News
- (Sep 2024) Visiting Simons Institute for the semester as part of the "Modern Paradigms in Generalization" and "Special Year on Large Language Models and Transformers" long programs!
- (Aug 2024) Wrapped up my internship at Amazon.
- (May 2024) DiracDiffusion (Poster) and Adapt-and-Diffuse (Spotlight) got accepted to ICML 2024!
- (Jan 2024) Will be joining Amazon in the San Diego office as an Applied Science Intern for the summer of 2024!
- (Dec 2022) Obtained my M.Sc. degree in EE!
- (Apr 2022) Will be attending Princeton ML Theory Summer School organized by Boris Hanin this June. Excited to visit beautiful campus of Princeton University and IAS!
- (Dec 2021) Passed SIPI screening exam (ranked 1st in the department)!
- (May 2021) Will be attending CIFAR's Deep Learning + Reinforcement Learning (DLRL) and MLSS summer schools.
- (Apr 2020) Website is live! Received offers from UCLA, USC, and UBC. Very excited to join USC for my PhD studies next fall.
|
Research
My current research focuses on analyzing the convergence of shallow neural networks with small initialization, as well as developing algorithms for inverse problems, such as denoising, deblurring, and MRI reconstruction. I also have experience working with large language models (LLMs) from past projects, including knowledge injection via continual pretraining during my internship at Amazon and investigating their ability for self-feedback. Recently, I've become interested in the mechanistic interpretability of vision-language models (VLMs) and diffusion models. Selected papers are shown below.
|
|
Adapt and Diffuse: Sample-adaptive Reconstruction
via Latent Diffusion Models
Zalan Fabian*,
Berk Tinaz*,
Mahdi Soltanolkotabi
(* denote equal contribution)
ICML (Spotlight), 2024
NeurIPS Deep Inverse Workshop, 2023
GitHub
/
Paper Link
Latent diffusion based reconstruction of degraded images by estimating the severity of degradation and initiating the reverse diffusion sampling accordingly to achieve sample-adaptive inference times.
|
|
DiracDiffusion: Denoising and Incremental
Reconstruction with Assured Data-Consistency
Zalan Fabian,
Berk Tinaz,
Mahdi Soltanolkotabi,
ICML, 2024
GitHub
/
Paper Link
Novel framework for solving inverse problems that maintains consistency with
the original measurement throughout the reverse process and allows for great flexibility in trading
off perceptual quality for improved distortion metrics and sampling speedup via early-stopping.
|
|
HUMUS-Net: Hybrid Unrolled Multi-scale Network
Architecture for Accelerated MRI Reconstruction
Zalan Fabian,
Berk Tinaz,
Mahdi Soltanolkotabi,
NeurIPS, 2022
GitHub
/
Paper Link
A hybrid architecture that combines the implicit bias and efficiency of conbolutions with the power of Transformer blocks in an unrolled and multi-scale network to establish SOTA on fastMRI dataset.
|
|