Starting a startup is hard. This thread will contain some key advice I think are particularly noteworthy:

1. “If you’re not embarrassed by the first version of your product, you’ve launched too late.” - @paulg The first version of my product was not great, but hey, it worked.

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.

10 Signs of Social Intelligence 🧠

1-Smile
2-Good posture
3-Strong eye contact
4-Fluid hand gestures
5-Let’s others talk more
6-Listens & relates
7-Comfortable Touching people
8-Walk with a swagger
9-Speaks at slower pace
10-Keeps calm under criticism

[ QUICK THREAD ]

1-Smile

How do you feel when you see someone that looks angry?

It’s a downer

No one will want to interact with you if you look pissed all of the time, it’s just negative energy

Plus, smiling is a sign of confidence and boosts others moods

2-Good posture

Don’t walk Into any social setting looking like the hunchback

It’s weak and makes you look low energy

Plus, you’re displaying low status with weak posture

Stand up straight with your chin up & chest out like a god

Walk like you’re proud of yourself

3- Strong eye contact

Ever shake someone’s hand & their eyes are either looking down or darting?

It’s weak & you probably won’t trust them

Or in conversation?

Eye contact while talking means you’re listening and paying *Attention*

Strong eye contact displays confidence

4-Fluid hand gestures

Use your hands to talk & express your points

Don’t stand there like a robot

Using the hands can make what you’re talking about exciting

It can also be used to emphasize

Exciting = positive emotion

Remember, making people FEEL is what’s important

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.