Notes from 21/10–28/10

Hamming, “You and Your Research” (June 6, 1995)

4 min readOct 28, 2019

The transcript is abridged/says some different things.

Do things that are by your definition, significant.
And I will cite Pasteur who said, ``Luck favors the prepared mind.’’ And I think that says it the way I believe it. There is indeed an element of luck, and no, there isn’t. The prepared mind sooner or later finds something important and does it. So yes, it is luck. The particular thing you do is luck, but that you do something is not.
Very able people work very hard all the time.
Once you get your courage up and believe that you can do important problems, then you can. If you think you can’t, almost surely you are not going to.
Become comfortable with ambiguity
Keep your door open — then you know what problem to work on
Know how to communicate!
Examine your life — you have only one life to lead
Spend Fridays on ‘big problems’.
Originality is being different. You can’t be an original scientist without having some other original characteristics. But many a scientist has let his quirks in other places make him pay a far higher price than is necessary for the ego satisfaction he or she gets. I’m not against all ego assertion; I’m against some.

AI

Foundations of ML: Parameterized Functions

A huge amount of what we do in machine learning is finding particular functions. We choose a model class (parameterized function class) and we optimize it (find a certain value for the parameters).

By viewing machine learning through this lens, it becomes apparent that all we are doing is searching through parameter space for a function that matches our data!

One commonly taut’ed fact of neural networks is that they are universal function approximators. Which means that, given enough hidden nodes, they can approximate any function. The precision with which this function is approximated is based on the number of hidden nodes in the network. This means that neural networks are a class of functions that are sufficiently generic to model an arbitrarily complex function. Getting them to do so, however, is an art in-and-of itself!

Blockchain

How to Learn About Blockchain If You Didn’t Study Computer Science

Participate in a structured learning experience. Unchained or Crypto 101 c

2. Read papers. Blockchain Basics. Decrypt. Crypto Canon. EthHub.

3. Go to events.

Ethereum Studio

Interesting way to learn, wonder how they built it.

Ethereum resources for Devs

How have I not come across this before?!

Data

Building a Successful Modern Data Analytics Platform in the Cloud

Lots of backend magic. Need to shift away from typical RDBs to noSQL for greater scalability. So you injest data in its raw form, make it usable, then make it served to who needs it.

Interactive Data Visualization with Modern JavaScript and D3

Very much a good tutorial on viz. Actually D3 can just pluck out sample code and fiddle about. But proper data science you’ll have to know how it actually works.

Development

Interview with a Pornhub Web Developer

Most of our sites use the following as a base:

Nginx
PHP
MySQL
Memcached and/or Redis

Other technologies like Varnish, ElasticSearch, NodeJS, Go, Vertica are used where appropriate.

For frontend, we run mostly vanilla Javascript, we’re slowly getting rid of jQuery and we are just beginning to play with frameworks, mostly Vue.js

I found our interview really enlightening. I was a bit surprised they didn’t use images while developing features and designs. It’s exciting to see that Pornhub continues to push the bleeding edge of the web with WebXR, WebRTC, and Intersection Observer. I was also happy to see that they consider the current set of web APIs sufficient to start dropping jQuery.

Make Your Web App Accessible With the Axe Accessibility Library

Axe is an open-source project.
It returns zero false positives (bugs notwithstanding).
It’s designed to work on all modern browsers and with whatever tools, frameworks, libraries, and environments you use today.
It’s actively supported by Deque Systems, a major accessibility vendor.
It integrates with your existing functional/acceptance automated tests.
It automatically determines which rules to run based on the evaluation context.
Axe supports in-memory fixtures, static fixtures, integration tests, and iframes of infinite depth.
Axe is highly configurable.