Ciro Santilli @cirosantilli 37

A parameter that you choose which determines how the algorithm will perform.

In the case of machine learning in particular, it is not part of the training data set.

Hyperparameters can also be considered in domains outside of machine learning however, e.g. the step size in partial differential equation solver is entirely independent from the problem itself and could be considered a hyperparamter. One difference from machine learning however is that step size hyperparameters in numerical analysis are clearly better if smaller at a higher computational cost. In machine learning however, there is often an optimum somewhere, beyond which overfitting becomes excessive.

The total derivative of a function assigns for every point of the domain a linear map with same domain, which is the best linear approximation to the function value around this point, i.e. the tangent plane.

E.g. in 1D:and in 2D:

Every vector space is defined over a field.

E.g. in ${\mathbb{R}}^3$, the underlying field is $\mathbb{R}$, the real numbers. And in ${\mathbb{C}}^2$ the underlying field is $\mathbb{C}$, the complex numbers.

Any field can be used, including finite field. But the underlying thing has to be a field, because the definitions of a vector need all field properties to hold to make sense.

Elements of the underlying field of a vector space are known as scalar.

For an example with context, have a look at E. Coli K-12 MG1655 and the second protein of the genome, e. Coli K-12 MG1655 gene thrA.

Group of the unitary matrices.

Complex analogue of the orthogonal group.

One notable difference from the orthogonal group however is that the unitary group is connected "because" its determinant is not fixed to two disconnected values 1/-1, but rather goes around in a continuous unit circle. $U(1)$ is the unit circle.

2013 notes: www-thphys.physics.ox.ac.uk/people/AlexanderSchekochihin/A1/A1MT13.html

Created by Dr. Rod Nave from Georgia State University, where he worked from 1968 after his post-doc in North Wales on molecular spectroscopy.

While there is value to that website, it always feels like it falls a bit too short as too "encyclopedic" and too little "tutorial-like". Most notably, it has very little on the history of physics/experiments.

Ciro Santilli likes this Rod, he really practices some good braindumping, just look at how he documented his life in the pre-social media Internet dark ages: hyperphysics.phy-astr.gsu.edu/Nave-html/nave.html

The website evolved from a HyperCard stack, as suggested by the website name, mentioned at: hyperphysics.phy-astr.gsu.edu/hbase/index.html.

Shame he was too old for CC BY-SA, see "Please respect the Copyright" at hyperphysics.phy-astr.gsu.edu/hbase/index.html.

exhibits.library.gsu.edu/kell/exhibits/show/nave-kell-hall/capturing-a-career has some good photo selection focused on showing the department, and has an interview.

Kell hall is a building of GSU that was demolished in 2019: atlanta.curbed.com/2020/1/31/21115980/gsu-georgia-state-atlanta-kell-hall-demolition-park-library-north

Basically means "company with huge server farms, and which usually rents them out like Amazon AWS or Google Cloud Platform

TODO what's the largest molecule done on a classical computer?

On particle exchange:so it is a generalization of bosons and fermions which have $\theta =0$ and $\theta =\pi$ respectively.

Key physical experiment: fractional quantum Hall effect.

This was the first major commercial computer hit. Stlil vacuum tube-based.

SGML predecessor.

The traffic is designed for cars, which makes many red stops for bicycles completely stupid.

In a bicycle you just have too much more control and awareness than in a car, so if the way is completely clear, you should be allowed to stop, look if the way is clear, and then run reds.

Of course, this does increase the chances of hitting pedestrians a little bit. But the risk change feels so little that it would be worth it. Studies quoted by Wikipedia corroborate. It just feels extremely unintuitive to make cyclists stop in certain places when the street is clear.

Unlike abugida, these actually make you guess vowels, which are mostly or all not written down in any way. Terrible.

E.g.: the main Arabic script.

Essentially, defining an holomorphic function on any open subset, no matter how small, also uniquely defines it everywhere.

This is basically why it makes sense to talk about analytic continuation at all.

One way to think about this is because the Taylor series matches the exact value of an holomorphic function no matter how large the difference from the starting point.

Therefore a holomorphic function basically only contains as much information as a countable sequence of numbers.

A (multi-user) blog is the hello world of the web, so creating one of those is the best way to quickly evaluate web technology, i.e. time to Hello World.

Some new frameworks like FeathersJS are making a chat app instead, as that highlights the push notifications a bit better.