Stories
Slash Boxes
Comments

SoylentNews is people

posted by janrinok on Monday June 29 2015, @03:17PM   Printer-friendly
from the I'm-glad-you-asked dept.

I'm a neuroscientist in a doctoral program but I have a growing interest in deep learning methods (e.g., http://deeplearning.net/ ). As a neuroscientist using MR imaging methods, I often rely on tools to help me classify and define brain structures and functional activations. Some of the most advanced tools for image segmentation are being innovated using magical-sounding terms like Adaboosted weak-learners, auto-encoders, Support Vector Machines, and the like.

While I do not have the time to become a computer-science expert in artificial intelligence methods, I would like to establish a basic skill level in the application of some of these methods. Soylenters, "Do I need to know the mathematical foundation of these methods intimately to be able to employ them effectively or intelligently?" and "What would be a good way of becoming more familiar with these methods, given my circumstances?"


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 0) by Anonymous Coward on Tuesday June 30 2015, @01:41AM

    by Anonymous Coward on Tuesday June 30 2015, @01:41AM (#203128)

    Yes! Yes, to use any tool, you must understand what it is doing.

    Now, you don't need to know the intimate details of how a particular library has optimized a technique.

    "How To Lie With Statistics" is famous but there's no equivalent for machine learning. The idea behind it, though, that by misusing techniques one can claim almost anything, absolutely 100% holds with machine learning. I've seen people repartition datasets into training/testing, over and over, until the train and test groups gave the desired outcome...

    Don't be one of those people! Understand the ideas behind what's going on.

    You can swing by any university bookstore and check the titles that their machine learning classes use, if you don't get a title you like here.

  • (Score: 0) by Anonymous Coward on Tuesday June 30 2015, @07:20AM

    by Anonymous Coward on Tuesday June 30 2015, @07:20AM (#203215)

    There is a difference between knowing and checking assumptions and knowing proofs of a method.

  • (Score: 3, Informative) by fritsd on Tuesday June 30 2015, @08:25AM

    by fritsd (4586) on Tuesday June 30 2015, @08:25AM (#203241) Journal

    I've seen people repartition datasets into training/testing, over and over, until the train and test groups gave the desired outcome...

    Then they're doing it wrong :-)

    Cross-validation [wikipedia.org] is a very useful technique to repartition your dataset into training and testing sets, but the idea is to prove a more robust result, i.e. "is your predicted result really as good as you found". In other words, it's meant to make your result *worse* but a bit more certain. (This is also in the Hastie & Tibshirani book, half of chapter 7 is about "does cross-validation really work and how should you do it")

    I had to LOL at this XKCD cartoon: https://xkcd.com/882/ (significant) [xkcd.com] which shows exactly what you get if you're using it wrong.

    • (Score: 0) by Anonymous Coward on Tuesday June 30 2015, @09:37PM

      by Anonymous Coward on Tuesday June 30 2015, @09:37PM (#203517)

      Cross Validation is not (at least on its face) a hard concept. I've used it, and used it correctly in an imaging methods paper. Obviously test data and training data need to be separated within each cross-validation experiment.