Something I want to flag for #MAEdu this morning:

The Board of Ed meets next Tuesday; agenda is here:

MAHARAANI TARARANI: THE QUEEN OF MARATHAS

Rani Tarabai was the brave daughter-in-law of Chhatrapati Shivaji and one of the greatest medieval Monarchs India ever had.
She was born in 1675 to one of the bravest commander-in-chiefs of Shivaji's Army named Hambir Rao Mohite.

Tarabai was just 8yrs old when she was married to Shivaji's younger son, Rajaram. Mughals and Marathas were constantly at war with each other. Shivaji was crowned the ruler of the Independent Maratha Swarajya in 1674 which was established by him.

However,the great leader died in 1680&the Maratha Samrajya went into a complete&total chaos.Marathas saw some very tough times ahead. In 1689,when Raigad Fort fell to the Mughal Army of over 15000,Shivaji's first wife Saibai and his elder son Sambhaji were captured&put to death.

Sambhaji's wife and son Shahu were taken prisoners to the Mughal court. During the same battle, Shivaji's second son Rajaram(Tarabai's husband) , who was the heir to the throne now, managed to escape using a disguise. He made his way to the Gingee Fort in Tamil Nadu.

Mughal General Zulfikar Khan, however, followed them to Gingee laying siege to the Gingee Fort. From 1690-1698,This was the longest siege by a Mughal Army of any Fort which stands testimony to the military and diplomatic credentials of Rani Tarabai.

8 March 1679 - The date on which 300 saffron clad warriors gave their life to protect temples from mugal army. Thread➡️
Aurangzeb's huge army, which was adorned with artillery, elephant and horses, under the leadership of Commander Darab Khan had marched to punish the ruler of/1

Khandela Raja Bahadur Singh, who became a rebel due to the anti-Hindu policies of Aurangzeb, and to destroy the temples of Shekhawati. Bahadur Singh to compete with such a large army, vacated Khandela for the purpose of conducting guerilla warfare and went Kot Sakrai /2

located in the inaccessible hills of Sakrai. People of Khandela also vacated Khandela and hid in nearby villages.

But some Shekhawat warriors did not like this as they knew that the royal army would not leave a single temple located in Shekhawati unbroken./3

so soon the teams of self-sacrificing Dharma-protectors Shekhawat heroes started reaching Khandela to protect the Dev temples.

Sujan Singh Shekhawat, the young thakur of Chhapoli, was coming back from his wedding when he heard about the imperial army from Shepherds./4

झिरमीर झिरमिर मेहा बरसे,मोरां छतरी छाई |
कुल में है तो आव सुजाण, फोज देवरे आई ||

if you are born in your glorious clan, then come, the army have come to destroy temple. /5

1/ Throughout my startup career, I’ve seen many founders have an aversion to sales. This is a mistake that can kill your startup.

2/ Sales is often viewed as either a saving grace or proof that the product isn’t good enough (because it should sell itself). Neither are ever true. Some common mistakes that result in...

3/ Mistake 1: Hire a sales rep before reaching product/market fit to get your initial batch of customers. This is a mistake because founders need to work through their MVP with early adopters to truly understand what it is they’re selling.

4/ Mistake 2: Reach product/market fit, need to scale, and rely entirely on self-serve. For enterprise products that require big commitments and internal shifts, almost no product is self-explanatory enough to sell itself.

5/ Mistake 3: Make a first sales hire who isn’t scrappy enough to help mold the sales process from scratch. Some salespeople are amazing at their jobs, but not cut out to establish the processes that others end up following. This skillset is what @rdedatta calls a “sales ninja”.

If there is one thing I have learned it is that people love bullshit. They buy bullshit stories from professional bullshit artists, accept bullshit philosophies, purchase bullshit how-to-books, vote for bullshitting politicians.

At first, I resisted believing this because it is illogical and dumb and frankly, doesn't reflect well on the human race of which I am a member (and thus have long been hopeful that it is more of a distinction than it actually turns out to be.)

But if you actually are interested in facts you begin to notice how little most people care for them. Sometimes it is because they are heard to learn and bullshit is so much more widely and easily available. Sometimes it is because facts make them uncomfortable about themselves.

Sometimes it is because just one tiny fact will upset a whole worldview that has grown as comfortable as an overstuffed chair. (Settling into such worldviews is a lot like settling into chairs like that because the longer you're there and the deeper you sink...

...the harder it is to hoist yourself up and out of your fat-assed intellectual laziness.) But I actually think the reason people have such an appetite for bullshit goes deeper, goes to an aspect of ourselves we just can't accept.

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.