Yagiz Devre

PhD Student at Yale, Princeton CS '26
I'm an incoming PhD student in CS at Yale. I study theoretical machine learning, LLMs, and high-dimensional systems.
Check out my work
yagiz.devre [at] yale [dot] edu

I recently graduated from Princeton University with a B.S.E. in Computer Science, with minors in Mathematics and Statistics and Machine Learning. I graduated summa cum laude (highest honors) and was inducted into Sigma Xi. My work is mostly about understanding how modern machine learning systems learn, scale, and reason.

At Princeton, I worked with Prof. Elad Hazan on Spectral Transform Units and scaling laws for large language models. I also wrote my junior thesis with Prof. Robert Tarjan on approximate nearest neighbor search and hierarchical vector retrieval algorithms, which earned the Best Thesis Award. During my exchange at ETH Zürich, I worked with Prof. Florian Tramèr on knowledge cutoffs of language models.

Before focusing on ML, I worked on statistical physics with Prof. A. Nihat Berker. We studied phase transitions and Potts models, which led to a paper in Physical Review E. That background still shapes how I think about ML. I am interested in scaling behavior, entropy, and simple mathematical models that explain complicated systems.

This fall, I will start my PhD at Yale University, where I plan to work on theoretical machine learning, high-dimensional probability, and physics-inspired approaches to learning systems.

Outside research, I like reading, writing, and talking to people about films. I keep a running log of everything I watch on Letterboxd.

Experience

Research and engineering roles across academia and industry.

Sep 26 - present

PhD Student @ Yale University

Incoming PhD student in Computer Science. I plan to work on theoretical machine learning, AI, high-dimensional systems, and the mathematical foundations of learning.

Theoretical Machine LearningHigh-Dimensional ProbabilityAI TheoryScaling LawsStatistical Physics
Mar 24 - May 26

Student Researcher @ Princeton and Google DeepMind

Worked in Hazan Lab on Spectral Large Language Models and Spectral Transform Units. My work focused on long-context sequence modeling, scaling behavior, and the design of efficient sequence mixer architectures. I also contributed to Flash STU: Fast Spectral Transform Units, which was invited to the 64th IEEE Conference on Decision and Control.

PythonPyTorchJAXLLMsSpectral Methods+1
Jan 25 - Aug 25

Student Researcher @ ETH Zürich

Studied Computer Science and Mathematics at ETH Zürich as a full-scholarship exchange student. I took graduate-level courses in large language models, high-dimensional statistics, graph theory, computational semantics, and 3D vision. I also worked with Prof. Florian Tramèr in the Secure and Private AI Lab on knowledge cutoffs and memorization profiles of language models.

Large Language ModelsPrivacyMemorizationHigh-Dimensional StatisticsNLP
Aug 25 - present

Student Researcher @ New York University

Worked with Prof. Ravid Shwartz-Ziv on looped reasoning architectures, including LoopedGPT and TRM-style models with looped adapter blocks. I studied iterative reasoning on structured tasks such as CLRS-Text, Assembly, Chess, and A5.

Reasoning ModelsTransformersInformation BottleneckAlgorithmic Tasks
Jun 24 - Aug 24

Data Science and Machine Learning Intern @ Pinecone

Worked on production data and machine learning pipelines for workload classification, revenue forecasting, and performance optimization. I contributed across SQL, Python, and JavaScript, and worked closely with production systems.

Machine LearningData ScienceSQLPythonVector Databases
May 23 - Aug 23

Student Researcher @ Princeton Plasma Physics Laboratory and HMEI

Developed simulations for quantum-mechanical systems and machine-learned atomic interaction potentials. I also wrote a white paper on DFT-based machine-learned reactive force fields for water and carbon dioxide-water systems.

Scientific Machine LearningDFTPhysics SimulationPythonVASP