Table 2: Single-unit brain region classification for the IBL dataset.

The accuracy and F1-scores are reported for three conditions: (i) a linear layer and (ii) MLP on top of the frozen pretrained representations (for VAE and NEMO), and (iii) after MLP finetuning. We only ran PhysMAP one time as it is deterministic.

Model	Linear		MLP		MLP fine-tuned
	Acc	F1	Acc	F1	Acc	F1
Supervised	N/A	N/A	N/A	N/A	0.45 ± 0.01	0.42 ± 0.01
PhysMAP	0.311	0.271	N/A	N/A	N/A	N/A
VAE	0.33 ± 0.00	0.29 ± 0.00	0.40 ± 0.01	0.37 ± 0.00	0.43 ± 0.00	0.40 ± 0.00
NEMO	0.42 ± 0.00	0.40 ± 0.00	0.45 ± 0.01	0.42 ± 0.00	0.47 ± 0.01	0.44 ± 0.01