Future Projects

I aim to implement a conversational bot that will assist with search on StackOverflow using Natural Language Processing, AWS and Telegram.

By calculating similarity for pieces of text, I aim to find duplicate questions from StackOverflow using StarSpace, Gensim and Nltk

In this project, I aim to create and deploy a model to App Engine for serving and create Cloud Composer and Apache Airflow environments to automatically retrain and redeploy my recommendation model with TensorFlow on GCP.

I plan to design an agent that can fly a quadcopter, and then train it using a reinforcement learning algorithm.

I wish to engineer features that can help identify cases of plagiarism in text and deploy a trained plagiarism detection model using Amazon SageMaker.

I wish to use a recurrent neural network to solve Named Entity Recognition (NER) problem by recognizing named entities from Twitter.

In this project, I wish to define and train an image-to-caption model, that can produce descriptions for real world images using Vision Deep CNN and Language Generating RNN

I seek to build a model that can read in some text and make a prediction about the sentiment of that text, where it is positive or negative by training the model on a dataset of movie reviews from IMDB that have been labeled either "positive" or "negative". Since this is text data, words in a sequence, I wish to use an RNN to build a model that doesn't only consider the individual words, but the order they appear in.

Facial composites are widely used in forensics to generate images of suspects. Since victim or witness usually isn't good at drawing, computer-aided generation is applied to reconstruct the face attacker. One of the most commonly used techniques are evolutionary systems that compose the final face from many predefined parts. In this project I wish to try to implement an app for creating facial composite that will be able to construct desired faces without explicitly providing databases of templates. I aim to apply Variational Autoencoders and Gaussian processes for this task.

I aim to use neural networks to solve sequence-to-sequence prediction tasks. Using sequence-to-sequence modeling, I aim to build a calculator for evaluating arithmetic expressions, by taking an equation as an input to the neural network and producing an answer as it's output.

This time, I plan to solve a problem of transribing hebrew words in english, also known as g2p (grapheme2phoneme) word (sequence of letters in source language) -> translation (sequence of letters in target language). Unlike most deep learning researchers do, I won't only train it to maximize likelihood of correct translation, but also employ reinforcement learning to actually teach it to translate with as few errors as possible.

As a drill exercise, the goal of this project is to implement and compare two ensembling schemes: simple linear mix and stacking.

In the part 5 of this project, the aim is to optimize a simple tracking system using Bayesian optimization with Gaussian processes, which can also be used for more complex systems optimization.

In the part 4 of this project, I seek to design a prediction model that is able to discriminate the basetracks belonging to the electromagnetic showers from the background basetracks. The goal is to get the best possible score (ROC AUC).

In the part 3 of this project, the aim is to design prediction model that will give best score (ROC AUC) and will meet the constraints of similar performance on simulated and real data (agreement check) decorrelation with the mass (correlation check).

In the part 2 of this project, I plan to train a classifier to identify type of a particle among electron, proton, muon, kaon, pion and ghost. Ghost is a particle with other type than the first five or a detector noise. Different particle types remain different responses in the detector systems or subdetectors such as tracking system, ring imaging Cherenkov detector (RICH), electromagnetic and hadron calorimeters, and muon system. The aim is to identify a particle type using the responses in the detector systems.

In the part 1 of this project, I seek to analyze the mass spectrum of particle decaying into muon-antimuon (dimuon) pairs and figure out the mass of Z boson by fitting the mixture of 1) Gaussian distribution and 2) flat distribution using scikit-optimise toolkit.

In this project, I plan to generate my own Simpsons TV script using RNNs. I wish to use part of the Simpsons dataset of scripts from 27 seasons. The Neural Network built will generate a new TV script for a scene at Moe's Tavern.

In this project, I deep dive into the realm of neural network machine translation. I wish to train a sequence to sequence model on a dataset of English and French sentences that can translate new sentences from English to French.

In this project, I use generative adversarial networks to generate new images of faces. In particular, I seek to employ a DCGAN on the CelebA dataset to generate images of novel and realistic human faces.

This project aims to classify images from the CIFAR-10 dataset. The dataset consists of airplanes, dogs, cats, and other objects. I begin with preprocessing the images, then train a convolutional neural network on all the samples. The images need to be normalized and the labels need to be one-hot encoded and then build a convolutional, max pooling, dropout, and fully connected layers. At the end, you'll get to see my neural network's predictions on the sample images.