Deep Learning Frameworks

There are a lot of deep learning frameworks out there, depending on your usage purpose or the familiarity of your programming languages or working tasks, here I only summarize ones that I am familiar with:

1) TensorFlow by Google (released on 10 Nov 2015): http://tensorflow.org/

Comment: 

2) VELES by Samsung (released on 11 Nov 2015): https://velesnet.ml/

Comment:

3) cnn (lightweight and very fast neural network library in C++, also in Python, works both on Windows and Linux machines): https://github.com/kaishengyao/cnn

Comment: cnn has been proven to be much faster than Theano both with and without GPU. Also, it offers the advantage for software production of neural network models since it has been developing in C++ and more importantly, it supports both Windows and Linux platforms. 

4) to be updated

Torch vs. Theano vs. Caffe

Link: http://fastml.com/torch-vs-theano/

Here is my summary:

- Torch and Theano are better to be used for research purpose on deep learning (DL) whereas Caffe is more scaled for DL application development.

- Torch and Theano are competitive in terms of speech and performance via different benchmarks. Hence, choosing one of them depends the ease of use from users.

(to be updated)

Representation Learning

This research topic is very important in machine learning and a first step for all kinds of machine learning algorithms. Having a robust representation definitely plays a vital role for good performance/accuracy.

Recent progress of representation learning is using deep architecture (normally referring to deep learning). Here I collect some of very good papers worth reading:

1) Representation Learning: A Review and New Perspectives (Yoshua Bengio et al., 2014)

2) Deep Learning of Representations for Unsupervised and Transfer Learning (Yoshua Bengio et al., 2012)

3) ...

Deep Learning for NLP

1) CSLM: Continuous Space Language Model toolkit
Link: http://www-lium.univ-lemans.fr/cslm/

Intro: CSLM toolkit is open-source software which implements the so-called continuous space language model.

The basic idea of this approach is to project the word indices onto a continuous space and to use a probability estimator operating on this space. Since the resulting probability functions are smooth functions of the word representation, better generalization to unknown events can be expected. A neural network can be used to simultaneously learn the projection of the words onto the continuous space and to estimate the n-gram probabilities. This is still a n-gram approach, but the LM probabilities are interpolated for any possible context of length n-1 instead of backing-off to shorter contexts. This approach was successfully used in large vocabulary continuous speech recognition and in phrase-based SMT systems.

2) Recurrent Neural Network LM (RNNLM)

Link: http://rnnlm.org/

Intro: Neural network based language models are nowdays among the most successful techniques for statistical language modeling. They can be easily applied in wide range of tasks, including automatic speech recognition and machine translation, and provide significant improvements over classic backoff n-gram models. The 'rnnlm' toolkit can be used to train, evaluate and use such models.

3) word2vec

Link: https://github.com/dav/word2vec or http://deeplearning4j.org/word2vec.html

Intro: This tool provides an efficient implementation of the continuous bag-of-words and skip-gram architectures for computing vector representations of words. These representations can be subsequently used in many natural language processing applications and for further research.

4) Long Short Term Memory (LSTM)
Link: http://www.bioinf.jku.at/software/lstm/
Intro: Software for the state of the art recurrent neural network. Long Short-Term Memory Software

5) DL4J Deep Learning for Java
Link: http://deeplearning4j.org/
Intro: Deeplearning4j is the first commercial-grade, open-source, distributed deep-learning library written for Java and Scala. Integrated with Hadoop and Spark, DL4J is designed to be used in business environments, rather than as a research tool. It aims to be cutting-edge plug and play, more convention than configuration, which allows for fast prototyping for non-researchers.

6) CURRENNT
Link: http://sourceforge.net/projects/currennt/
Intro: CUDA-enabled machine learning library for recurrent neural networks which can run both on Windows or Linux machines with CUDA-supported capability. CURRENNT is a machine learning library for Recurrent Neural Networks (RNNs) which uses NVIDIA graphics cards to accelerate the computations. The library implements uni- and bidirectional Long Short-Term Memory (LSTM) architectures and supports deep networks as well as very large data sets that do not fit into main memory.

7) ...

*** Deep learning materials

- For NLP

a) http://www.socher.org/index.php/DeepLearningTutorial/DeepLearningTutorial

b) http://cl.naist.jp/~kevinduh/a/deep2014/

c) http://ronan.collobert.com/pub/matos/2009_tutorial_nips.pdf

- For general background

a) http://cl.naist.jp/~kevinduh/notes/duh13deepadvances.pdf

b) http://www.trivedigaurav.com/blog/quoc-les-lectures-on-deep-learning/

c) http://deeplearning.net/reading-list/
d) (new book, 2014, draft  version) http://www.iro.umontreal.ca/~bengioy/dlbook/
e) http://people.idsia.ch/~juergen/deep-learning-overview.html 

- ...

(to be updated)

SENNA toolkit

Intro: SENNA is a software distributed under a non-commercial license, which outputs a host of Natural Language Processing (NLP) predictions: part-of-speech (POS) tags, chunking (CHK), name entity recognition (NER), semantic role labeling (SRL) and syntactic parsing (PSG).

SENNA is fast because it uses a simple architecture, self-contained because it does not rely on the output of existing NLP system, and accurate because it offers state-of-the-art or near state-of-the-art performance.

Link: http://ronan.collobert.com/senna/

Deep learning

I just find that this research topic is quite new.
I intend to get deeper into it, especially its impact in NLP research.

Some of review paper or tutorials:

1) http://deeplearning.net/
(tutorial: http://deeplearning.net/tutorial/)

2) ACL 2012 tutorial:
http://www.socher.org/index.php/DeepLearningTutorial/DeepLearningTutorial

3) Ronan Collobert (http://ronan.collobert.com/)
http://ronan.collobert.com/pub/matos/2008_nlp_icml.pdf
http://ronan.collobert.com/pub/matos/2009_tutorial_nips.pdf
http://ronan.collobert.com/pub/matos/2011_nlp_jmlr.pdf

4) ...