Name:Naomi Patterson

Bio:

Hello! I am a machine learning engineer and data scientist at Electronic Arts (EA). Here is a brief summary of my research and interests:

• Professionally, I focus on applying learning algorithms to solve problems in the fields of game testing and defect prevention.

  • For game testing, we automate the selection of automatic and traditionally manual tests using a combination of risk based testing, machine learning and deep learning.
    • This is done as described in our SUPERNOVA paper, with the primary difference between ours and other methods being the inclusion of manual tests and the manually set features that come with them.
  • Automatic tests are easy to frame in the form of a learning problem, as they are automated by default; manual tests are another story altogether. We refer to them as “traditionally manual,” since in the past, play-testing a video game or validating user interface components required a human tester to accomplish these tasks. However, due to the increasing cost of manual testing, a motivation to find a way to automate these processes has emerged.
    • We use an internally developed scripting tool based on traditional artificial intelligence principles to automate these manual tests. Yet despite this tool being capable of testing with nearly full functionality, it lacks the stochasticity to truly replicate what manual testing can achieve. Therefore, we are replacing elements of this scripting tool with agents trained through reinforcement learning, piece by piece.
  • With defect prevention, we parse a number of features from Perforce, Jira, Workday, etc and perform gradient boosting classification on them (as described in the SUPERNOVA paper). However, we are researching how to apply deep learning to improve our results, which involves abstract syntax tree (AST) parsing.
    • Since C++ is the primary language used at EA, that is also the language we are focusing on for defect prevention. We are using Clang to extract required information from these C++ ASTs.
    • The idea is to train embeddings on these ASTs to extract meaningful information about the code structure, its similarity to other known bug causing ASTs, etc.
  • We are also starting to use learning algorithms to infer human readable information from game or program crash dumps.

• Personally, I enjoy a number of hobbies that allow me to exercise both my body and my mind.

  • My athletic passions are squash, running, cycling and yoga, with beach volleyball and tennis incorporated in the summer.
  • I love reading, with non-fiction and science fiction being my go-to genres.
    • My favourite books are Creativity, Inc by Ed Catmull and the Robot series by Isaac Asimov.
  • Working at a video game company, it is no surprise that I enjoy playing games of all kinds. I love to bring out my expansive collection of board games and retro video games to play with friends and coworkers!
    • My favourite video games are The Elder Scrolls 3: Morrowind, Romancing Saga Minstrel Song, and the Shin Megami Tensei series.
    • My favourite board games are D&D, Arkham Horror, and social deduction games such as Blood on the Clocktower.

Country:Canada

Affiliation:Electronics Arts

Research interests:Machine Learning, Deep Learning, Computer Vision, Natural Language Processing, Reinforcement Learning