Mahdi Saeedi's Portfolio
Hey! Welcome to my corner of the internet. I like to research machine learning, create easy to use analytic dashboards, and conduct simulations on human interactions. This page showcases a sample of my work and interests. Feel free to reach out if you have any questions or just want to chat.
This study explores using the U-Net model for Diabetic Foot Ulcers (DFUs) image segmentation within a federated learning framework, contrasting it with traditional centralized training. Leveraging an open-source Medetec dataset for privacy-sensitive medical images, the research demonstrates that federated learning matches centralized methods in performance, achieving a dice score of 0.9 without the need for centralized data aggregation. This highlights federated learning's potential in medical image analysis while addressing privacy concerns. The research further contributes by providing a federated learning codebase, facilitating further exploration and development in the field.
Distrust with the electoral system in the United States has grown over the years. In this project, we'll explore some of the mathematics behind redistricting in order to find a solution for gerrymandering. I have prepared a more extensive discussion on the matter; please access it using the link below. I have also designed an online partisan gerrymandering calculator that lets us simulate an instance of state redistricting at various levels of resolution for party preference spread among the population. Please find a demo and link to the app below.
This study analyzed the gene expression profiles of prostate cancer and benign prostate hyperplasia (BPH) patients using RNA sequencing. It identified differentially expressed genes and analyzed their functions, suggesting potential diagnostic, prognostic, and treatment targets for both conditions.
To identify jammed signals, we need to ask a series of questions about the signal data. Decision tree algorithms are great tools for this type of classification. In my project, I used Python's scikit-learn library's decision tree package to design a model that classifies healthy and jammed signals. This was part of more extensive work studying machine learning capabilities in detecting jamming attacks, which you can read about here. I've included the code and additional notes on the decision tree algorithm below. This work also explores jamming, particularly the impact of high-powered jammers on ADS-B signals, using MATLAB to simulate these effects. The dataset features 10 columns with signal information, such as energy and bad packet ratio. This research contributes to understanding how machine learning can detect jamming attacks. The simulation source code is available for use and further development into more complex jamming scenarios. For more details, my project notes are attached.
Given the rise of retail investors, I've been wondering if there are more effective ways of analyzing company data and making sense of the prospect of investment. My goal with Angur is to create a one-stop shop where users can analyze company data from financial statements, stock market data, and relative news. Angur is a python web application that allows users to access financial data and news data about a company traded on NYSE. Predict the next day's closing price of a particular stock using an LSTM neural network trained on trading data available on 30 previous days. I am extending this work to include different comparisons of the data points available, due to that, I have shut down the server. If you would like to test this application, please reach out to me.
I created an engine that translates and reduces pascal code to executable MIPS Assembly code in this project. Pascal code can contain basic arithmetic, functions, variable assignment, etc.
I designed a bot that plays a tic tac toe game with the user. At its heart, the algorithm utilizes a series of decision tree nodes at every move and prioritizes blocking possible winning moves of the human player.
Mock user interface for an air traffic control unit. Shows airplane entering and exiting an air space and associated flight data monitors plane movement for potential hazards. It warns the user in case of a possible crash, as shown in the image on the right. It also allows for modifying the background map.
This study employs the ns-3 library to simulate drone networks and ground station communication, extracting signal data like signal-to-noise ratio and drone specifics. The ns-3's object factory feature enables flexible creation of various entities, including UAVs. The linked script is compatible with ns-3 version 3.21.
