Developed multimodal architecture for automated assessment of tear film infrared images to detect/segment out the eye gland area, provide ophthalmologist quality assessment score (Meiboscore) and remove specular refection.

Dataset of 1000 images released.

[Model: Encoder-Decoder Structure, Resnet50, GAN architecture with image inpainting].

Images taken by long distance object object is effected by turbulenec and come out distorted and blurry. We are working on image restoration from dynamic moving scene of different depth using self supervised deep learning approach.

This involve moving object separation from turbulence motion by trained optical flow network and self supervised way of image restoration.

Onsite Experiment: Setup onsite team experiment with several telescopes, weather stations, and scintillometers.

Data Analysis: Analyze data taken with telescope, drones, cameras, weather stations, and scintillometers.

ML Model: Design ML model for atmospheric turbulence estimation with focus, light and motion correction.

Spectral Analysis: Analyze absorption and emission spectroscopy data of viruses from saliva and buer solution

Covid-19 Classication: Simulate dataset from limited spectra; AI for Covid-19 classication from spectral signatures.

Funded by: Lightsense Technology

Developed an architecture to analyze images of custom contact lenses and predict blood glucose level with 85% accuracy

Measured with spectroscopy system and got better result from image analysis based methods.

I built abby's diffraction limited microscopy consisting of two camera and different lens elements with supervision from Muhammad Mohsin Qureshi and Professor Euiheon chung.

Assisted lab members building different optical system including Confocal microscopy(with integration of scanimage), Light sheet microscopy and two Photon Microscopy.

We have proposed a LSTM based deep learning approach where no need to employ any feature extraction and reduction framework. In this approach, raw ECG data have been employed to the LSTM model.

Two classes motor imagery ECG has been classified using LSTM based deep learning approach.

Utilize NLP to assess the tweets of influential personalities and how it influences public sentiment
Understand public perception and the influence of misinformation on public sentiment to

• Help Policymakers with data-driven decision making to check spread of misinformation.
• Inspire Media and influential personalities to be cognizant and share the right news.
• Wake-up General Public to be more conscious about their news consumption.

This project was focused on measuring different neural network model available today and find the best suited model for tabular format data.

12 data sheet was collected from kaggle for Evaluation

Compared SVM, Attention Net, ResNet, CNN, Attention+Resnet with different learning rate, optimizer, loss function, decay, dropout, cross validation to find the best neural network basnd on One hot encoding was used.