Jasmeet Singh’s Portfolio

Academic Qualification

Bachelor of Engineering in Electronics and Communication
Bennett University
Aggregate/CGPA: 7.93 August 2023

Work Experience

Research Intern at IIT KGP (KLIV)

Feb 2023 – Jun 2023

Implemented and analyzed computer vision techniques to reconstruct 3D scenes from 2D images and videos.
Conducted comprehensive research and experimentation on state-of-the-art techniques.
Synthesized findings into concise block diagrams and tables for presentation.

Deep Learning Research Intern at Bennett University, Greater Noida

Mar 2022 - Dec 2022

Utilized Python libraries for data visualization, generating interactive 3D pose points.
Conducted a study on Image Super Resolution models and evaluated PSNR, SSIM scores.

Academic Projects

Arduino Health Care System for COVID-19

Team size: 2
Developed a sensor-based system using Arduino to gather and analyze user health data.
Implemented classifiers to enhance system accuracy and effectiveness.

RGB to Hyperspectral Imaging

Team size: 3
GitHub Link
Developed an 8-Layer 2D Convolved autoencoder model for image transformation.
Assessed model performance using key metrics.

Domain Adaptation with CycleGAN

GitHub Link
Created a CycleGAN-based project for domain adaptation with adversarial training.

Pix2Pix Image Generation

GitHub Link
Implemented a Pix2Pix project for realistic image generation from sketches.

Neural Style Transfer

GitHub Link
Developed a project using VGG19 for blending styles with original content images.

Paper Submission

1. Track Name: Archives of Computational Methods in Engineering.

Paper Title: Deep learning based Single Image Super-Resolution: A comprehensive survey.

Description: This paper reviews deep learning-based algorithms for single image super-resolution (SISR), highlighting key components, learning strategies, datasets, and quality assessment metrics. Future insights are presented for improved performance with low computational cost.

Authors: Ankit Shukla (Primary), Avinash Upadhyay, Jasmeet Singh, Manoj Sharma.

2. Track Name: ICCV2023

Paper Title: Absolute 3D Pose Estimation from Multi-View Synchronized Videos Using Transformer-based 3D Pose Baseline for Temporal and Spatial Coherence.

Primary Subject: Human pose/shape estimation.

Authors: Avinash Upadhyay (Primary), Ankit Shukla, Udyan Sharma, Jasmeet Singh, Manoj Sharma.

Skills

Technical Skills

Languages: Python, SQL, MATLAB, C, C++, HTML

Libraries and Packages:

Machine Learning and Deep Learning: ScikitLearn, Pandas, Numpy, Pytorch, Keras, Tensorflow
Data Visualization: Matplotlib, Seaborn, Plotly, pythreejs
Image Processing: OpenCV, Pillow, ScikitImage
Others: Tkinter, BeautifulSoup, pymysql, Selenium

Tools: VS code, Jupyter Notebook, Tableau, MS Excel, Power BI, GitHub, PopSQL

Certifications

Medium Articles

1. Denoising Diffusion model Implementation from scratch

Article Summary: Denoising Diffusion Models are generative models that convert data into noise and learn to reverse this process. They use Markov chains to add and remove Gaussian noise progressively. This guide covers theory, training, and sampling, with results on the CIFAR-10 dataset using PyTorch. Read more

2. LLM fine-tuning methods

Article Summary: Explored the pre-training and fine-tuning processes for large language models (LLMs), emphasizing methods to mitigate catastrophic forgetting. Highlighted parameter-efficient fine-tuning techniques, including LoRA and prompt tuning, for improving model adaptability and efficiency. Demonstrated advancements in making LLMs versatile and resource-efficient. Read more

3. Evaluating LLMs

Article Summary: Explored evaluation techniques for large language models (LLMs) in NLP, focusing on metrics like ROUGE and BLEU. Reviewed benchmark datasets such as GLUE and Super GLUE for comprehensive performance assessment. Highlighted key insights into model capabilities and improvements. Read more