Just as no amount of flat temperature data would dissuade the climate cult, so no amount of flat (or even negative!) data on mask effectiveness will shake the faith of the Church of Covid. Beautiful Theories cannot be disproven by mere information.

(MAJOR NOTE) The reason Manafort took over Trump's campaign in just 3 weeks despite being hired as a mere delegate counter is that ROGER STONE initiated a whisper campaign against Corey Lewandowski. That campaign started...days after Manafort returned from meeting with WikiLeaks.

2/ Stone and Manafort are longtime business partners; we'd expect Manafort to tell Stone what Assange told him, and for him to expect Stone to then pressure the campaign not only to hire him but quickly elevate him.

And who was Stone in touch with at the campaign? Trump himself.

3/ I bet you Manafort's late March emails are very interesting and that some of them are to/from his longtime business associate Roger Stone. And I bet Roger Stone's late March phone calls are very interesting and that some of them are to/from his longtime friend Donald Trump.

4/ At the end of March 2016, Trump suddenly convenes a NatSec meeting. At that meeting he directs his NatSec team to change the RNC platform in July to benefit Putin. Who later takes credit for that change?

Paul Manafort's business associate, Kremlin agent Konstantin Kilimnick.

5/ Why did Donald Trump suddenly convene a NatSec meeting 3 days after he finally hired Manafort? Why did he issue a pro-Kremlin directive at that meeting that Manafort's camp would later take credit for? What did Manafort tell Trump in the days leading up to that NatSec meeting?

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.

SRISAILAM: KASHI OF THE SOUTH (AP)
#andhratemples

The divine abode of Mallikarjuna Swamy (Jyotirlinga) and Devi Brahmaramba (Shaktipeeth).

When Ganesha was married before Kartikeya, he left in anger to Kraunch mountain. Shiva and Parvati also followed him there.

Shiva assumed the form of Jyotirlinga and resided on the mountain by the name of Mallikarjuna. Mallika means Parvati while Arjuna is another name of Shiva
It is also said that once, a princess Chandravathi, saw a cow shedding milk on a rock resembling a shivlingam.
@DubeyYash1911

She started praying to the lingam daily and used to offer jasmine (mallika pushpam) daily. Pleased Shiva appeared before her and agreed to wear jasmine garland (mallika pushpmala) on his head permanently. Since then, Shiva became known as Mallikarjun Swamy.

At this place, Shakti took the form of Bhramaram (bee) to kill Arunasur and settled as Bhramaramba Devi. The place is a Shakti Peeth as the neck portion of Devi Sati fell here.

It is said that by merely seeing the hill, one is emancipated from all his sins and worries.