Alea Axis Released

I’m excited to announce my new company’s first tool, Alea Axis1. Alea Axis brings the power of deep learning to your browser. What would have required installing multiple libraries and extensive knowledge of deep learning can now be done with several clicks.

Alea Axis is more than just a user interface for TensorFlow functions. It implements workflows I spent years learning/perfecting as a deep learning researcher2, and having the entire process in one place allows for a consistent philosophy.

I’ve had a lot of success in my career analyzing data that I wasn’t supposed to be, and I believe there are a lot of discoveries that aren’t being made due to the high barrier to entry. This barrier still exists at Alea Axis; however, it is much lower since the need for writing code has been removed.

Alea Axis is a bet against the current trend of automation. Reasonable defaults are often provided, but the data doesn’t analyze itself. Instead, users are assisted throughout the process with helpful messages.

I don’t have a specific roadmap for new features, so feedback is welcome at X, or YouTube comments.

Footnotes

  1. Deep learning can feel a bit like throwing dice since it’s not clear exactly how to build your model, and the weights are randomly initialized. The name comes from the Latin phrase “Alea iacta est”, i.e. “the die is cast”. ↩

  2. Alea Axis provides complete control over batching/epochs, resets initial weights and optimizer state each fold split, restores best weights even if epoch limit is reached or NaNs are encountered, plots live metrics, prints per fold metrics and final metrics, provides normalization option that correctly calculates parameters from training fold. For survival: provides an efficient C-index metric, an efficient minibatch-compatible Cox loss, and a final C-index calculation resistant to small test folds. ↩