Ask HN: machine learning success stories?

33 points by cageface ↗ HN
There seems to be a bit of a buzz around machine learning these days. The combination of cheap compute clusters and lots of easy available, potentially mineable data from social networking, e-commerce etc seems to present some juicy new opportunities for these techniques.

So, are there some good examples of successes with this kind of approach lately or is this the overpromise and underdeliver of AI all over again?

20 comments

[ 3.5 ms ] story [ 56.6 ms ] thread
(comment deleted)
(comment deleted)
At least basic machine learning seems to have become ingrained enough in a lot of areas that it's not really a separate thing anymore, just part of how people do things, which I suppose is one kind of success. E.g. lots of business-analytics departments will use things that at least 10 or 20 years ago would've been considered "machine learning", but now are just part of how you analyze and visualize data.
Though techmeme is guided a by a human hand, I'm pretty sure much of its power and intelligence is via machine learning (for determining related stories, and their basic order).
Some fairly well-known consumer-facing things that use machine learning are spam filters, recommendation engines, speech recognition systems (speech-to-text or customer service stuff), internet advertising, news clustering (Google News), related stories, handwriting recognition, questionable content identification, automatic closed captioning, and machine translation. These are not all equally successful or sophisticated, but are ML-based and mostly good enough to use.

There are a number of examples that are not consumer-facing, like credit card fraud detection, snail mail routing, quantitative trading, market segmentation analysis, demand prediction for inventory control, and other things. It is also used for scientific data analysis in several areas, with bioinformatics being the really big one. There are other examples.

There are also applications that are not considered machine learning, but use the same ideas for different purposes. An example would be modern codes, which are used for things like compression and satellite communications, and are based on the same `graphical models' pervasive in machine learning.

There is hype, and some applications need only a little bit, but it is at least used in some real stuff.

Handwriting detectors implemented by the Postal Service. My ML professor worked on algorithms with USPS for over 20 years and gave our class a sub-problem for a term project. He once told us a story that the computer scientists were all up in arms because they wanted the post office to design letter boxes on envelopes to encourage people to write their characters in legible english. Obviously that didn't happen and thus it took quite a bit longer to design software that would reliably recognize v's from u's and b's from 6's and r's from t's and c's from a's under the ridiculous constraints that sorting millions of packages of mail requires.
Bayesian spam filtering - machine learning 101
Also clustering algorithms (k-means et al) used to do a preliminary analysis of NMRI scans
IF you round ML into "computerized statistics" you'll find endless success stories in scientific research. Particularly interesting might be f/MRI image analysis
I think there is a bit of a hype, and people specifically try to deemphasize any relation to AI promises. IMHO, many things that were once associated with statistics are now rather referred to as machine learning, and the range of applications tends to become very broad.

One could refer to the unsupervised adaptation of new HMM-based speech synthesis (HTS) voices as a promising machine learning application. It's not a success story, but imagine being able to create new synthesis voices given only a few minutes of speech by any person.

Want to try out how well that works? Go to [1] and select the voice "GWB (HTS 2007)", and enter any english text you want. Sounds familiar? And that's just an academic demo page...

[1] http://homepages.inf.ed.ac.uk/jyamagis/demos/page35/page35.h...

[disclaimer: I'm an optimist.]

I personally think that ML is on the verge of some major breakthroughs.

In particular, I think that results in "deep learning" are very promising. I've written about this approach in earlier HN comments.

"Deep learning" is the new big trend in Machine Learning. It promises general, powerful, and fast machine learning, moving us one step closer to AI. Deep learning has already made important advances in achieving state-of-the-art accuracy in vision and language, but with much less manual engineering that competing methods.

In fact, I think the major success of the deep learning movement has been to get the community to start focusing on figuring out how to get powerful learning algorithms actually to work. A lot of people used to (but many still do, sadly) work on making incremental improvements on learning algorithms that are implausibly simple. "Sure, we know this model can't achieve human level performance on vision or language or control (robotics) or planning, but with this neat refinement I can get a paper out of it." Deep learning begins its endeavor with the goal of AI, and rejects techniques whose upper bound isn't high enough. Simply the fact that the community is setting its sights high---not in terms of over-promising to the outside world, but merely in terms of the learning machinery being explored---and is actually trying to achieve AI is a step forward.

An algorithm is deep if the input is passed through several non-linearities before being output. Most modern learning algorithms (including decision trees and SVMs and naive bayes) are "shallow".

For intuition, imagine if I told you that your main routine can call subroutines, and your subroutines could call subsubroutines, but you couldn't have any more abstraction than that. You can't have subsubsubroutines in your "shallow" program. You could compute whatever you wanted in a "shallow" program, but your code would involve a lot of duplicated code and would not be as compact as it should be. Similarly, a shallow machine learning architecture would involve a lot of duplication of effort to express things that a deep architecture could more compactly. The point being, a deep architecture can more gracefully reuse previous computations.

Deep learning is motivated by intuition, theoretical arguments from circuit theory, empirical results, and current knowledge of neuroscience. Here is a video where I give a high-level talk on deep learning, and describe this intuition: http://bilconference.com/videos/deep-learning-artificial-int....

Here is more detailed information about deep learning: http://deeplearning.net/tutorial/

Another aside: My colleague Hoifung Poon published exciting work in semantic parsing. It received best paper award at ACL 2009, the most prestigious NLP conference. (http://www.cs.washington.edu/homes/hoifung/papers/poon09.pdf) You read it and you're like: "Really? You're doing that? You're actually trying to solve NLP using purely automatic techniques. Whoa. I forget that was the goal, I was too busy doing feature engineering!"

He achieves impressive results on question-answering, and beats other systems in recall, giving answers to many more questions, at the same level of accuracy at the competing methods.

The source code for his semantic parser is available (http://alchemy.cs.washington.edu/papers/poon09/) and you can use it to build a Q+A system. You can try a demo of it here, which I put up: http://bravura.webfactional.com/ He is about to talk about an updated version of this work, in which he induces ontologies purely automatically.

Thanks for the leads. Looks like some interesting reading.
The best case study I've ever heard is Tescos. Apparently they were a downmarket British retailer in the early 1990s. They teamed up wit a statistical consultancy called dunnhumby and now they're the third biggest retailer in the world.

Apparently one per cent of marketing mail is opened, while 98 per cent of Tesco mailouts are opened (because they offer people targetted discounts). No surprise large retailers and grocery chains are rushing the adding significant ML capabilities.