Portfolio

---

Pattern Recognition

Expectation Maximization

In this project, I implemented Expectation Maximization algorithm as a solution to cluster a dataset which is a Gaussian Mixture model, includes three different Gaussian distribution.

---

Logistic Regression and Gradient Descent from Scratch

This project is about implementing a logistic regression model by scratch and updating its gradients using stochastic gradient descent method.

---

Computer Vision

Image Classification Using Traditional CV Methods

In this project, the aim is to classify a set of images using various methods. The pipeline consists of 4 steps: Feature Extraction, Finding Dictionary Centers, Feature Quantization and Classification. (In my pipeline, I used SIFT (OpenCV implementation), K-Means Algorithm (my implementation), Bag of Visual Words (my teammate’s implementation) and Random Forest (Sklearn implementation) respectively.) For training and testing, we used “Caltech20” dataset provided by TAs.

---

Deep Learning

MLP as a Neural Language Model

In this project, I implemented a neural language model using a multi-layer perceptron. This network receives 3 consecutive words as the input and predicts the next word.

---

Convolutional Neural Network(CNN) from Scratch
I implemented a convolutional neural network (CNN) architecture from scratch, using Pytorch. Tried different data augmentation and optimization techniques and to boost the performance on CIFAR10 dataset.

Source for CIFAR10 dataset examples: https://www.cs.toronto.edu/~kriz/cifar.html

---

Spatio-Temporal Attention for Manipulation Failure Detection (Bachelor’s Thesis)

---

[Implementing a Variational Auto Encoder(VAE)] The aim of this project is to implement a VAE, where the encoder is an LSTM network and the decoder is a convolutional network. Training and testing was made on MNIST dataset.

Page template forked from evanca

Hosted on GitHub Pages — Theme by orderedlist