Computer Vision and Deep Learning for Moderating Visual Content - BC-348

Desired discipline(s): Engineering, Computer science, Engineering - computer / electrical
Company: Two Hat Security
Project Length: 4 to 6 months
Preferred start date: As soon as possible.
Language requirement: Flexible
Location(s): Kelowna, BC, Canada
No. of positions: 1

About the company: 

Two Hat Research has developed the next generation of moderation tools for virtual worlds and social networking apps. Our Community Sift product lets clients define which kinds of chat messages are acceptable and which are high risk. We are seeking MSc and PhD students or Post-Docs to work on this research projects. We are particularly interested in pursuing multi-year collaborations. We are committed to eliminating bullying from the web. We work with some of the largest video game, social media, and messaging companies on the internet and are scheduled to process 4 billion chat messages per day. We are deeply interested in publishing papers to showcase strong Canadian research. The work that these researchers will be doing will have a real-world impact.

Please describe the project.: 

Two Hat Security is a company that develops next generation moderation tools for social networking apps. Since visual content (e.g. images, videos) is one of the most important types of data shared by social networking apps, an important problem for the company is to identify images/videos that are offensive or inappropriate. For example, certain images/videos might contain violence, nudity, or certain objects (knife, gun, bikini, etc.) that are considered offensive. It is obviously unrealistic to have human annotators to manually sift through all images/videos online and flag those offensive contents. The goal of the proposed research program is to develop a series of computer vision and machine learning (especially deep learning) technologies that the partner company can utilize to build the state-of-the-art moderation tools for social network apps.

This is a great opportunity for a graduate student or Post-Doctoral Fellow to conduct cutting-edge research in the Rapidly-Growing Tech Hub of the Beautiful Okanagan Valley.

Research Objectives/Sub-Objectives:

Image recognition to discover high risk content
Contextual learning (text, text + image)

Methodology: We are experimenting with a variety of cutting edge deep learning techniques for image recognition using both pre-trained and proprietarily-trained convolutional neural networks.

Required expertise/skills: 

Computer vision and deep learning. Convolutional neural networks. Training binary and multiclass systems. Exceptional communication skills (the intern will have an opportunity to work directly with many notable organisations while conducting research and experimenting). Preferred disciplines: Engineering, Computing Science, ECE (Masters, PhD or Post-Docs)