Graph machine learning researcher @Livermore_Lab. @UMich ‘20 PhD @WUSTL ‘15 AB/MS. Chess master and powerlifter.
Joined January 2016
- Tweets95
- Following118
- Followers185
- Likes749
Multilevel framework to improve accuracy and/or runtime of several network alignment algorithms. Great work during your time at @Livermore_Comp Jing! arxiv.org/pdf/2208.10682.pdf
A great combination of theoretical analysis, examination of intuition from other domains, empirical insights, and new benchmarks for graph contrastive learning—all in one paper. Outstanding work led by Puja!
New Paper Alert #NeurIPS2022: Self-supervised learning with discrete, structured data comes with some twists! In our paper, we explore how data-centric properties known to be crucial to success of visual CL can fail to hold for graph CL. 🧵Abs:arxiv.org/abs/2208.02810
Mark Heimann retweeted
2. We take a deep look into graph contrastive learning and find that data-centric properties known to be crucial to the success of visual CL can fail to hold for graph CL. Paper: arxiv.org/abs/2208.02810 @puja_computes @lastgoodcaesar @EkdeepL @danaikoutra
Mark Heimann retweeted
Our fourth tutorial is held by Mark Andrew Heimann, Junchen Jin and Danai Koutra (@danaikoutra). The tutorial is about
+++ Network Embedding for Role Discovery: Concepts, Tools, and Applications +++.
Link:
markheimann.github.io/tutori…
To appear in TKDD!
Structural role-based node embedding: extensive benchmarks, insights, and easy-to-use codebase. Led by undergrad alum Junchen Jin whose organization awes me 🙂 with Di Jin and @danaikoutra.
arxiv.org/abs/2101.05730
github.com/GemsLab/StrucEmbe…
Mark Heimann retweeted
Thanks @tangjiliang for sharing this interesting and thought-provoking work. In light of these new findings, we revisit the problem of heterophily for GNNs and discuss the reasons behind the seemingly different takeaways from different works. Check it out:
jiongzhu.net/revisiting-hete…
GNNs are widely believed to work well due to the homophily assumption and recent works design new architectures to overcome such heterophily-related limitations. Is Homophily a Necessity for Graph Neural Networks? the answer is in our new preprint: arxiv.org/abs/2106.06134
arxiv.org/abs/2102.13582 Use node proximities to learn embeddings modeling node proximity OR structural roles. Kind of like using 🍕dough to make 🍕OR cinnamon rolls. @JingZhu85095487 did an outstanding job on this project and will present it today at #SDM2021
Credit to @DMthechemist for inspiring me to start making my own pizza in grad school and for telling me about the cinnamon roll thing
Benchin’ is like attention because it’s all you need. Get ready to disambiguate forthcoming raps about “weight training”
Featuring the work of @jiong971 and @YujunYan4 -- you love to see it
This Post is from an account that no longer exists. Learn more
I think I found my new place in science
OMG this is so meta. I just found that in the validation set from the VATEX Video Captioning dataset there's a video of @umsi #NLProc researcher @david__jurgens WEIGHTLIFTING 405 LB (!!!) youtube.com/watch?v=cfBtVPEw…
You're contributing really from the core, David! 😂😂🤣🤣
I tried to go beyond two layers and what do you know, oversmoothing issues. It’s like I learned nothing from all this graph machine learning research.
#SDM2021 paper *not* titled “Want Your Network Alignment Algo To Be Its Best Self? Try This One Weird Trick!” (yes, a few lines of code can help find much better alignments, but Xiyuan Chen @vahedian_f63 @danaikoutra and I back it up with theory and exps) markheimann.github.io/papers…
Structural role-based node embedding: extensive benchmarks, insights, and easy-to-use codebase. Led by undergrad alum Junchen Jin whose organization awes me 🙂 with Di Jin and @danaikoutra.
arxiv.org/abs/2101.05730
github.com/GemsLab/StrucEmbe…
Really proud of all the amazing undergrads I've had the privilege of working with who are already producing top-notch scientific research. Y'all are much cooler than I was when I was your age (not that taking an intro watercolor class with @cghodgepodge wasn't cool 😎) #SDM2021
Jiong is a BOSS (as are other collaborators, like @YujunYan4 and as always @danaikoutra). Congrats to all involved, and please do check out the work!
Excited to share our #NeurIPS2020 paper, in which we look into theoretically and empirically supported designs for Graph Neural Networks in a world beyond homophily, where dissimilar nodes are more likely to connect. Poster on Tue. 9am PST / 12am EST! Paper, code & talk 📺 in 🧵
To my loss I don't personally know the U of M researchers who titled their #NeurIPS2020 paper "Tree! I am no Tree! I am a low dimensional Hyperbolic Embedding" but I pronounce them the winners of machine learning
The entire GEMS Lab knows there is literally no way I would have made it #beysterkitchencruising
