www.braintank.ca

Author: JonathanVinden

  • BrainTank Deep Learning Five Week Course

    BrainTank Deep Learning Five Week Course

    BrainTank Deep learning is a 5 week workshop that was develped designed and taught by myself, with overview and guidance from faculty at the University of Guelph’s computer science department. It was presented in October of 2021 to a class of ~20 students. The course’s goals were to:

    1. Minimize the barrier of entry, that plagues deep learning. We aimed to create a curriculum that gave grounded lessons about the field without advanced mathematical knowledge.
    2. Give students a theoretical foundation to the basic’s of working with deep learning.
    3. Provide students with hands-on problems that can be solved with deep learning.
    4. Get students up and running with the technical ability to make their own neural networks in pytorch.

    Each workshop class consisted of a one hour lecture, and one hour of hands on coding time.

    Topics covered were from using machine learning in simple linear regression, to convolutions.

    The recordings of the lectures are available below.

    Collaborators: University of Guelph’s Society of Computer and Information Science

    Job Title: Founder

    University: University of Guelph

  • Deep Q Learning and Self Play Beat Humans at Super Auto Pets

    Deep Q Learning and Self Play Beat Humans at Super Auto Pets

    Super Auto Pets is a popular online zero-sum extensive form game that emphasizes taking discrete actions to ensemble a cohesive team of pets to win.

    To better understand various principle related to Deep Q Learning, I designed and built a AI modet to complete agains humas in the game. My work:

    • Broke the challenge into sub-components to minimize cost of computation.
    • Reduced a large action space into one dramatically smaller, but maintaining all the actions available in the game.
    • Created a domain-specific search algorithm that increased performance from 3.6 to 5.5 points per game.
    • Oversaw optimization. Tested how hyperparameters of the model affected AI performance.

    I was successfully able to develop an AI that outperforms human players in diverse game scenarios.

    This video explains the project in detail.

  • Using Transformers to Generate Crossword Puzzles

    Using Transformers to Generate Crossword Puzzles

    For this project, I trained a set of transformers to generate crossword puzzles, akin to the New York Times crosswords.

    • Used Huggingface’s API to utilize transfer learning for the NL tasks.
    • Used beam search to create multiple board candidates that were evaluated at each step.
    • Puzzles are available to play at nickvinden.com/crossword.
    Play Now

    How to Play

    Select cells with your mouse and type them in with your keyboard. To get a new puzzle click “New Puzzle”. There are 3 different options for puzzle clue generation.

    1. GPT4: Best Model.
    2. GPT3.5: Good Model.
    3. T5: Miserable solving experience.

    To see if your answers are correct select “autocheck”.

    Give a go at cracking these completely AI generated crossword puzzles.

  • Comparing Instance Segmentation with Models Trained with and without GAN training samples

    Comparing Instance Segmentation with Models Trained with and without GAN training samples

    Note that this is a project that is currently under development.

    • Extension to paper Exploring and Classifying Beef Retail Cuts Using Transfer Learning (2022), by Abdallah Abuzaid et al.
    • The goal of this project is to reduce the cost of manual human image annotation.
    • Used a StyleGAN2 architecture to take a set of 10 beef cut image-annotation pairs, and increase the set of images from 10 to as many as needed.
    • I intend to measure under which conditions using a GAN to create more data is beneficial to instance segmentation.

  • Using Visual Transformers in Sequential Salience Prediction

    Using Visual Transformers in Sequential Salience Prediction

    Collaborators: Dr. Neil Bruce

    Job Title: USRA Recipient, Undergraduate Researcher

    University: University of Guelph

  • Using Self Supervised Learning for Annotation Reduction in Segmenting Beef Cuts

    Using Self Supervised Learning for Annotation Reduction in Segmenting Beef Cuts

    Collaborators: Dr. Dan Tulpan, Dr. Luiza Antonie

    Job Title: Undergraduate Thesis Writer

    University: University of Guelph

  • Using Machine Learning to Predict Antibacterial Resistance

    Using Machine Learning to Predict Antibacterial Resistance

    Collaborators: Dr. Dan Gillis, Dr. Theresa Bernardo, Matthew Kreitzer, Rashi Mathur, Xavier Ifill, Luc Dube

    Job Title: Undergraduate Data Scientist

    University: University of Guelph

    Program Overview

    Problem / Description

    The widespread and often unnecessary use of antibiotics in humans, livestock, and agriculture accelerates the evolution of bacteria, leading to the development of resistant strains, making antibacterial resistance progressively worse.

    Specifically, the overuse of antibacterials on companion animals is under researched.

    Research team of Dr. Theresa Bernardo and Matthew Kreitzer, is working with the class of CIS*4020 to create a dashboard prototype that vets can use to predict which drugs are resisted by which bacteria.

    We are given access to a private dataset of patient information, and infection information as features, and are tasked on predicting susceptibility test results.

    Method

    We designed an experiment to compare the efficacy of neural networks vs. k-clustering techniques on predicting antibacterial resistance.

    We find large gains in both of the models, by adding interdependent meta information from other entries in the dataset. (e.g. How do patients near you effect the diseases you can get? How does diseases change over time? How do repeat patients differ from one-time patients?)

    We will present our findings in a dashboard, and a presentation to the research team.

    Results

    Results are not yet public. Will be updated



  • Reward Augmented Expectations in Decision Making Problems for Long-Horizon Planning

    Reward Augmented Expectations in Decision Making Problems for Long-Horizon Planning

    Collaborators: Dr. Animesh Garg, Raeid Saqur

    Job Title: Undergraduate Research Assistant

    University / Lab: PAIR Labs, Vector Institute, University of Toronto

  • Analysing Siamese Neural Network Architectures for Computing Name Similarity.

    Analysing Siamese Neural Network Architectures for Computing Name Similarity.

    Collaborators: Dr. Luiza Antonie, Jeremy Foxcraft

    Job Title: Undergraduate Researcher

    University: University of Guelph

  • Roboethics Guest Speaker Talk

    Roboethics Guest Speaker Talk

    Collaborators: Roboethics Club at University of Guelph

    University: University of Guelph

    Description

    Note that as of September 28th 2023 this talk has not yet taken place.

    Preparing a 30 minute presentation on the history of automation and its possible effects on wealth inequality. This talk is based off of the research done at MIT, entitled “Automation drives income inequality” (link)