antreasantoniou / deepclassificationbot

A deep learning powered bot capable of classifying images into user-specified categories

Antreas Antoniou

I am a Machine Learning Research Associate at the University of Edinburgh, supervised by Prof. Amos Storkey. I am a member of the BayesWatch research group and the Adaptive and Neural Computation (ANC) research institute.

Leading my research interests is Multi-Modal Learning, specifically targeting the synergistic integration of text, images, audio, and video data. This is followed by the development of Self-Supervised Methods, inspired by mechanisms of infant learning and principles of evolutionary computation.

Additional key areas include Meta-Learning, Adversarial Learning, and Optimization Techniques Inspired by Evolutionary Optimization. These are applied across both differentiable and gradient-free optimization paradigms. Other relevant research dimensions include Inductive Biases, Scalability, Computational Efficiency, and Memory-Augmented Neural Networks.

For more information see my website: https://antreas.io/home/

deepclassificationbot's Issues

Serialize and save model when initializing workspace

Motivation

deploy.load_model takes an input shape argument, which is hardcoded in bot/webapp. This value is not something dependent on the code, but rather tied to the model that goes with the saved weights. Furthermore, the saved weights are tightly coupled with the model itself.

Coupled data should saved in proximity with each other, not updated/specified by human hands.

Proposed changes

When initializing a workspace, user can specify the model builder name and input/output shapes. Build a model and save it in the workspace.

All other code should load the model from the workspace, and get input/output shapes from the model.

Portable workspace directory

Motivation

Data files created by this framework are currently saved in multiple places, which are hardcoded:

./data
./downloaded_images
./pre_trained_weights

In particular, files needed for deployment are in ./data and ./pre_trained_weights.

This leaves these user stories unfulfilled:

• When I deploy a model to a remote location, I want a simple way to specify all the files I need, so that it's easily automated and future-proof.
• When I'm experimenting with multiple classification projects, I want to easily switch between projects, so that progress is saved and easy to pick up later.

Proposed changes

Provide a command that initializes a workspace directory with the expected directory structure.

workspace/
  downloaded_images/
  wip/
    data.hdf5
  dist/
    latest_model_weights.hdf5
    model_weights.hdf5
    categories.p

All of the entrypoint scripts should take a --workspace=dir argument and look for files within the workspace:

google_image_scraper.py
train.py
deploy.py
deepanimebot/bot.py
deepanimebot/webapp.py