About Me

I am a PhD student at Chalmer University of Technology, advised by Fredrik Johansson and Devdatt Dubhashi. I have received Hon. BS from the University of Toronto in Nov 2018, where I majored in physics and minored in mathematics and statistics. I obtained my master's in statistics at Yildiz Technical University in Aug 2022. I have extensive machine learning experince from NLP and outlier detection to neural graphics.

Currently, I am in the Healthy AI Lab in Chalmers. My research is focused on non-linear independent component analysis, and causal representation learning. You can see my teaching materials and some of my chosen research projects here.

shape

Publications


  • Identifiable latent bandits: Combining observational data and exploration for personalized healthcare joint with Newton Mwai, Emil Carlsson, and Fredrik Johansson; preprint available on arxiv.
  • On a notion of outliers based on ratios of order statistics, joint with Oğuz Gürerk; preprint available on arxiv.
  • Order Statistics Based Training and Scoring Algorithms for Deep Outlier Detection; presented in ICAS 2022 conference, available upon request.
  • Adaptive Slot-filling for Turkish Natural Language Understanding; IEEE Xplore, 7th International Conference on Computer Science and Engineering (UBMK), 2022.

    • Teaching & Talks

    Teaching

    Chalmers University of Technology, Department of Computer Science and Engineering:

  • TIN 092 – Algorithms, (Winter 2024, Fall 2024, Winter 2025)

    • Yildiz Technical University, Department of Statistics:

  • IST 3121 - Regression Analysis I, Fall 2019
  • IST 5110 - Advanced Regression Analysis, Spring 2020

    • Selected Talks

  • Noise Contrastive Estimation and Applications to Representation Learning, DSAI Seminars at CSE, Chalmers — November 2024
  • Identifiable Latent Bandits, Vector Institute University of Toronto — October 2024
  • Time Contrastive Learning for ICA, Healthy AI group, Chalmers — October 2023
  • Outliers and Anomalies, Probability and Statistics Seminars, Boğaziçi University, Istanbul — Jan 2021
  • Concentration of Measure and Almost Spherical Sections, Student Seminars in Analysis, Boğaziçi University, Istanbul — Jan 2021
  • Adaptive Slot-filling for Turkish Natural Language Understanding, 7th International Conference on Computer Science and Engineering (UBMK) — September 2022
  • Order Statistics Based Training and Scoring Algorithms for Deep Outlier Detection, 8th ICAS Conference 2022 — May 2022

    • Research

    Downstream Guarantees for Contrastive Learning (Jan 2025–Present)

  • Studying linear downstream guarantees for contrastive learning from robust optimization perspective focusing on bounds for minmax type bounds for worst case downstream task.
  • Applications to well-studied contrastive learning applications such as non-linear ICA.

    • Identifiable Latent Bandits (Nov 2024–Present)

  • Making use of observational data for online-decision making by training an identifiable latent varaible model. Latent variable model allows for sample efficiency in decision time and makes exploaration possible for ml based approaches.
  • Applications to personalized healthcare study on semi-synthetic Alzheimers data.

    • Outliers, order statistics, and concentration inequalities (2021-2022)

  • Main goal of our project is to have a distribution invariant robust estimator for anomalies based on order statistics.
  • Studying the distribution and the asymptotic behaviour of the chosen statistic.
  • Developing a python library for heavy-tailed data, anomalous data generation and order statistic simulations.

    • Turkish Bert transformer models for a low data setting (2020)

  • Main goal of the project is to establish a fast training model for an interactive dialogue system.
  • We developed a model which used transfer learning based on pre-trained transformers and used a hyper-parameter optimization like training on Conditional Random Fields.
  • Using word-piece tokenization to help differentiate between word roots and suffixes agglutinative languages such as Turkish.

    • Teaching Materials and Personal