Deep Learning for Perception
Georgia Tech, Spring 2015
Zsolt Kira
Software Packages

1 Recurrent Neural Networks (Language/Sequences)

• RNNLM- Language modeling toolkit
• RNNLIB- Sequence learning (speech, handwriting are typical applications)
• Passage- Text analysis using RNNs
• Stanford Sentiment Analysis and CoreNLP
• CSLM: Continuous Space Language Model Toolkit
• Senna- Neural networks for many different tasks: part-of-speech (POS) tags, chunking (CHK), name entity recognition (NER), semantic role labeling (SRL) and syntactic parsing (PSG)

2 General Libraries used by Researchers

• Theano- Python-based library, mainly an implementation of a symbolic expression compiler so very different from traditional “coding”-based libraries for deep learning
• Pylearn2- Another python-based library
• Torch- State of the art learning algorithms, combination of Lua/C.

3 Convolutional Neural Networks

• Caffe- Mostly used by computer vision researchers and applications.
  • Nice support for specifying architectures via text files.
  • Has a model zoo with pre-trained networks
  • Note you can implement lots of things (RNNs, semantic segmentation, etc.) but typically it’s not well-documented so you have to figure out how to bend Caffe to achieve it
• MatConvNet - Matlab-based library
• cuda-convnet2 - Fast cuda-based library