NOTEBOOKS.md

Notebooks for $\text{IAM-CL}^2\text{R}$ training and evaluation

We provide both Google Colab and IPython Notebooks. They contain the same code.

Google Colab links

• IAM-CL2R with d-Simplex-HOC
• IAM-CL2R with ER baseline
• Compatibility evaluation script

IPython Notebooks

Create virtual env and install dependencies

The following command will create a virtual enviroments and automatically install the required dependencies.

make install

Remember to select this virtual env as notebook kernel.

Download pretrained-models

Before executing any training script please use the following command to download the checkpoints of the pretrained models that are used to replace the fine-tuned model.

make download-pretrained-models

Implementations

These are the currently available implementations:

• ER baseline
• d-Simplex-HOC

Using this code we got the following results:

	Model	7 Tasks	31 Tasks
	ER baseline	$\mathit{AC}: 0.00 , \mathit{AA}: 35.18$	$\mathit{AC}: 0.00 , \mathit{AA}: 30.32$
	d-Simplex-HOC	$\mathit{AC}: 0.95 , \mathit{AA}: 68.90$	$\mathit{AC}: 0.65 , \mathit{AA}: 68.07 $

Download results

Use the following link to directly download the checkpoints that we got from the notebooks.

make download-results