Diyar Taskiran
I am an MSc student in computer science at the University of Zurich.
My major is AI and machine learning, and my minor is data science.
My master's thesis studies the limits of recurrent reasoning models. I test Tiny Recursive Models (TRMs) on N-Queens completion tasks. Using proxy measures, I estimate task complexity and identify the levels at which the models fail. Prof. Roger Wattenhofer supervises this work at the Distributed Computing Lab at ETH Zurich.
For a master's group project, I built a 3D latent diffusion pipeline that generates synthetic CT scans from MRI scans. I wrote a wrapper around NVIDIA's MAISI 3D VAE to process full scans. I used a custom DDIM model to translate MRI data to CT data in latent space. We developed the pipeline for the SynthRAD2025 challenge with University Hospital Zurich. Prof. Günther supervised the work at UZH.
For my bachelor's thesis, I trained large populations of vision models. Researchers can use these models to learn representations and predict model properties. The repository contains code and links to the datasets. I co-authored the Model Zoos paper, published in the NeurIPS 2022 Datasets and Benchmarks track. Prof. Borth and Konstantin Schürholt supervised the work at the University of St. Gallen.
In a previous life, I interned at Credit Suisse, where I contributed to strategic initiatives in fixed-income investment banking and private mortgages. I also interned at EY Zurich, where I worked on digital transformation projects for clients in wealth management, insurance, and reinsurance.