About Me

I’m Shreyans Jain, a third-year undergraduate at IIT Gandhinagar pursuing a B.Tech in Electrical Engineering. My passion lies in cutting-edge research in machine learning, computer vision, and signal processing. I have been actively involved in innovative projects—from developing robust AI image classifiers and artifact detection frameworks to designing hardware-accelerated neural network models on FPGA.

My work has earned recognition such as a Silver Medal at the Inter IIT Tech Meet 13.0, and I continually strive to build solutions that merge advanced research with real-world application.

Connect with Me:

Email: shreyans.jain@iitgn.ac.in
LinkedIn: Shreyans Jain
GitHub: shreyansjain04

Put your address / P.O. box / other info right below your picture. You can also disable any of these elements by editing profile property of the YAML header of your _pages/about.md. Edit _bibliography/papers.bib and Jekyll will render your publications page automatically.

Link to your social media connections, too. This theme is set up to use Font Awesome icons and Academicons, like the ones below. Add your Facebook, Twitter, LinkedIn, Google Scholar, or just disable all of them.

latest posts

May 12, 2023	a post with custom blockquotes
Apr 25, 2023	a post with table of contents on a sidebar
Apr 25, 2023	a post with audios

selected publications

2024

NeurIPS Workshop

Variational Diffusion Unlearning: A Variational Inference Framework for Unlearning in Diffusion Models

S. Panda, M.S. Varun, Shreyans Jain, and 2 more authors

In NeurIPS Workshop on Safe and Responsible Generative AI, 2024

Abs PDF

We present a variational inference framework for unlearning specific concepts from diffusion models while minimizing interference with unrelated concepts. Our approach uses an adapter network and novel loss function to reduce the likelihood of targeted concepts.
ICVGIP

WavShadow: Wavelet Based Shadow Segmentation and Removal

Shreyans Jain, Viraj Vekaria, Karan Gandhi, and 1 more author

In 13th Indian Conference on Computer Vision, Graphics and Image Processing, 2024

Abs PDF

We introduce WavShadow, a novel approach for shadow segmentation and removal using wavelet transforms. Our method effectively identifies and eliminates shadows while preserving image details and texture information.
ISTU

Suppressing Streak Artifacts in Ultrasound Images during Therapy Guidance by a Hybrid U-Net and Enhanced Masked Autoencoder

Shreyans Jain, Mihir Agarwal, Ruchika Dhawan, and 2 more authors

In International Society for Therapeutic Ultrasound Conference, Apr 2024

Abs

We present a hybrid U-Net and enhanced masked autoencoder approach to suppress streak artifacts in ultrasound images during therapy guidance, improving visualization for therapeutic ultrasound applications.
IEEE SAUS

Suppressing Streak Artifacts Generated by the Interference of Imaging and Therapy Fields: Initial Findings Using a Hybrid U-Net and Diffusion Model

Mihir Agarwal, Shreyans Jain, Ruchika Dhawan, and 2 more authors

In IEEE South Asian Ultrasonics Symposium, Mar 2024

Abs PDF

We present a combined U-Net and diffusion model to remove streak artifacts in therapeutic ultrasound imaging. Using a denoising U-Net with skip connections and text-based prompts for enhanced inpainting, initial results show effective artifact removal, an improved Signal-to-Noise Ratio by 6.0 ± 3.4 dB, and an IOU score of 0.698. Rapid model convergence suggests potential for real-time clinical use. Future work will validate the model on diverse datasets and optimize performance.
ISTU

Reconstruction of Ultrasound Super-Resolution Images using a Hybrid Attention-based U-Net Architecture applied to sparse data

Ruchika Dhawan, Mihir Agarwal, Shreyans Jain, and 2 more authors

In International Society for Therapeutic Ultrasound Conference, Apr 2024

Abs

We developed a Hybrid Attention-Based U-Net architecture to enhance Ultrasound Localization Microscopy by accurately tracking microbubbles in noisy image sequences. Our approach reduces data requirements by 40 times and achieves a peak SNR enhancement of 14.73 dB in simulated data.