Zarif Ikram / জারিফ ইকরাম [scholar, github, x, youtube, zarifikram003 [at] gmail [dot] com]

hi, i am zarif. i am a first-year Annenberg PhD fellow at USC, with Paria Rashidinejad. i hold a B.Sc. in CSE from BUET (class of '25).

right now i am also a research intern at Microsoft Research AI Frontiers, NYC, working with John Langford, Tim Pearce, Siddhartha Sen, and Riashat Islam.

i like fundamental AI problems in world modeling and continual improvement. i recently took a liking to heaps of data and navigating through them.

Zarif Ikram with a cat
me + a cat

previously, i studied world models under Dianbo Liu at NUS and behavior modeling under Jim Whitehead at UCSC.

outside research, i do music!!


Papers:

  1. World Model Anomaly Detection with a Latent Linear Prior — TMLR'26
    Zarif Ikram, Harry Zhao, Ling Pan, Alex Lamb, Dianbo Liu
  2. JEDI: Joint Embedding Diffusion World Model for Model-Based Reinforcement Learning — preprint'26
    Jing Yu Lim, Rushi Shah, Zarif Ikram, Samson Yu, Haozhe Ma, Tze-Yun Leong, Dianbo Liu
  3. CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing — ICML'26
    Zarif Ikram, Arad Firouzkouhi, Stephen Tu, Mahdi Soltanolkotabi, Paria Rashidinejad
  4. Evolution Guided Generative Flow Networks — TMLR'25 — best poster award at EEML!!
    Zarif Ikram, Ling Pan, Dianbo Liu
  5. Gradient-guided discrete walk-jump sampling for biological sequence generation — TMLR'24 — RISE'24 grant!
    Zarif Ikram, Dianbo Liu, M Saifur Rahman
  6. PedGrid - A Simple yet Expressive Simulation Environment for Pedestrian Behavior Modeling — IEEE ITSC'23
    Golam Md Muktadir, Taorui Huang, Zarif Ikram, Abdul Jawad, Jim Whitehead
  7. Procedural Generation of Complex Roundabouts for Autonomous Vehicle Testing — IEEE IV'23
    Zarif Ikram, Golam Md Muktadir, Jim Whitehead

some unpublished/workshop papers...


Some talks / lectures:

  1. Masked Generative Priors Improve World Models Sequence Modelling Capabilities, ICLR World models workshop Oral (2025) / video / slides
  2. How to Train Your World Models, Beijing Institute of Technology (2024) / slides

i review NeurIPS, ICLR, ICML, and TMLR since 2024.



anonymous feedback

last compiled 2026-08-08.