jhrmnn

25 ms·

301.
by jhrmnn 8y ago
Optimizing a neural network is not really minimization. Sampling the training data in batches introduces a stochastic element, so a better analogy is perhaps something like Langevin dynamics. Decreasing the learning rate would then very rou
302.
by jhrmnn 8y ago
A macrostate is any particular probability distribution of microstates. Usually these are picked to reproduce some macroscopic observable, such as temperature, pressure, volume, magnetisation, etc.
303.
by jhrmnn 8y ago
And with something like Mypy, you can get most of the benefits of static typing even in Python, if you want to. But Rust has so much more to offer, such as memory ownership.
304.
by jhrmnn 8y ago
DL is a branch of ML.
305.
by jhrmnn 8y ago
Which is necessary because each EU country still has its own local laws, which the local T&C must abide. If you travelled to Germany and opened a bank account at a local branch, you’d get German T&C.
306.
by jhrmnn 8y ago
Germany is still nowhere close to what's common in other EU countries. Pretty much all Germany's neighbors have much higher adoption of terminals accepting MC/Visa.
307.
by jhrmnn 8y ago
That's a false analogy. No one forces you to stay with your bank. If you are an EU citizen, you can open an account in any other EU country. Mastercard was forcing merchants to only use banks within their respective countries.
308.
by jhrmnn 8y ago
Geometric proofs of problems that do not seem to be geometric are the most satisfying in my experience.