Fatwir Sheikh Mohammed Email & Phone Number

Seattle, Washington, United States

Engineer in the Representation Learning Team!

Seattle, Washington, United States

· Enhancing the performance of a generalizable Foundation Model using contrastive self-supervised learning for Optical Coherence Tomography (OCT) Images· Utilized Vector Quantized Generative Adversarial Networks (VQGANs) and Multimodal Quantized Variational Autoencoders (MQVAEs) for reconstruction; qualitatively, decent reconstruction performance was obtained in a lesser number of epochs· Incorporated stage two training for downstream tasks, such as endemic disease prediction, highlighting the critical importance of OCT data and its correlation with systemic diseases

Seattle, Washington, United States

· Made progress on the neural style transfer problem for stains using supervised and adversarial approaches. This could potentially lead to a 90% reduction in staining time and a 1000-fold decrease in staining costs. Further, this could also allow for virtual staining on the fly.· Experimented with normalization and chunking for large 3D datasets (∼250 GB), leading to a 10% improvement in training loss· Employed supervised learning with UNets, incorporating losses such as KLD, weighted MSE, and trained on 7500 images of size 512 × 512· Implemented Pix2Pix and CycleGAN, outperforming the supervised method by almost 15% on a validation set of 750 images· Investigated various guided diffusion-based models, including pre-trained and custom-built, for zero-shot stain transfer of tissue images· Developed a prototype for a content-based autocropping feature that aimed towards optimizing the data processing pipeline for large datasets. This led to a 40% reduction in processing time and saved almost 30 − 50% in storage space.· Contributed to development of the GUI for microscope software and enhanced the saving module for laser calibration and scan settings

Seattle, Washington, United States

· Worked on the self-checkout problem using synthetic Image Generation (2D/3D) to improve training of object detection models· Accelerated training by almost 20% using Fully-Fused MLPs to train a custom retail product dataset of around 140 images· Obtained a validation PSNR of 25.96 dB in training a Neural Radiance Field (NeRF), deviating by just 0.6 dB from original implementation· Effectively harnessed 93% of the GPU’s capacity and slashed the training time by almost 42%, marking notable improvement in efficiency

Seattle, Washington, United States

· Developed the largest public vision-language histopathology dataset, comprising approximately 1, 000, 000 image-text pairs· Fine-tuned a pre-trained CLIP model (QuiltNet) on zero-shot classification across 12 datasets, achieving an improvement of 10% in accuracy· With our LMM, QuiltNet, we obtained around a 5% improvement on 4 datasets in linear probing using {1, 10, 100}% of data and boostedcross-modal retrieval by an average of 36% in image-text and text-image retrieval on the Quilt-1M holdout dataset· This LLM and ASR-based multimodal data generation approach was bestowed with an oral publication in the datasets track of NeurIPS 2023

Seattle, Washington, United States

· Developed a system incorporating an IWR6843 chip and bioradio sensor pod for non-invasive physiological sensing· Mastered the hardware setup and data collection process, optimizing transmission parameters for enhanced data reliability

Spectrum Lab

• Worked on optimizing the performance of GANs as a precusor to which Wasserstein Generative Adversarial Networks (WGAN) were implemented for 1D and 2D Gaussian Target Distributions whilst analyzing snakes using gradient vector flow (GVF) fields· Solved the optimal discriminator PDE by bridging the gap between GVFs and GANs - finite differences and grid inversion in 2D· Accelerated convergence within 100 iterations (vs. State-Of-The-Art) on 2D Gaussian targets with active contours using WGANs· Using tiling, improved the efficiency for the SnakeGAN to learn the latent space distribution of SVHN and CelebA datasets by around 8%· This snake based GAN optimization was bestowed with an oral publication at the Adversarial Learning Track of ICASSP 2023Extra-Curricular :• Represented the electrical department and bagged the silver medal in the Inter-departmental table tennis competition (Spectrum '22)

Mangaluru, Karnataka, India

I am a part of the special interest group - CHARGE that caters to the field of electronics. There were various workshops and Knowledge Exchange Programs organized in order to teach juniors about interesting topics in the fields of analog and digital electronics along with embedded systems.

Peter L. Reichertz Institute For Medical Informatics

· Performed Blood Pressure estimation from photoplethysmogram (PPG) signals taken from the M IM IC − III waveform dataset· Devised a spectro-temporal Deep Neural Network and attained near perfect correlation (R ≈ 0.95) between BP with PPG signals· Boosted R value between BP and PPG by 0.15, capitalizing on frequency information of PPG and its derivatives using GRUs and spectrograms· Enhanced the performance of the DeepNet by 10% leveraging Python, MATLAB and Bash scripts for extensive data preprocessing usingclassical signal processing techniques (Elgendi Algorithm, Hampel and Frequency Filters)· Trained the network on 251 patient records (approximately 2 hours of time-data) using leave-one subject out cross-validation (LOSO-CV)

Centre For Computational Imaging

Video Magnification was applied to ultrasound images of the Common Carotid Artery to analyse subtle motion present. The novel technique of enhancing oscillations and estimating the displacements along with grid distortion and subsequent interpolation was carried out. Various image processing techniques were also utilised. This project was guided by Dr. Mahesh Panicker.

Spectrum Lab

In this internship, various mathematical models that predict the spread of infection of the COVID-19 were studied and implemented. The critical parameters of the model in the governing differential equations were changed to see how sensitive the model was to these parameters. Subsequently, the influence of certain factors such as social distancing and quarantining on the spread of the disease was estimated.

Master Of Science - Ms, Electrical Engineering, 3.88/4

Specialization: Machine Learning and Computer Vision Advised by Dr. Linda Shapiro

Bachelor Of Technology - Btech, Major In Electrical And Electronics Engineering With Minor In Computer Science And Engineering, 9.71/10 | Class Rank : 2/108

Activities and Societies: Member of the Table Tennis team, SPICMACAY NITK and ISTE NITK 𝗥𝗲𝗹𝗲𝘃𝗮𝗻𝘁 𝗖𝗼𝘂𝗿𝘀𝗲𝘀 Computer Vision, Control Theory.

Senior Secondary-Cbse, Pcm With Computer Science

Primary And Secondary- Cbse

Activities and Societies: Cyber Club, Music Club, Sports Club, House Captain