Google AI: Quick, Draw! Doodle Recognition Challenge

Aleksey participated in the Quick, Draw! Doodle Recognition Challenge put on by Google AI and Kaggle in late 2018. He earned a gold medal, finishing 8th out of 1,316 teams (leaderboard).

You can find a runnable version of his best single model here on GitHub

• Date: Fall 2018
• Technologies: Convolutional Neural Networks (CNN), Python, Pytorch

VanCityGrowler

VanCityGrowler is a website and twitter bot that helps you find the beer you're looking for at Vancouver's craft breweries.

The project runs entirely serverless on AWS.

Aleksey is the sole developer on the project and collabroates with another local craft beer enthusiast.

• Date:2017
• Link: VanCityGrowler.ca
• Technologies: AWS (especially Lambda), Python

Landmark Lens

Landmark Lens is an iOS app. Take photos in crowded places. Landmark Lens removes people, cars, and moving objects. ​Use a tripod, make a stand, or lean your phone against something. Press start and the app does the rest. ​Get the perfect shot of a landmark without any pesky tourists in it. No getting up at sunrise or waiting forever for all of the people to leave.

Aleksey worked with two other collaborators to bring the app to market. Aleksey contributed in image processing research, product management, and limited production code.

• Date:2016
• Link: Landmark Lens
• Technologies: Matlab, iOS, Swift

Hailsense

Hailsense is NGRAIN's new AI-driven hail damage assessment product. We use machine learning to size and count hail dents in 3D scans of hail damaged vehicles in order to create accurate, objective, and consistent repair estimates.

Hailsense is a web application with an AWS back-end for heavy lifting.

Aleksey led the machine learning research effort, faces the customer and leads development as Product Manager, and writes production code when necessary.

• Date: 2016/17
• Link: Hailsense
• Technologies: 3D Triangular Meshes, Python, XGBoost, Theano/Lasagne, AWS, Django, 3D Scanners

NIPS 2017 Adversarial Images Challenge

Aleksey participated in the 2017 NIPS Competition Track Adversarial Images Challenge has hosted on Kaggle. After winning $1,000 Google Cloud Platform credits by finishing 1st in the first defense development round, he teamed up with Ross Wightman for additional submissoins. Together they placed gold in all three competitions, finishing 5th, 9th, and 9th.

They published a paper on their attack here.

• Date: Summer 2017
• Technologies: Convolutional Neural Networks (CNN), Adversarial Images, Python, Pytorch, Tensorflow
• Sub-Competitions: Targeted Attack, Non-Targeted Attack, Defense

ISIC 2018 Disease Classification Challenge

As part of the MetaOptima R&D team, won the ISIC 2018 Disease Classification Challenge put on by the International Skin Imaging Collaboration. The task was to design and build a system to autmoatically diagnose a photo of a mole as one of seven diseases: Melanoma, melanocytic nevus, basal cell carcinoma, actinic keratosis, benign keratosis, dermatofibroma, and vascular lesion.

We won with a score of 88.5%, a significant 2.9% better than the second best team as visible on the Task 3 Leaderboard.

• Date: Summer 2018
• Technologies: Convolutional Neural Networks (CNN), Python, Pytorch

Interactive Web Visualizations

In 2012-2014 Aleksey put out a flurry of interactive web visualizations built on D3.js

K-means Clustering

Global Games, Regional Sports - London 2012

Global Games, Regional Sports - Vancouver 2010

Passing Direction

Paid to Win

English Football Value

Who won the Olympics?

Create Info Wheel (4-level)

Create Info Tree (4-level)

• Date: 2012-2014
• Technologies: D3.js, Javascript

Ontario Sunshine List

In 2012-2014 Aleksey did some extensive analysis of the publicly available Ontario "Sunshine List". A public disclosure of the salaries of the best paid public workers in Ontario. His work was referenced in both the National Post and in the CBC.

For all of the blog articles written during that period, click here.

• Date: 2012-2014