I remember the Pulse terrorist attack vividly.
I remember walking into work and seeing the quiet, concerned look on my coworker's faces unsure of what to say or do.
I remember my friends calling me to ask if I was ok and suddenly feeling like the LGBT community was one.

1. There are more people added on the list of arrests and executions of famous people but no further intel is available at this time.

ARRESTS and EXECUTIONS OF FAMOUS PEOPLE 2020 UPDATED
Explanation of the latest military tribunal arrests and executions of famous people in 2020.

2. The tribunals are arresting these people in plain clothes. We saw the Obama’s arrest, although we didn’t know it was an arrest at the time. The arrest was done nonchalantly so as to not get noticed by the public.
All arrests of these famous people were done this way.

3. If the person is convicted of child sex trafficking, crimes against humanity or treason, they will be executed.
If the person arrested has evidence on someone else to help their prosecution, or is willing to help Trump, & they want to cooperate, they may be allowed to do so.

4. A sentence of execution down to life in prison may be accepted if their evidence is worth anything.
When a plea deal is done, the defendant must present evidence they have to make their sentence lighter. This information is given to the prosecution, defense, and to the judge.

5. If evidence is accepted, they are presented the plea deal. They will have to make a guilty plea to the crime with a video confession & written statement. That is how all plea deals work.

All famous people arrested were either treasonous or involved in satanic worship rituals.

You want to know something about how bullshit insane our brains are?
OK, so there's a physical problem with our eyes: We move them in short fast bursts called "saccades", right? very quick, synchronized movements.
The only problem is: they go all blurry and useless during this

having your vision turn into a blurry mess every time you move your eyes is obviously not a good idea, so our brains hide it from us. Now, imagine you're an engineer and you have this problem.

You've got some obvious solutions you could do.
1. make the vision go black during movement. (Some VR games do this!)
2. just keep showing the last thing we saw prior to movement

both are good options with different downsides, but OH NO. this is assuming everything makes sense and is chronological and (regular) logical.
Your brain does neither of these options, really.

first, it basically puts your visual system on "pause". You're not seeing blackness or even nothing, you're just not seeing period.
then when you finish your saccade, it shows you what you now see at the new position. and then it pretends it can time travel.

So, today, for the first time in 25 (!) years of Apple, I downgraded. From the 2016 MacBook Pro to my 2013, which I had kept in a drawer...

It was pretty simple to do—Apple Time Machine backups let me do it with one click.

That first tweet captures, in two pictures, how badly Apple has “lost the plot” (to quote @wylieprof). On the right is the Apple MagSafe adapter, from 2013. On the left, what I had “upgraded” to.

Thanks, Apple! I really was nostalgic for worrying about yanking my computer off the table.

Oh and I really appreciated not knowing if my computer was charging. What was great was the little whoop sound you used, so that the speaker before me could be informed I was charging my laptop.

Important paper from Google on large batch optimization. They do impressively careful experiments measuring # iterations needed to achieve target validation error at various batch sizes. The main "surprise" is the lack of surprises. [thread]

https://t.co/7QIx5CFdfJ

The paper is a good example of lots of elements of good experimental design. They validate their metric by showing lots of variants give consistent results. They tune hyperparamters separately for each condition, check that optimum isn't at the endpoints, and measure sensitivity.

They have separate experiments where the hold fixed # iterations and # epochs, which (as they explain) measure very different things. They avoid confounds, such as batch norm's artificial dependence between batch size and regularization strength.

When the experiments are done carefully enough, the results are remarkably consistent between different datasets and architectures. Qualitatively, MNIST behaves just like ImageNet.

Importantly, they don't find any evidence for a "sharp/flat optima" effect whereby better optimization leads to worse final results. They have a good discussion of experimental artifacts/confounds in past papers where such effects were reported.