Kayhan Behdin
 |
Senior Software Engineer - Machine Learning, LinkedIn |
About me
I'm a senior software engineer (machine learning) at LinkedIn. My research at LinkedIn focuses on neural network efficiency through model compression and pruning. I received my Ph.D. from MIT in operations research in 2024 under the supervision of Prof. Rahul Mazumder. My research at MIT was focused on statistical learning with discrete structures, such as sparsity, low-rankness, fusion etc. During my studies, I explored a wide range of problems, including well-known problems from statistics methodology such as sparse PCA, to emerging problems such as neural network compression. I was an AI/ML Engineering Intern at LinkedIn from June 2022 through August 2022, and from May 2023 through August 2023.
Research
Broadly, I'm interested in studying statistical learning problems that possess discrete structures. I'm interested both in the methodology, as well as the real-world applications.
From a methodological perspective, I'm interested in:
Designing efficient algorithms, including exact and approximate discrete optimization methods,
Developing a theoretical understanding of statistical and computational trade-offs under discrete structures.
From a practical perspective, I'm interested in understanding how discrete structures benefit model interpretability, and computational efficiency. Examples include:
Efficient neural network training and inference
Interpretability for Biostatistics
Some of my papers that demonstrate my major research interests include:
K. Behdin, G. Loewinger, K. T. Kishida, G. Parmigiani and R. Mazumder "Multi-Task Learning for Sparsity Pattern Heterogeneity: Statistical and Computational Perspectives ", The Journal of Royal Statistical Society, Series B (minor revision). [arxiv], [CRAN], [R package]
K. Behdin, W. Chen and R. Mazumder "Sparse Gaussian Graphical Models with Discrete Optimization: Computational and Statistical Perspectives ", Operations Research (minor revision). [arxiv], [code]
K. Behdin and R. Mazumder "Sparse PCA: A New Scalable Estimator Based On Integer Programming", Annals of Statistics (to appear), 2025. [arxiv], [code]
K. Behdin et al. “Scaling Down, Serving Fast: Compressing and Deploying Efficient LLMs for Recommendation Systems”, EMNLP 2025 Industry Track (to appear). [code]
Google Scholar.
Full list of publications.
CV (Oct 2025).