Ashkan

Hi, I’m Ashkan 👋

CS PhD @ Vanderbilt | ML Intern @ Instacart • LLMs, attention, & optimal transport

Transformers & LLMs Optimal Transport Applied Mathematics

I am a Ph.D. candidate in Computer Science at Vanderbilt University, where I work in the Machine Intelligence and Neural Technologies (MINT) Lab advised by Dr. Soheil Kolouri. My research focuses on designing new attention mechanisms and uncertainty estimation methods for large language models, often drawing on ideas from optimal transport and geometric deep learning. Before starting my Ph.D., I completed a B.Sc. in Biomedical Engineering at Amirkabir University of Technology (Tehran Polytechnic) in 2021. Outside of research, I enjoy playing Dota 2 as a carry and exploring the world of coffee.

News

Joining Instacart (Fall 2025)

Updated: Sep 7, 2025

I’m joining Instacart as a Machine Learning PhD Intern (Fall 2025) to work on LLMs and prompt optimization.

ICML 2025: ESPFormer accepted 🎉

Updated: Sep 7, 2025

Our paper “ESPFormer: Doubly Stochastic Attention with Expected Sliced Transport Plans” has been accepted to ICML 2025.

Shifu says hi

Updated: Sep 7, 2025

Chief Morale Officer 🐈‍⬛ checking in and purring about attention heads.

Research Focus

Attention Mechanisms

  • ESPFormer — Doubly-stochastic attention with Expected Sliced OT, achieving O(n log n) inference (ICML 2025).
  • LOTFormer — The first linear doubly-stochastic attention via low-rank OT projections (under review).
  • LUNA — Linear Universal Neural Attention with meta-learned kernels across tasks (under review).

Selected Papers

Teaching

Vanderbilt University

  • Geometric Deep Learning — Teaching Assistant, Spring 2024

Amirkabir University of Technology

  • Statistics and Probability — Teaching Assistant, Spring 2021

Oral Presentations

Geometric Neural Simulators for Modeling Liver Tissue Deformation

Presented at the 13th Annual Symposium on Surgery, Intervention, and Engineering, Vanderbilt University, Nashville, TN — December 2024

Academic Services

Reviewing

  • AAAI 2026
  • NeurIPS 2025
  • ICLR 2025

Mentorship