Algorithm 1.

Online learning algorithm for training population-based encoding models

1:	G0 ← 0, ${\mathit{w}}_v^0\leftarrow \mathbf{0}$, n0 ← 0, λ0 = 0
2:	While new data* is available: X, ${\mathit{r}}_v^1$, n1
3:	$\theta =\frac{n^1}{n^0+n^1}$
4:	F1 = DimensionReduction(ResNet(X))
5:	G1 = [F1]TF1/n1
6:	G = (1 − θ)G0 + θG1
7:	${\mathit{w}}_v=(1-\theta ){(\mathbf{G}+\mathrm{\lambda }\mathbf{I})}^{-1}({\mathbf{G}}^0+{\mathrm{\lambda }}^0\mathbf{I}){\mathit{w}}_v^0+\theta {(\mathbf{G}+\mathrm{\lambda }\mathbf{I})}^{-1}{[{\mathbf{F}}^1]}^{\mathrm{T}}{\mathit{r}}_v^1/n^1$ with cross validation
8:	G0 ← G, ${\mathit{w}}_v^0\leftarrow {\mathit{w}}_v$, n0 ← n0 + n1, λ0 = λ
9:	Output: wv

^*X is the new visual stimuli, ${\mathit{r}}_v^1$ is the cortical response, and n¹ is the number of samples