ilyaeck

11 ms·

61.
by ilyaeck 10y ago
The hype is likely largely due to AlphaGo. It was a big win and used RL ==> RL must be a silver bullet! But more generally, when you don't have enough training data ahead of time but do have the benefit of lots of user interactions,
62.
by ilyaeck 10y ago
Give it some time. ML improves with usage and training data.
63.
by ilyaeck 10y ago
Technically, this could/should have been already solved with voice authentication. Nuance and other companies have been claiming/pushing it for a few years now: http://www.nuance.com/for-business/customer-serv
64.
Bugsee is now the best crash reporting tool for iOS, according to Microsoft(bugsee.com)
3 points by ilyaeck 10y ago|0 comments
65.
by ilyaeck 10y ago
How surprising and refreshing it is to find an essay about Pushkin on HN!
66.
Natural language neural networks: how deep is too deep?(ilyaeck.github.io)
4 points by ilyaeck 10y ago|1 comments
67.
Snapchat acquires Vurb for $110 million(venturebeat.com)
5 points by ilyaeck 10y ago|2 comments
68.
by ilyaeck 10y ago
Question to Y-hat folks: why cluster in 2D? Granted, clustering in 300D is hard :) Still, the 2D projection must add a significant metric distortion. Why not a middle ground, say, 5-10D ?
69.
How to shut down a startup in 36 hours(entrepreneurs.maqtoob.com)
5 points by ilyaeck 10y ago|0 comments
70.
Volkswagen's U.S. diesel emissions settlement to cost $15B(reuters.com)
292 points by ilyaeck 10y ago|339 comments
71.
Ghostbot pretends to be you, talking to people you want to avoid(theverge.com)
7 points by ilyaeck 10y ago|3 comments
72.
by ilyaeck 10y ago
They didn't throw "millions" at the idea. If you read the article carefully, they stopped after spending $1M, i.e., with 2/3 of their investors' money still in the bank. Yes, maybe they could've stopped/pi
73.
by ilyaeck 10y ago
In fact, you have both sides of the table to satisfy, and therein lies one of the difficulties.
74.
by ilyaeck 10y ago
Very good read, but at the end, there is no universal recipe for either success or failure.
75.
Gmail 'Smart Reply': the paper [pdf](kdd.org)
4 points by ilyaeck 10y ago|3 comments
76.
by ilyaeck 10y ago
Interesting results - it's a pity there is no theoretical justification whatsoever, like with most of the deep-nets papers.
77.
AI Resources(airesources.org)
22 points by ilyaeck 10y ago|2 comments
78.
by ilyaeck 10y ago
Are they actually open-sourcing it? There doesn't seem to be link to the code in sight.
79.
by ilyaeck 10y ago
There is value and demand for both approaches. For the foreseeable futures, we are going to need lots of applied mathematicians to figure out computational models that will eventually put them out of business (same as programmers ;)) Someda
80.
by ilyaeck 10y ago
Here is the direct link (paywalled, see summary below): https://www.theinformation.com/apple-opening-siri-developing...
81.
by ilyaeck 10y ago
Summary: 1. Enable Siri-2-app integration (likely similar to Android's intent URIs): "To work with Siri, the third-party apps will need some kind of search-query box that Siri could fill with the user’s verbal request once the app
82.
Apple to release Siri SDK(go.theinformation.com)
7 points by ilyaeck 10y ago|5 comments
83.
by ilyaeck 10y ago
It looks like Apple's plan is to enable something similar to Android's intent URIs: "To work with Siri, the third-party apps will need some kind of search-query box that Siri could fill with the user’s verbal request once the
84.
by ilyaeck 10y ago
It is not misleading at all, that is the project.
85.
by ilyaeck 10y ago
OpenAI Gym is for Reinforcement Learning - a different kind of learning, where you don't have ground truth, but the agent gets a positive reward when it makes good guesses. Is this what you are looking for with NLP?
86.
by ilyaeck 10y ago
Do you know what kind of classification method it is using at the moment?
87.
by ilyaeck 10y ago
SyntaxNet is, by definition, for syntactic analysis - it would likely not help you much with semantics, to extract meaning. It could maybe help you automatically determine is a sentence is grammatically correct, though.
88.
TensorFlow: the good, the bad and the ugly(indico.io)
3 points by ilyaeck 10y ago|0 comments
89.
by ilyaeck 10y ago
With just 72 hours of training data, this is an extremely impressive result, scientifically speaking. However, in terms of quality control/best practices in automotive, it's simply unacceptable to trust human lives with a monolith
90.
Victory Day anniversary: WWII diaries of Soviet children translated to English(ibtimes.co.uk)
2 points by ilyaeck 10y ago|0 comments

More ›