Comment by aaron-lebo | Hacker News Reader

Comment by aaron-lebo | Hacker News Reader

For a recent practical example of their usefulness:

This paper presents the Militarized Interstate Dispute (MID) 4.0 research design for updating the database from 2002-2010. By using global search parameters and fifteen international news sources, we collected a set of over 1.74 million documents from LexisNexis. Care was taken to create an all-inclusive set of search parameters as well as a sufficient and unbiased list of news sources. We classify these documents with two types of support vector machines (SVMs). Using inductive SVMs and a single training set, we remove 90.2% of documents from our initial set. Then, using year-specific training sets and transductive SVMs, we further reduce the number of human-coded stories by an additional 21.6%. The resulting classifications contain anywhere from 10,215 to 19,834 documents per year.

http://steventlandis.weebly.com/uploads/1/2/1/4/12144932/dor...

5 years isn't a recent example. my impression is that deep nets ate everyone's lunch (including svm).

Not at all, deep nets are difficult to train and they need lot's of processing before they learn same kind of classifying features than ie. SVM has. So yes you can simulate SVM with deep networks but usually it's not very good solution.

SVM can be also used as part of the neural network such as in classifying layer

Deep nets ate everyone's hype. The lunch is still there. SVMs have many advantages over ANNs that recommend themselves to practical applications still.

What advantages do SVMs have at this stage?

training speed, training data requirements, and runtime resource requirements are the big ones. And on domains that are not raw image processing, SVMs are still often quite competitive when it comes to accuracy.

They really aren't appropriate for the same set of problems, even if there is some overlap.

There is a lot of buzz around deep learning right now, but the SNR isn't great.

On the research frontier, this is pretty much true. But, for what it's worth, I'm spearheading some machine learning efforts at my current company, and most of my initial production models have not been deep networks but rather classical approaches like boosted trees or SVMs. Actually, gradient boosted trees in particular are one of the most powerful general-purpose models out there, and there are some really fantastic distributed implementations available now that routinely win Kaggle competitions (xgboost, Spark MLLib's version).

I will note though that the problems I'm tackling do not involve any image processing or recognition. Convolutional networks really have completely dominated that area both in research and practice in the last few years.

Reply on news.ycombinator.com