Did we get dietary saturated fats all wrong? The #HADLmodel provides a new understanding and an opportunity to get it right. THREAD👇👇👇
@simondankel @kariannesve

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.

'Tera mera rishta kya, hindu, hindu,hindutva' is a clear take on 'tera mera rishta kya,la ilaha illallah'. As much as Hindutva hates the monotheistic religions especially Islam, there's a strange wistfulness and subliminal need at display now, to BE that they hate. 1

Charlie Hebdo doesn’t come up as a cautionary tale now but rather 'why cant we be like that'?
Growing up, Hanuman was an adorable deity for us, who swallowed the sun, or who couldn’t figure out what was Sanjeevani booti so he picked up a whole mountain and came back with it. 2

Or he was loyal, admirable and when he cleaved his chest, you saw both Ram and Sita. Now we have a lot of “Angry Hanuman” stickers. Why is a cute childlike God angry? We don’t know. It's a convenient blank space where to paint all the angst of Hindu Khatre mein hain narrative. 3

The other strange thing that has started happening is pictures of Ram appearing sans Sita. “Jai Siya Ram” has become “Jai Shri Ram”. Some people disingenuously explain it as being the same thing - because Shri is a name of Lakshmi. Not really. Siya is an apbhransh of Sita. 4

Shri is a prefix of respect in this case. Hinduism is being militarised. A woman/Sita has no place in it. Even the militaristic Goddess that is Durga has no place in this conception of neo-Hindu divinity, as evidenced by Dilip Mandal wondering where Durga came from. 5

You want to buy a house.

You take a loan from the bank.

The bank asks you, “what if you don’t pay back?”

You say, “take my home and sell it”.

(Thread)

That makes sense. The bank gives you a loan and in case you don’t pay, the bank will take over your house and sell it and recover its money.

That’s collateral.

Banks also lend money to big corporations. With them, banks don’t always have collateral.

So a loan where the bank has collateral is considered safer for the bank.

In the early 2000s, the banks in the USA were giving out loans to people for buying houses.

The economy had just recovered from the tech bubble in 2000 where many banks had given out loans to businesses that had shut down.

In that backdrop, the banks found giving home loans safer.

why do I hear boss music