mrdrozdov

David Rosenberg's Machine Learning course is an excellent intro. Will give you the foundations that you need for anything else. Has a few links for additional resources, but the slides are mostly sufficient. https://davidrosenberg.github.io/ml2015

Hi Michael,

What advice do you have for some who is happy with their current trajectory, but might want to start their own business in say 10 years?

Thanks!

This would fit so well into George Carlin's Euphemisms bit:

https://www.youtube.com/watch?v=vuEQixrBKCc

I'd go ahead and say that a lot of ambiguous news will not be marked anyway. Although any news that says I've made alien babies should obviously be marked disputed.

What are the two sides? Why should they have a say? Some information can be definitely marked as False without taking opinion into account.

I have three feet. <= False

Yeah, I mean being able to find the origin of information for any piece of text online is obviously better, but that's hard.

Marking articles which are clearly incorrect is probably a good start.

The sky is checkered like a chess board if you are in the right part of Minnesota on a Monday evening. <= Fake News

The court.

What do you mean by source?

Considering that anyone can publish anything online, and then collect money for advertising dollars on that topic, this seems like a very reasonable feature regardless of the political climate.

Spam detectors are clearly also political in nature. And I know some Nigerian princes that you really need to meet.

I'd expect they have a reasonably effective approach for A/B testing. Essentially, you want a uniformly random sample of your users, but you also want to enforce some properties about "connectedness" of the users in the sample, since user-user interaction might be an interesting metric to track. I think the buzzwords related to this sort of thing is "causal inference."

Based on the performance of some sample, you might decide to tweak the feature and release it to a new sample. The buzzwords for this would be "bandit algorithms."

The Onion is fake news, so it probably should be disputed, especially considering there are many instances of people considering it as real news.

Wouldn't this logic apply to their previous reporting feature?

It's really a great approach that they're taking. I do think classification of "True" is very difficult, although "Disputed" seems reasonable. For instance, if I wrote an article saying I was a billionaire, I'd assume this would eventually receive the "Disputed" label. Articles not marked as "Disputed" are clearly not necessarily "True".

Anyone using this yet to automatically track SotA results on machine learning tasks?

Only place I've seen them used has been to annotate the notes of a deep learning course at Stanford. http://cs231n.github.io/

How do you know it's from Uber? Don't they have a partnership program where you can collect some $ from convincing new users to join? Would be easy for anyone to sign up, get a bunch of a coupon codes, then post these reviews.

To clarify, given the net is not neutral, Facebook will be able to provide better services because they pay internet providers some fee for extra bandwidth and reduced latency. While Snapchat would decide to provide worse services at a more reasonable cost to them.

This helps Facebook in the battle against Snapchat, but hurts Facebook in that they have to pay an extra fee. A similar argument could be made with other tech companies and their rivals.

Is this a correct interpretation?

Would this only effect up and coming telecom firms and high bandwidth services (video streaming, audio streaming, gaming, etc.), or also your every day SaaS startup?

Edit: how exactly would this have a negative impact on SnapChat?

Double edit: I think net neutrality is a good thing, and worth finding clear arguments to fight for.

Triple edit: Reasons I believe the net should be neutral...

1. Consumers should be able to use any internet service they choose.

2. Consumers should pay the same for data usage regardless of what the service is.

3. Businesses should be able to provide their service for the same data usage cost as any other business.

This seems like a specific case of Curriculum Learning from the machine learning literature.

What is the content page?

Thanks. According to Snopes, the story is False [1]. It's pretty ridiculous that "credible" news sources can write a story like this and then take no responsibility when it's wrong. I don't think it's enough to just correct a story after the fact. There should be a fine or search engine punishment, or something along these lines.

For instance, I read this story earlier on Variety and it was presented as fact. If you go back to the same url [2] the story and headline have completely changed.

Previous Headline: Cable Operator Snafu in Boston Leads to Porn Airing on CNN...

Current Headline: CNN, RCN Deny Reports of Porn Airing on Channel in Boston

[1]: http://www.snopes.com/cnn-half-hour-porn/ [2]: http://variety.com/2016/legit/news/cnn-porn-boston-rcn-cable...

Unclear. Was the story accurate?

> Eventually the scientists went so far as to develop an in-house programming language for their models rather than settle for a numbercentric option such as ASCII, which was popular at the time.

Did not realize you could use MXNet declaratively (like Tensorflow/Theano) and imperatively (like Torch/Chainer). Can anyone speak more of their imperative usage of MXNet?

CNTK from Microsoft Research has been rebranded as Microsoft Cognitive Toolkit.

Here's the github repo: https://github.com/Microsoft/CNTK

CNTK homepage (http://www.cntk.ai/) now redirects to https://www.microsoft.com/en-us/research/product/cognitive-t...

> CNTK (http://www.cntk.ai/), the Computational Network Toolkit by Microsoft Research, is a unified deep-learning toolkit that describes neural networks as a series of computational steps via a directed graph. In this directed graph, leaf nodes represent input values or network parameters, while other nodes represent matrix operations upon their inputs. CNTK allows to easily realize and combine popular model types such as feed-forward DNNs, convolutional nets (CNNs), and recurrent networks (RNNs/LSTMs). It implements stochastic gradient descent (SGD, error backpropagation) learning with automatic differentiation and parallelization across multiple GPUs and servers. CNTK has been available under an open-source license since April 2015. It is our hope that the community will take advantage of CNTK to share ideas more quickly through the exchange of open source working code.

Occam's razor implies that the simplest explanation is usually the correct one.

I wonder how much Trump could get paid to Tweet.

> The rewards to attracting the right people in the corporate world can more than justify a six-figure speech. Hedge funds often invite potential clients to events featuring prominent speakers. As one hedge fund manager has explained, if just one client “decides to invest $10 million… the firm will snag a 2 percent management fee—which works out to $200,000” per year.

This on the other hand seems like a very legitimate explanation and parallels the logic behind paying speakers to attend large tech events (like one Salesforce might have, which is mentioned elsewhere in the article).