mehmet f. demirel

demirel at cs dot wisc dot edu  |  resume  |  gscholar  |  github  |  twitter

Mehmet F. Demirel
I am a PhD student in computer science at University of Wisconsin—Madison. I am fortunate to be advised by Yingyu Liang. My general research interests include machine learning, representation learning, self-supervised learning. More specifically, I have been focusing on learning representations of graph-structured data, and therefore graph neural networks. In my spare time, I tutor students on Varsity Tutors.

Publications 📝

you are seeing [all] papers

Attentive Walk-Aggregating Graph Neural Networks
Mehmet F. Demirel, Shengchao Liu, Siddhant Garg, Zhenmei Shi, Yingyu Liang.
Transactions on Machine Learning Research (TMLR), 2022
[OpenReview] [arXiv] [code]

Graph Neural Networks for an Accurate and Interpretable Prediction of the Properties of Polycrystalline Materials
Minyi Dai, Mehmet F. Demirel, Yingyu Liang, Jia-Mian Hu.
NPJ Computational Materials 7: 103 (2021).
[NPJ CM] [arXiv] [code]

N-Gram Graph: Simple Unsupervised Representation for Graphs, with Applications to Molecules
Shengchao Liu, Mehmet Furkan Demirel, Yingyu Liang.
Neural Information Processing Systems (NeurIPS), 2019. (Spotlight)
[NeurIPS] [arXiv] [code]

Polycrystal Graph Neural Network
Minyi Dai, Mehmet F. Demirel, Xuanhan Liu, Yingyu Liang, Jia-Mian Hu.
Preprint, 2022

Analysis of Sparse Subspace Clustering: Experiments and Random Projection
Mehmet F. Demirel, Enrico Au-Yeung.
Preprint, 2022

News 🔥

09.22 - Started working at Meta as a software engineer intern!
08.22 - Our paper where we introduce and analyze a novel attentive walk-aggregating algorithm has been published at TMLR.
05.22 - Started working at Google as a software engineer intern!
07.21 - Our paper has been published in npj Computational Materials! ✅
05.21 - Got my Master's in computer science! 💪
01.21 - Passed my qualifying exam! 🙌
08.20 - I received a grant from the IFDS for Fall 2020!
05.20 - Among a number of great colleagues, I have received a teaching award from our department! 🎉
09.19 - Our paper has been accepted for spotlight at NeurIPS 2019! ✅ 🙌

Talks 🗣️

11.21 - Lab Meeting -Recent literature
10.21 - Lab Meeting -Recent literature
10.21 - Lab Meeting -Recent literature
10.21 - Lab Meeting -Recent literature
09.21 - Lab Meeting -Recent literature
07.21 - Lab Meeting -Recent literature
04.21 - Lab Meeting -Recent literature
01.21 - Qualifying Exam -Walk Aggregation Graph Networks
12.20 - IFDS Ideas Forum -N-Gram Graph: Simple Unsupervised Representation for Graphs, with Applications to Molecules

Academic Service️ ‍🔧

Reviewed in the following conferences/workshops:
2022 - AISTATS (Top 10% reviewer), ICML, NeurIPS, DGM4HSD (ICLR workshop), AI for Science (ICML workshop)
2021 - AI for Science (NeurIPS workshop)

More 🔍

Welcome Weekend Optimization: We built an optimization model to optimize welcome weekend schedules of graduate programs e.g. setting up professor/student meetings based on such constraints as professor availability, student availability, students' interest in professors, professors' interest in students and so on. We implemented the model in Julia using JuMP.
3D Reconstruction of Chest X-rays: In a class project, we have reconstructed a neural network model to create clinically-relevant 3D volumes (CT-scans) from single 2D X-ray inputs.
Polygrain Structure Dataset: A collection of polygrain structure graph data points. It is a well-maintained dataset that we have generated for a project where we employ graph neural network models to learn polygrain structure and its properties (in collaboration with Mesoscale Computational Modeling Research Group).
xView2 Competition: Matt Dutson, Shri Shruthi Shridhar, and I have implemented a solution for the xView2 competition. We used a U-Net architecture for pixel classification of post-disaster imagery. Our model performed well enough to spend several weeks on the xView2 leaderboard. The source code and report can be found here.

Miscellanea 😄

I am a SACM volunteer!
I watch, play, enjoy, and live soccer! In fact, one of my biggest passions is for Galatasaray!
I write poems (well, if you'd call them poems). As a matter of fact, I have been compiling my work for years and hope to print a book for myself and my loved ones as a reminiscence.
I play the Bağlama e.g. (1), (2).
Playing around with ML libraries and trying to implement useless-but-cool stuff is a hobby of mine.
(Handsomeness Detector) (Robot 1) (Robot 2) (Simple Pong)

Page layout is a fork of minimal (CC BY-SA 3.0)
A lot of thanks to Aws Albarghouthi for letting me steal his webpage.