- Collaborated to create a unique neural network to solve the fundamental failure scenario of mode collapse in GANs- Created a type of GAN that performs much faster than most other GAN models by using an innovative pre-training process - Introduced an innovative loss function that accurately detects and quantifies mode collapse, allowing stable GAN training in sparser and further complex datasets

Dementia Risk Detection with Deep Neural Networks; using a customized CNN, using transfer learning via AlexNet architecture; MRI image dataset from Kaggle.Two CNN architectures (a custom model and AlexNet) were used in the development of this project with the goal of predicting the risk of an individual who has mild cognitive impairment developing Alzheimer’s disease with higher precision. One of the CNNs is a custom structure that was created by us and the other CNN model used is AlexNet, originally created by Alex Krizhevsky in collaboration with Ilya Sutskever and Geoffrey Hinton.

Project: Designing Neural Network Based DecodersThe project was purposed to reproduce the results from a paper that applied Recurrent Neural Networks (RNNs) in Quantum Computing. - Obtained understanding of fitting and overfitting in linear regression, logistic and softmax regression, regularization and generalization power- Applied deep learning through: classification and pattern recognition, feedforward and backpropagation, CNNs and RNNs- Used convex optimization through 1st and 2nd order methods, stochastic gradient descent, average gradient methods, and natural gradient- Built generative models using: energy-based models, variation auto-encoders, and generative adversarial networks - Collaborated in a team-wide software project that require us to reproduce the results in a machine learning research article using an RNN to program a quantum computational simple decoder with the Qiskit library and used Google Cloud Platform to train large datasetPaper: Designing Neural Network Based Decoders for Surface CodesAuthors: Savvas Varsamopolous, Koen Bertels, Carmen G. AlmudeverDelft University of Technology, Netherlands

Supervisor: Professor Brian McNamaraProject Title: X-Ray Analysis of Galaxy Cluster Atmospheres- Utilized CIAO and XSPEC software to analyze galaxy cluster atmospheres from the Chandra Data Archive- Reprocessed observation IDs from the CDA and reprojected all observation IDs for one cluster onto the one with the longest exposure time before combining multiple observations- Found blank-sky background files and combined ones that are normalizing to the source image count rate in the 9.5-12 keV band- Created combined event image and combined background image in the energy range 0.5-7 keV for Abell 2029 and also created the background-subtracted image- Created an exposure map for all observations, made exposure-correction for the area lost on point sources, chip gaps, bad pixels and modified this to the corresponding spectra- Used XSPEC environment to obtain projected gas properties of the ICM, used an absorbed single temperature PHABS*APEC model and produced plots of gas properties against radii- Wrote a final term report and presented orally online in front of supervisors and peers at the end of the project; presenting the relationship between temperature, electron density, cooling time and gas mass for the galaxy cluster Abell 2029

Course: PHYS 375 Stars, Department of Physics and Astronomy- Solved for the Standard Main Sequence using the adaptive step Runge-Kutta-Fehlberg 45 method to show a sequence of stellar models and reproduce the Main Sequence as well as the properties of the stars- Presented HR diagrams, L-M plots, and R-M plots for the Main Sequence and used OPAL tables to accurately model the internal structures of stars- Effectively provided a way to construct the luminosity function where, depending on the location of the surface, the remaining quantities were obtained by interpolation and the surface boundary condition is satisfied for the specific central density- Modified gravity parameters and inferred limits on the modification length scale; using two different lambda constraints for the effect of gravitational forces on small and large scales- Generated a variety of main sequences as functions of gravity parameters and explored implications for the HR diagram and the L-M and M-R relations

- Collaborated with supervisors to improve databases and optimized digital communication system - Kept meticulous records of all correspondence between management and tenants- Oversaw monthly collections of over $340,000 in funds, maintained high payment rates and positive relationships with tenants- Toured property with prospective tenants and provided abundant information in regards to its key features

- Created a 6-month customized curriculum and delivered weekly 1-hour presentations to 21 elementary and high school students to introduce fundamental coding techniques using MIT Scratch 2 - Mentored 12 students aged 8-17 on underlying concepts and critical skills of high-level programming using Python

- Assisted debtors with setting up payment plans to recover overdue balances- Worked in call centre environment handling manually- and automatically-dialled outbound calls- Delivered exceptional customer service on all calls while maintaining calm and professional demeanour in challenging circumstances

- Reviewed and updated client correspondence files and database information to maintain accurate records- Retained interest and maximized receptive learning by educating students using hands-on instructional techniques- Supported student learning objectives through personalized and small group assistance