I interned at Numenta [0] in 2012. Numenta is building an open-source machine intelligence product [1] based on the human brain. Specifically, the algorithms are based on the theory of "hierarchical temporal memory" (HTM) [2] as described by Numenta founder Jeff Hawkins [3] in his book On Intelligence [4]. The basic idea is that the neocortex has a generalized learning framework that acts on generalized input from al…
You're aware that numenta is considered a joke in the serious machine learning community, right?
I don't think your second point follows from your first. ML is not easy to set up or maintain, even if we're just talking about linear regression. Companies trying to make 'data scientists obsolete', usually frame ML as a black box that you can plug data in one end and spit out profits the other. But in reality it doesn't work like that, significant effort has to build and maintain data pipelines, to make sure data q…
It can be easier to set up and maintain than hand-rolled solutions. Best example I can think of is speech recognition - previous systems used hidden markov models, gaussian mixtures, triphones, and all sorts of complex and obscure things that had to be tuned by experts. Now they just use an end-to-end neural network that goes directly from sound to letters (almost anyway; I believe the input is still MFCCs). I agree…
> Now they just use an end-to-end neural network
You've missed my point. The thing is that building the ML model is NOT the hard part of machine learning in industry. The hard part is building an infrastructure that can make that machine learning model do something useful. It is much harder than people imagine. See for example this great paper by google for more details : https://static.googleusercontent.com/media/research.google.c...
The DeepMind datacenter project was very interesting, but a lot of the ML people I spoke to were quite dubious about how much of it was genuinely down to new AI/neural networks, and how much of it was Google PR to justify how much they spent on DeepMind. > DeepMind trained a neural network to more accurately predict future cooling requirements, in turn reducing the power usage of the cooling system by 40 percent. But…
I was going to say, I feel like this could be done with excel and some historical data combined with weather forecasting. that being said, even if ML gets you a 0.5% advantage over spreadsheet math, that is a non-trivial amount of savings on a national scale.
What I'm wondering is why I can't find anything about ML applied to finding and curating data, which is the most tedious part of data science. That would be an interesting way of using ML without fuzzy stuff.
I interned at Numenta [0] in 2012. Numenta is building an open-source machine intelligence product [1] based on the human brain. Specifically, the algorithms are based on the theory of "hierarchical temporal memory" (HTM) [2] as described by Numenta founder Jeff Hawkins [3] in his book On Intelligence [4]. The basic idea is that the neocortex has a generalized learning framework that acts on generalized input from al…
You're aware that numenta is considered a joke in the serious machine learning community, right?
Yes, and quite unjustifiably so in my opinion, which is why I sourced all those links in my comment. From what I've seen, Numenta is very good at a specific class of problems (time series prediction and anomaly detection).
Feel free to underestimate them. It's really not my problem.
And do pompous, abrasive comments like yours really add anything to the discussion? Why don't you elaborate on why they are a "joke" to serious intellectuals like yourself.
I was going to say, I feel like this could be done with excel and some historical data combined with weather forecasting. that being said, even if ML gets you a 0.5% advantage over spreadsheet math, that is a non-trivial amount of savings on a national scale.
What I'm wondering is why I can't find anything about ML applied to finding and curating data, which is the most tedious part of data science. That would be an interesting way of using ML without fuzzy stuff.
There are a few current projects, particularly in database research. I don't know how many of them use ML in the traditional understanding. Current projects I know of are Wrangler, Mimir, Katara, MayBMS (in no particular order).
I suppose you are joking, as double-glazing seems to be a popular meme in the UK (with people being convinced they don't need it and scammy vendors pushing it)? Must admit I never completely understood it, coming from a country where double glazing is the norm.
Can't understand your comment. You are agreeing with me that, in most other parts of the world, double-glazing (aka a glass that won't break if you elbow it by mistake!) is the norm. But in the UK, it's sadly the opposite. So we agree that they should replace their crappy windows with double-glazed, properly insulated, ones...
>> But in the UK, it's sadly the opposite.
Any data to back this up? Personally I find it a pretty rare sight to see windows that aren't double glazed and if I came across a property without them it's absurd enough I wouldn't move into it.
I know a couple people working on using deep learning for financial timeseries. It can be beneficial even with relatively limited data (a couple of years with high granularity). I also think the 10% figure is quite ambitious; especially given that they are at an early stage of negotiations.
Yeh I'd guess electronic exchange financial time series have a lot of interesting dynamics at very fine detail because of high frequency trading, so it's a very different system than a power grid in that respect. And for the record I regard HFT as mostly being a wasteful activity that effectively amounts to a tax on using electronic exchanges - ok there's a fuzzy line between arbitrage (useful) and full on HFT (which…
I'm involved in very-unsexy not-Google-level smart grid research using machine learning. A Phasor Measurement Unit (PMU) like those being deployed in America takes sixty measurements a second for each signal, and this is in no way overkill for the granularity of information required.
I don't know much about electric grid engineering - are there opportunities for a ML approach to increase efficiency in ways other than the sort of better supply forecasting implied by this article. The article does mention the losses involved in long distance transmission, but surely traditional approaches can already yield fairly well optimised planning for improving this sort of efficiency? (Finally, this article…
(copy-pasted disclaimer: I'm involved at a junior level in very-unsexy not-Google-level smart grid research using machine learning, and my power grid engineering knowledge is painfully limited.) Predicting demand is dicey dicey stuff, but throwing sensors on the grid gives a lot of information that can imply other factors. Just pulling this out of the air for an example, but: UK power is famous for having to deal with a sharp spike in demand when a big TV event is going on and then goes to commercial, because everyone in the country in synchronized fashion gets up, goes to the kitchen, and turns on the kettle for tea. (Really, it's a big deal!) Now, that's not the best example because of the dramatic nature of the spike, but you can see how power information might on some level reflect the state just before that spike: people aren't moving around their homes, vacuuming, w/e, they're in front of the TV, right? So our demand forecaster learning from the data might not be able to tell that a new season of Sherlock is airing, but it might learn enough paranoia about everyone-watching-TV-at-once patterns to be useful.
It can be easier to set up and maintain than hand-rolled solutions. Best example I can think of is speech recognition - previous systems used hidden markov models, gaussian mixtures, triphones, and all sorts of complex and obscure things that had to be tuned by experts. Now they just use an end-to-end neural network that goes directly from sound to letters (almost anyway; I believe the input is still MFCCs). I agree…
> Now they just use an end-to-end neural network You've missed my point. The thing is that building the ML model is NOT the hard part of machine learning in industry. The hard part is building an infrastructure that can make that machine learning model do something useful. It is much harder than people imagine. See for example this great paper by google for more details : https://static.googleusercontent.com/media/re…
I don't think either of you are strictly wrong, though--the comment's point seems to be that doing things by hand was so much harder that it couldn't even be done to the same level of success that machine learning achieves.