Danila Mishin

Current Role: Software Engineer @ IMC Trading
Location: Switzerland
Areas of Interest: Low-Latency Systems, Scalable Distributed Systems

Danila Mishin currently works as a software engineer at IMC Trading.

He holds a Master’s in Computer Science from EPFL, where he was awarded a Research Scholarship to eventually be a part of collaborative projects on efficient LLM inference and training with Google, AMD, and Korea University at the PARSA lab under the supervision of Prof. Babak Falsafi. He previously optimized large-scale core infrastructure at major tech firms including Oracle, Meta and Yandex.

Rethinking Scale Selection in Microscaling Formats for LLM Inference
Master's Thesis
link

• Mathematically formulated the shared scale selection problem for micro-scaling formats, deriving an optimal algorithm leveraging second-order information.

Effective Interplay between Sparsity and Quantization: From Theory to Practice
Spotlight at ICLR 2025
Simla Burcu Harma, Ayan Chakraborty, Elizaveta Kostenok, Danila Mishin, et al.
link

• Formulated a unified mathematical framework to analyze the joint impact of sparsity and quantization, demonstrating their non-orthogonal nature in model compression.

MXSens: Sensitivity-Aware Mixed-Precision Quantization for Efficient LLM Inference
Pre-print
Simla Burcu Harma, Danila Mishin, et al.
link

• Designed a novel mixed-precision quantization framework that leverages layer-wise sensitivity analysis for optimal bitwidth allocation, establishing a new Pareto frontier for model accuracy vs. average bitwidth.

Languages: C++, Python, C, Rust
Tools: Git, Docker, Linux

M.S. Computer Science (2023-2026)
École Polytechnique Fédérale de Lausanne (EPFL)

B.S. Applied Mathematics and Computer Science (2019-2023)
Moscow Institute of Physics and Technology (MIPT)

• Email: dkmishin at gmail dot com
• LinkedIn: linkedin.com/in/danmishin