Dear friends, hello! I have seen some Matlab examples of Deep Learning, but unfortunately didn't understand them clear. Please, answer for my questions: 1. Deep Learning is some Generic Approach or it is applicable only for Neural Network Method ? I have seen examples only according Neural Network Method. Is it possible to use Deep Learning also for Logistic Regression, SVM and other Machine Learning methods? 2. Please, sent me (or give me links) some tutorials about Deep Learning, especially for Deep Learning for Logistic Regression and SVM. 3. Is Deep Learning really applicable only for Image Processing/Recognition problems, or it is also applicable for other Data Mining tasks (e.g., Text Mining, Data Classification/Diagnostics, etc.)
Thanks in advance for your answers.
Dear Alex Yiolkin,
Look the link, may be useful.
https://www.oreilly.com/ideas/what-is-deep-learning
Regards, Shafagat
Dear Shafagat, thank you very much for your fast and useful answer. I carefully read attached article. Is it right, that I have understood (?) :
1. Deep Learning is applicable only for Neural Network
2. Deep Learning is applicable for problems with very-very high dimension (Very Large vocabulary on the text mining task, million pixels on the image recognition task,...) and a lot of objects (millions documents, images,...)
3. For regular tasks (e.g., with 10...100 parameters) isn't necessary to use Deep Learning.
What is your opinion? Thanks in advance for your answer.
Dear Alex Yiolkin,
I found this on line book very useful:
http://neuralnetworksanddeeplearning.com
In short deep learning is a form of neural network architecture with many hidden layers. Currently various deep learning implementations have the highest accuracy for unstructured data sets such as image classification or language processing.
However Neural Networks are hard to tune, and are very intimidating for novices in ML.
some useful alternatives in my opinion are:
boosting techniques - xgboost (http://xgboost.readthedocs.io/en/latest/model.html) is a powerful implementation. From my experience it performs better compared to elastic nets (glmnet - https://cran.r-project.org/web/packages/glmnet/index.html), random forests (randomForest - https://cran.r-project.org/web/packages/randomForest/randomForest.pdf), or SVM (liquidSVM - https://cran.r-project.org/web/packages/liquidSVM/index.html, e1071 - https://cran.r-project.org/web/packages/e1071/vignettes/svmdoc.pdf), at least on the data sets I tested. I am providing links for R packages since I am not a MATLAB user.
It all depends on the time you are willing to spend solving the problem - understanding the algorithm, the tuning process, validating the results.
elastic nets work out of the box, and are quite fast, random forests also. SVMs have usually 1-3 hyper-parameters and several kernels to try out. Xgboost has a lot of parameters but if you do not have time to fiddle with them it performs very decent out of the box..
Regards,
Milan
Dear Milan, thank you very much for your answer. I will carefully study it. Regards, Alex.
Hi! My answers are below. Regards, Sergey.
1. Deep Learning is advanced method of NN (rather it is based on NN) so it isn't applicable for SVM, LogRegr, etc.
2. There are a lot of tutorials for Deep Learning, please use Google
3. Generally speaking, it is possible to use Deep Learning for regular tasks, main advantage of Deep Learning is to solve Large Dimension tasks.
Dear Sergey, thank you very much for your answer. I will carefully study it. Regards, Alex.
- Badges
- Science topic