Zarif Ikram (জারিফ ইকরাম)

I'm an undergraduate senior at CSE, BUET and a visiting scholar at NUS Yong Loo Lin School of Medicine and NUS AI Institute . From 2022 to 2023, I have done research in collaboration with Augmented Design Lab under Jim Whitehead, publishing in top intelligent vehicles conferences such as IV 2023 and ITSC 2023.

My current research focuses on world models and learning representations for downstream tasks. My ongoing undergraduate thesis, under Dr. Mohammad Saifur Rahman , focuses on biomolecular design, exploring better ways to sample antibody sequences in a conditional setting. Besides, my work at CogAI4Sci Lab focues on hierarchical world modelling, causal task understanding, and task specific representation learning under the amazing Dianbo Liu and Ling Pan.

My research goal is to build generalist agents (!). My subgoal is to (i) construct dynamic hierarchies for time-continuous world models, (ii) integrate intrinsic motivation with planning feasibility for embodied agents, and (iii) discover more effective methods to enhance the time horizon for model predictive control (MPC).

As a sixth-year Indian classical musician trainee under Asit Dey, I belong to the Indore gharana. I also have a first-class specialization in Nazrul Sangeet from Chhayanaut Music School. As a multi-intrumental vocalist, I play harmonium, guitar, and piano, and I love organizing musical perfomances. When I am not doing research and music, I practice calisthenics, play the latest Zelda titles, and keep up with the newest meme trends.

Contact: zzzarif.ikram[ a t ] gmail.com
GitHub  /  Google Scholar  /  X  /  YouTube

Highlights / News

Research

I'm interested in robust and accurate world modelling with generative modeling and self-supervisted learning. My interest also includes discrete generative modeling techniques such as generative flow networks and the algorithmic improvement of it. Most of my applied research is about modeling the real world with discrete objects (i.e., bio-molecules, road objects).

Combining denoising score model and discriminator model for discretized Langevin MCMC, this method generates high quality biomolecular sequences that optimize a given objective.

Taking inpiration from animal brain’s ability to perform temporal prediction in the latent space, we propose a simple unsupervised augmentation to existing methods for better downstream sample efficiency of world models.

Incorporating MaskGIT as a dynamics prior to enhance sequence modeling in world models, we address the limitation of transformer-based world models in continuous action environments and the inadequacies of prior methods, like STORM, in capturing effective state representations, suggesting that Masked Generative Priors could be a powerful inductive bias for world models, supporting broader applicability across diverse RL tasks and environments.

A powerful method for the GFlowNets training using Evolutionary algorithms that enhances its robustness to long trajectory and sparse rewards

GFlowNets enable diverse generation of HD maps that are structurally valid.

We design and propose a grid-based simulator that is lightweight and can be used to model pedestrian behavior in a variety of scenarios.

Procedural content generation method to generate infinite complex roundabout HD maps that can be used in autonomous vehicle testing simulators.

Selected Talks/Lectures

I love to share my experiences with the audience and learn from them.

Academic Services

Projects

These are mostly my side projects, which I did for hackathon or university course -- but mostly for fun.

Learn tensor manipulation with PyTorch with our automatic puzzle generator, or practice problems made by friends!

Named Bidirection Autonomous roBot with Obstacle avoidance (BABO), it uses Hierarchical Q learning to learn how to crawl forward while avoiding obstacle.

Built on breadboard with 19 ICs, this MIPS can run MIPS code with a 4-bit ALU.

Supply chain management is important for businesses as it manages the flow of goods and materials from production to consumption. However, the current system has problems that can cause inefficiencies, delays, and increased costs. To fix these issues, we plan to use blockchain and IoT technology to automate the supply chain and create a more efficient, cost-effective, transparent, and secure system. Our solution combines the digital, virtual, and physical worlds with RFID, blockchain, and IoT technology as part of Industry 4.0.

Astha proposes a reliable and sustainable method to implement something that should be the norm: refugees’ right to own their identity. While one problem can have multiple solutions, this paper shows that this problem can be suitably solved using a fully decentralized blockchain. Refugees need to be protected, and the first step to do so is to protect their identity.

Built for CSE 216 (Database) final project, it is a website for attending events together with like-minded buddies.

In the busy schedule of today’s life, a number of people does not want to go to hospitals. This is why, we propose ReMeDi, an end-to-end system that handles routine check and doctor interaction. It uses AI powered chatbot to talk to patients about symptoms and suggests tests. On the other hand, doctors can monitor their patients on the dashboard and can quickly take actions if needed. We thank Tazneen Tamanna for her contribution to the early part of this project.

An app for Nazrul Sangeet musicians, containing over 3000 detailed information

A desktop application for buying and selling players with threading and networking using java.

Mentoring

I am in no way accomplished for mentoring, but I do love to share insights from my experiences in a way that helps young researchers with limited capacity due to geographical/economic situation. If you think I can help you in any way in your research, feel free to send an email about anything -- I reply fast.

Music

My music interest is shaped by music lessons from the age of five till today and mostly comprises of classical and comtemporary. I listen to everything.

Selected honors

Selected Television and Stage Appearances

this guy's website is kinda cool.