About Mert

I am an undergraduate student at Bogazici University, studying Computer Science and Industrial Engineering as a double major. In Fall 2021, I will start my PhD in Computer Science at Stanford University.

My ultimate research goal is enabling greater use of machine learning in human-critical applications, particularly healthcare. While there are many diverse problems to tackle, I am broadly interested in building models that are robust to distribution shifts and various adversaries, that can be deployed in data-lacking regimes, and that offer explanations to their decisions.

Most recently, I was a Visiting Researcher at MIT - Poggio Lab and Sinha Lab where I worked with Dr. Xavier Boix and Prof. Pawan Sinha. Before MIT, I was a Research Collaborator at Harvard/Mathis Lab working with Prof. Mackenzie Mathis and Prof. Alexander Mathis. Previously, I worked as a Research Intern at MPI/Bethge Lab and Mathis Lab in Summer 2019 on out of domain robustness in pose estimation, supervised by Prof. Matthias Bethge. In Fall 2019, I was a research Assistant at Bogazici University working with Prof. Mustafa Gokce Baydogan, working on interpretability and multiple instance learning.

In the industry, I tackled various problems ranging from Recommendation Systems, Energy Forecasting, Vehicle Routing, Sports Analytics. Past internships include: Data Scientist at Algopoly, Supply Chain Optimization Analyst at Getir, ML Engineer at Armut, Data Analyst at QNB Finansbank.


  • Trustworthy machine learning
  • Human-critical applications of artificial intelligence
  • Robustness and explainability


  • Double Major: BSc in Computer Engineering and BSc in Industrial Engineering GPA:3.97/4.00, 2016 - Present

    Bogazici University

Recent Publications

On the Poisonous Neural Activity Driving Adversarial Attacks

Theories explaining the susceptibility of CNNs to adversarial attacks often characterize the impact of the inputs, i.e. dataset …

A Mixed-Integer Programming Approach to Example-dependent Cost-sensitive Learning

This paper studies example-dependent cost-sensitive learning that brings about varying costs/returns based on the label- ing decisions. …

Pretraining boosts out-of-domain robustness for pose estimation

Neural networks are highly effective tools for pose estimation. However, as in other computer vision tasks, robustness to out-of-domain …

Learning Maximally Predictive Prototypes in Multiple Instance Learning

In this work, we propose a simple model that provides permutation invariant maximally predictive prototype generator from a given …

ImageNet performance correlates with pose estimation robustness and generalization on out-of-domain data

Neural networks are highly effective tools for pose estimation. However, robustness to outof-domain data remains a challenge, …



Visiting Researcher

Massachusetts Institute of Technology - Poggio Lab

Jul 2020 – Oct 2020 Cambridge, MA, USA
(Remotely) Working on robustness and invariance.

Research Intern

Bethge Lab & Mathis Lab

Jun 2019 – Sep 2019 Tubingen, Germany
Worked on robustness in pose estimation.

Research Assistant

Bogazici University

Mar 2019 – Jun 2019 Istanbul
Worked on multiple instance learning and time series classification problems.

Reserch Assistant

University of Toronto

Sep 2018 – Nov 2018 Toronto
Worked on path planning problems.

Exchange Student

University of Toronto

Sep 2018 – Jan 2019 Toronto
Exchange term at UofT Computer Science. Completed the semester with 4.0.

Data Scientist Intern


Aug 2018 – Sep 2018 Istanbul
Worked on time series classification.

Machine Learning Engineer Intern


Jun 2018 – Aug 2018 Istanbul
Worked on recommender systems.

Data Analyst

QNB Finansbank

Jan 2018 – Feb 2018 Istanbul
Worked on customer categorization and ad-hoc analysis in finance.


Bayesian Methods for Machine Learning, High Honors

See certificate