Student Researcher at ELLIS Institute Tübingen
Master’s in Machine Learning, University of Tübingen ELIZA Scholar
Hello there! I am Hossein! I’m currently a student researcher with my supervisor and thesis professor, Bernhard Schölkopf at ELLIS Institute Tübingen. I’m also a master’s student in Machine Learning at the University of Tübingen and an ELIZA Scholar. Previously, I completed my bachelor’s degree in Computer Engineering at Sharif University of Technology. I’ve also had the good fortune of working with Florian Shkurti at the University of Toronto on off-policy evaluation in reinforcement learning and robotics.
I generally relish mathematically inspired fields. Because of my roots in physics, one of the things I most enjoy is combining beautiful mathematical concepts from physics and the perspective of a physicist with computer science. I enjoy both the theoretical and practical sides of computer science, as the enjoyment of seeing your algorithm work is like no other.
I have worked on reinforcement learning and am now particularly interested in world models: how to make them reliable, safe, and theoretically robust. I am also exploring causality and how causal reasoning can be used to build world models that better capture the effects of actions and generalize to new environments.
GUIDED DIFFUSION × POLICY EVALUATION
STITCH-OPE is a model-based generative framework that leverages denoising diffusion for long-horizon OPE in high-dimensional state and action spaces via trajectory stitching and combination of positive and negative guided diffusion.
SPATIAL STRUCTURE × CERTIFIED ROBUSTNESS
Exploiting the spatial structure of images through super-pixel partitions to improve certified adversarial robustness.
ROBUST STATISTICS × OOD DETECTION
Finding outliers in unlabeled, mixed data using median-based gradient statistics, with theoretical guarantees and empirical evaluation.
GENERATIVE MODELS × SPECTRAL DIVERSITY
A framework for steering generative models from imitation toward greater spectral diversity, using entropy-constrained target distributions.
Department of Empirical Inference · Supervisor: Bernhard Schölkopf
Exploring computational epistemology at the intersection of theoretical machine learning, causal inference, and the philosophical foundations of artificial intelligence.
Robot Vision and Learning Lab · Florian Shkurti
Off-policy evaluation and generative approaches to reinforcement learning. This work includes STITCH-OPE, a NeurIPS 2025 Spotlight paper.
Farzan Farnia July 2023 – December 2024
Certified adversarial robustness using image structure and super-pixel partitioning.
Master’s in Machine Learning
Supervisor and thesis professor: Bernhard Schölkopf
Master’s thesis at ELLIS Institute Tübingen.
Awarded the ELIZA Research-Oriented Master’s Scholarship.
B.Sc. Computer Engineering 2020 – 2025
Highest distinction · GPA: 19.7/20.0 · Ranked 2nd out of 192 students
Graduated Summa Cum Laude (with Highest Distinction) from B.Sc. in Computer Engineering
GPA: 19.7/20.0 — Ranked 2nd out of 192 students
In Iran, a competition consisting of three rounds is held annually, in which more than 10,000 participants are tested on their theoretical and practical knowledge of physics. Following the third and final round, only 40 participants are selected to attend the Physics Olympiad Summer School, where they have the opportunity to explore the subject in greater depth during the summer. Fortunately, I was selected as one of the 40 participants, and after several midterm and final exams, I was awarded a silver medal.
Ranked 32nd in the National University Entrance Exam on Mathematics (Konkour) among more than 150,000 students nationwide. (Top 0.02%)
Max Planck Institute
CMMRS 2024 is a Pre-Doctoral School on Computer Science organized by Cornell University, Max Planck Institute for Software Systems, and University of Maryland, College Park. I was among about 80 talented students from all over the world who were selected to participate in this school.
View Certificate ↗I enjoy exploring history, especially Japan’s Sengoku era and China’s Three Kingdoms. I like imagining the different paths history could have taken, and bringing those worlds into writing and drawing.
I also enjoy gardening and taking care of flowers. These are some Mexican peppers from my garden.
Graph-based image segmentation with superpixels, plus a compact neural radiance field implementation.
A distributed, fault-tolerant messaging system with Docker, inspired by Kafka.
Game logic, a graphical interface, and peer-to-peer multiplayer. Built as a course project.