JIS

Jibran Shah

Masters StudentUniversity of Toronto

I am a Masters student in ECE at the University of Toronto, with an undergraduate background in Mathematics. I am currently a deep learning researcher at Block, working on speech-to-speech models. Previously, I worked under Florian Shkurti at the RVL Lab on uncertainty calibration in perception-based motion planners and under Asif Zaman at the Fields Institute on probability theory.

Jibran Iqbal Shah

01

Research

02

Projects

TargetTTS

Target speaker extraction based ASR for full overlap speaker mixtures.

Reinforcement Learning finetunes Sparse Subnetworks in LLMs

Re-implementing the paper https://arxiv.org/abs/2505.11711 which says RL based finetuning changes only a sparse subnetwork in an LLM

Off-policy vs On-policy RL for Driving

We compare DQNs, DDQNs and PPO for self-driving on highway environments

HiggsArt, Voice-controlled Photoshop with GenAI

HiggsArt is a voice-controlled photo-editing platform that parses user intent to choose between native image edits or genAI edits after masking the object the user intends to edit.

03

Blogs

Why Laplace Transform works, a Linear Algebra approach

A proof of why the Laplace transform helps solve ODEs using Linear Algebra after viewing differentiation as a linear map.

Functional Analytic Proof of the Prime Number Theorem

Writeup of all the details of a functional analytic proof of the Prime Number Theorem sketched out in Terry Tao's blog.

> current cv

Curriculum Vitae