Jesse Perla

This account is verified. Learn more

@jlperla

Associate Professor, UBC Vancouver School of Economics. Interested in the (macro)economics of growth/AI/ML, and use of ML within economics itself. Swiftie.

jesseperla.com Joined February 2011

566 Following 1.5K Followers 746 Posts

Latest posts 5

Jesse Perla @jlperla Oct 17

Amazing. You must see lots of connections between macro time-series and how the physical sciences investigate the spectrum.

Mahdi E Kahou @MahdiKahou Oct 17

🧵1) Given the fantastic discussions I had CESG, I am reposting this:

Neural networks learn low frequencies first, and only later the high-frequency oscillations, if at all.

Left panel: training data and the neural-net fit.

Right panel: their Fourier transforms. code:

View quoted post on X

Jesse Perla @jlperla Oct 15

@ubcVSE is hiring this year with a focus on International Trade/Spatial Economics and Macroeconomics/International Macroeconomics. Please apply!

See econjobmarket.org/positions/12839 for more

Jesse Perla @jlperla May 15

Yes! Great paper on how overparameterization shows up in OLS; and when double descent DOESN'T occur.

When I ask about double-descent, ML people always talk about the null-space, how regularization modifies the spectrum of the Gram matrix to avoid eigenvalues near 0, etc.

Fedor Iskhakov @Fedor_Iskhakov May 15

arxiv.org/abs/2303.14151
View quoted post on X

Jesse Perla @jlperla May 14

My annual reminder that this is one of most important papers I have read on deep learning and its place in ML.

Non ML/CS/Stats readers can otherwise attribute properties to deep learning and "neural networks" that have nothing to do with that any particular hypothesis class.

Andrew Gordon Wilson Verified account @andrewgwils May 14

This is your annual reminder that we don’t need to speculate about whether we will have a “theory of deep learning” and what form it might take, because we already have a basic understanding of generalization in deep learning: arxiv.org/abs/2503.02113
View quoted post on X

Jesse Perla @jlperla Feb 24

In the latest @QuantEcon Julia release, you can now launch notebooks directly in @GoogleColab.

Just click ▶️ on the online lecture to open it in Colab.
On first run, the notebook automatically installs all required packages.