundamental tension in ML serving systems has been accepted to appear at SOSP 24! Congrats to the amazing co-primary authors Yinwei Dai and Rui Pan 潘瑞! #sosp #sosp24

Our work on improving inference throughput via per-input adaptation has been accepted to appear at SOSP. Thanks to all the reviewers for their insightful reviews, and special thanks to our champion! Excited to visit Austin in November. #sosp #sosp24

Our work on accelerating DNN inference on low-cost edges has been accepted to appear at Eurosys 25! Joint work with TanmoySen and HaiyingShen! ACM EuroSys 2025 #eurosys

Our work on accelerating distributed GNN training leveraging a communication-efficient partition refinement scheme and a better caching strategy has been accepted at Association for Computing Machinery #CoNEXT 2024!

I'm really looking forward to meeting people at Association for Computing Machinery #SOSP 2024. If you're around, come say hi!

MoEs promise reduced parameter activation, but in practice invoke all experts during batched inference. With my awesome students Vima Gupta, Kartik Sinha & collaborator Ada Gavrilovska, we show how MoEs can fulfill their promise in a workload-agnostic way arxiv.org/abs/2411.08982

It's PhD application season again. I'll be on the look out for motivated students. Located in the vibrant city of Atlanta, Georgia Tech Georgia Tech School of Computer Science is a top 5 school with an awesome set of people. Apply by Dec 15 and mention my name! scs.gatech.edu/phd-programs

If you're on Bluesky, follow me anand-iyer.com

We are fast approaching the PhD application deadline this cycle (Dec 15 for Georgia Tech Georgia Tech School of Computer Science, we're hiring so consider us if you're interested in systems!), good luck to all applicants!! #AcademicTwitter

Our recent Association for Computing Machinery SOSP papers on accelerating LLM inference are featured on Georgia Tech Computing news: cc.gatech.edu/news/new-resea…! Interested to work on similar systems? Georgia Tech School of Computer Science PhD application deadline is tomorrow (Dec 15): scs.gatech.edu/phd-programs #academia #AcademicTwitter

Serving GNNs on large graphs is challenging due to the high communication, computation, and memory overheads. Our new work, Omega, shows how to solve this. Joint work with awesome collaborators Daehyeok Kim, Aditya Akella and amazing students! arxiv.org/abs/2501.08547

Dynamic Graph Neural Networks (DGNNs) can be really useful in many domains, but they are hard to scale. Our new work shows how to achieve an *order of magnitude* speedup when training DGNNs! Joint work with awesome folks Mingyu Guan and Taesoo Kim! arxiv.org/abs/2501.15348