relate

FYI Here is an overview of all of his columns:

https://www.bloomberg.com/opinion/authors/ARbTQlRLRjE/matthe...

Here is an example of a demo we released that uses a tooltip to make it more obvious how it works:

https://hific.github.io/

You could allow for errors and use a bloom filter to avoid the space issue:

https://en.wikipedia.org/wiki/Bloom_filter

Would you prefer 11111+11111+33333 instead?

I've done whiteboard only interviews, as well as shared doc/coderpad only, and I find both limiting. When thinking about the problem, it's great to have a whiteboard to sketch examples etc, but writing code is much easier with a text editor.

Recently I interviewed at Google, and they had no problem fulfilling my request to do both: sketch the solution approach on whiteboard and then write the code with a laptop.

and whether you are willing to work 106 (40+56) hours a week...

I find it strange you mention transfer learning, since one of the reasons neural networks are so popular are because they tend to excel at it. Adapting (i.e. fine-tuning) networks trained on a task with a lot of data (e.g. image classification on ImageNet) to different tasks, such as image segmentation, has proven a very successful approach.

Could't they open source the design/implementation without a permissive licence? I.e. everyone could audit, but nobody could use it commercially.

In your example, this would correspond to having the number x = 0 +-1 and then wanting to compute 1/x. If your number can potentially be zero, why would you want to use it as a divisor?

I'm not sure what you are implying. Iceland suffered a hit in the 2008 crash, but has since almost fully recovered.

GDP Iceland over time (2017 missing): https://data.worldbank.org/indicator/NY.GDP.PCAP.CD?location...

GDP per capita comparison (Iceland ranking 6th): https://data.worldbank.org/indicator/NY.GDP.PCAP.CD?year_hig...

Hi, I'm one of the authors. That is correct in a strict sense, but we wanted to focus on the more 'creative' part of logos rather than the text. GANs are known to struggle with high resolutions, but we note that we show higher resolution logos later in the paper (see page 12 and 15 e.g) which is trained on the smaller but higher-resolution version of our dataset.

Suppose I have the set F = { all integers x where x can be written with a finite number of digits }

Are you saying the set F is finite?

Or are you saying that the set F contains transfinite numbers?

You Ain't Gonna Need It. (I think)

Shameless plug: we have a workshop today @ CVPR 2017 that explores the classification task of ImageNet using crawled (and noisy) data: https://www.vision.ee.ethz.ch/webvision/workshop.html

In contrast to ImageNet, only the validation and test data were manually labeled. Impressively, on such data, an accuracy of ~95% (top5) is achieved by the winning solution this year.

Interesting! I am wondering if it is easy to incorporate the visualizations of the models during training? If not, is this something on the roadmap?

Another issue is that browsers do not display the non-secure http:// prefix in the url bar (which should probably be red and striked through).

As a PoC, I bought the domain https.is, and now I can construct urls like https.is//accounts.google.com - which can look convincing when glimpsed over.

Well, if a large corporation infringes on your patent, in theory your patent is now valuable to its competitors as a "weapon" for their "patent arsenal".

What if someone edits an answer posted before you made it?

1. Set a minimum time until next stop can be announced 2. Add a button to skip a stop

I would have thought the same. However instead of adding random latency, one could have a buffer, that only allows a couple of fixed output rates (eg, 150,160,...,300 CPM), when typing. This would probably be quite annoying to type on though :-)

Thanks :-)

Yeah, we are not accounting for connection times right now. However, we have the data to make it more accurate, just needs a bit more coding ;)

Ideally I'd also want also to have cycling+train - since that is also a feasible option for some people. E.g. I live 1km away from a train station that is well connected to my workplace, so I can bike to the train station and commute the rest.

Thanks :-) The stack is based on:

* A Javascript based SPA with handlebars for template and leaflet.js for map

* Django + Postgres for the API

* Simple Vertex-Edge model of the train system, with shortest-path to estimate travel times (including walking distance)

-This is on our to-do list to improve, but we do this since querying all pairwise travel times from SBB/google would be too heavy.

* Numpy/Scipy for the machine learning

I'm one of the creators. This was built in 40 hours at Hackzurich 2015 (www.hackzurich.com) and combines the following:

    * The largest rent database of Switzerland 
    * A model of the train system of Switzerland
    * A simple recommender system
    * Google maps for name -> geolocation
    * OpenStreetMap for the map

There are some heavy calculations behind each query, so it takes about 15 sec to load. Furthermore, you need to like at least 3 apartments and dislike at least 3, before the system starts recommending.

Except that increased money supply means inflation which means that middle & upper class can buy less than before. E.g. someone with a medium salary might only get 20% higher income post-stipend but prices would rise more.

You took a poor example since pi is not known to be a normal number (see eg http://mathworld.wolfram.com/NormalNumber.html )

This number however is:

    0.0 1 10 11 100 101 110 111 1000 ...

This number contains all movies that have and will ever exist. However to specify where a 700mb harry potter movie is in this number, you'll need at least 700mb to represent the index. So in some sense harry potter 'exists' in this number - but in another sense its just a silly encoding method.

The comment I replied to mentioned a jumping whale, not exactly a still scene...

Well, if you have 16 quadcopters - each with one camera... :-)

This is a common criticism. However, almost all ML methods have some built in heuristic choices, that are the result of finding something that both works and is mathematically nice. Each of these choices restricts us to some family of functions where it's hard to justify why it's really relevant to the problem at hand, e.g. convex loss functions (l1, l2, ..), convex regularizers (l1,l2,..), gaussian priors, linear classifiers, some mathematically nice kernel functions, e.t.c. In the end, people usually statistically estimate the performance of the methods and use what works.