skaraoglu

Photometric-Stereo

In this paper, we implement a standard photometric stereo algorithm. With the albedo being unknown and not constant for the images, this program takes multiple images and light source direction information as an input and produce the albedo map, surface normals map, heights map and 3d plot of the surface for provided input. All the images used in this project are provided in the homework assignment and stored in "src" folder. Several output plots shown in the work are stored in the "out" folder.

Community-Detection-on-Graphs-with-Multiple-Layered-Community-Structure

This project focuses on the community structure analysis on graphs with multiple layers of community structures embedded in it. To conduct experiments on multiple layers of community structures in the graph, graphs that are based on the same nodes but wired with different edges are merged. Then the merged graph is analyzed by different techniques - such as community detection algorithms, overlapping community detection algorithms, nested community detection algorithms - to recover information about community structures set in the merged graph.

Approximate-Inference

Focusing on approximate inference methods and comparing their effectiveness on certain scenarios.

BernoulliNB-MultinomialNB

Bernoulli and Multinomial Naive Bayes classifiers are trained and tested on different datasets. We present and compare the accuracy scores for both Bernoulli NB and Multinomial NB models.

CNN-and-RNN

Experiment with PCA and CNN on CIFAR10 dataset, time series data analysis with LSTM and GRU

CNN-Experiment

Experimenting with fully connected NNs and CNNs, using the Keras open source package.

Community-Detection-of-Superimposed-Relational-Structures

Algorithms for “community” or cohesive subgraph detection are frequently used to analyze network representations of complex systems, and constitute an active area of methodological research. Algorithms for overlapping community detection are of particular interest, because it is often unrealistic to form strict partitions of natural systems, with at least some nodes participating in multiple clusters. However, the literature on overlapping communities does not distinguish different sources of potential overlap. This paper begins by discussing randomness, bridging nodes, and superimposed relational structures as three distinct sources of overlapping communities. It then focuses on the third source (neglected in the literature), by constructing test graphs that merge two distinct relational structures and comparing the communities identified by eight partition and overlapping community algorithms. Results show that most algorithms focus on the same source structure in each graph, ignoring the other, although a few algorithms find combined or nested communities from both structures.

Decision-Tree-and-ANN

This project focuses on the implementation and evaluation of two supervised learning schemes: Decision Trees and Feedforward, Multilayer Neural Networks.

Decision-Tree-and-Naive-Bayes

Decision Tree and Naive Bayes implementation and experiments.

drsenweb

Event-Maker

This web application is designed to generate event files with event information provided by the user. This application is capable of create calendar event files with ".ics" extension that can be emailed or shared and read into the recipient's calendar applications.

Face-Recognition-Analysis-of-Triplet-Loss-based-Neural-Networks

Face recognition gained popularity in computer vision and the number of methods and techniques used in the process are increased. This paper provides experimental analysis on face recognition process with Triplet Loss function based network architectures. We implemented several different techniques to improve the overall computation time of the process - to be able to implement live detection on low computational resources - and we compared the results.

Fairness-and-Bias

This report focuses on Fairness and Bias in Machine Learning (ML). Reviewed some researches that influenced the search of fairness in ML systems. We provide some insights to explain these concepts and how to avoid the bias and ensure fairness in ML system designs.

Forest-Cover-Type-Classification

In this paper, we analyze the Forest Cover Dataset provided by UCI, build numerous classifiers to classify the forest cover type of a parcel of land, and evaluate the performance of the classifiers. To achieve this goal, we applied a step by step approach to the problem and explained the process thoroughly. We start the process with the data preprocessing. After adding id column and headers to the data in MS Excel, the data is stored in a “Comma Seperated Value” (.csv file) format. Following, definition of the libraries used in this project are provided. After applying data processing to the dataFrame, we examined and explained some features about the data that is crucial in the classification processes. In this process; plot of the data, correlation in the data columns, distribution of the data are the main focus. Furthermore, data is classified by different classifiers and these classifiers are cross validated with several hyperparameters to find the best tuning for each classifier. Afterwards we trained the classifiers with best hyperparameters and tested the performance on data. Finally we compared and analyzed the results.

Harmonization-Project

This project creates a random melody and try to get a harmony between selected chord progression and evolved melody.

LordOfTheRings-EmpiricalStudy

The Lord of the Rings novels adapted to cinema with three movies. This project analyzes the structure with scene by scene analysis of the movies, characters with details and some other factors affect the relationship between characters. These variables are used to create a network to see the relations in the Lord of the Rings cinematic universe.

MDD-Analysis

Major Depression Disorder Analysis with Gene Expression and Demographic Symptom Data

Harmonization-Reloaded-Mathematica

Evolutionary music project to provide harmonization between user selected chord progression and random melody in Mathematica.

kMeans-and-ExpectationMaximization

Conducted the experiment on three part; first, evaluated the clustering results for Gaussian distribution datasets, following, utilized k-Means algorithm for image compression and employed the algorithms on Mall Customers dataset.

kNN-and-Local-Search

In this project, the main focus is the adaptation with nearest neighbor and local search methods.

LogisticRegression-and-GaussianNB

Gradient Naive Bayes and Logistic Regression with Gradient, L1 and L2 regularization implementations from scratch

MaximumLilkelihoodEstimation

Deep Learning Homework 1: Maximum Likelihood Estimation

Multi-armed-bandits

Multi-armed bandits experiment with epsilon-greedy, UCB, Thompson sampling, Bayesian-greedy and HA-UCB

NeuralNetwork

Simple Neural Network implementation with only numpy.

PCA-with-MultinomialNB

Trained and tested a Multinomial Naive Bayes algorithm on MNIST dataset, implemented PCA on MNIST dataset, then trained and tested the same Multinomial NB classifier with the results of different PCA applications.

project_smartyields

Data Science Foundations class project from Selim Karaoglu and Jasper Green

qmk_firmware

Open-source keyboard firmware for Atmel AVR and Arm USB families

SEITR

An Extended SIR Mathematical Model for Lumpy Skin Disease and Associated Properties

tuner

Warmth-Experiment

Website