Neural networks curse of dimensionality

ACL 2012: Tutorial on Deep Learning
Effect of the curse of Deep Learning for NLP


1. Basic

Representation

DBNs MRFs, multiple neural networks


1.1 Five reasons to explore deep learning

#1 Learning representation: byond handcrafted features

#2 Need for distributed representations

binary features are fragile

distributional similarity based word clusters

multi-clustering (multiple clusters per data)

curse of dimensionality

neural network learns kernel

#3 unsupervised feature and weight learning

#4 learn multiple levels of representation

#5 why now? : New methods for unsupervised pre-training (RMB, autoencoders, etc)


1.2 From Logistic regression to neural networks

natural language model

Training with backpropagation


1.3 Word Representations

1.4 Unsupervised word vector learning

1.5 Backpropagation Training


2. Recursive Neural Network

Learn Structure and Representation

Sentiment Analysis

Autoencoder


3 Applications

neural language models

speech and nlp applications

resources

tricks of the trade

discussion

Source: Japanese language processing blog

You might also like:

Lecture 03, part 1 | Pattern Recognition
Lecture 03, part 1 | Pattern Recognition
How to derive the volume of an n-dimensional hypersphere in 3 minutes.
How to derive the volume of an n-dimensional hypersphere in 3 minutes.
Nekrosys - Curse of Dimensionality (Indian Metal)
Nekrosys - Curse of Dimensionality (Indian Metal)

Stock market analysts on trial

2002-02-09 14:24:25 by on-trial

The amount of poor and self-interested advice that is being issued by brokerages and their analysts. To this day, the majority of stockbrokers are compensated on the number of trades their customers make, not on the returns they generate for them or on the quality of the advice they provide. We believe that the price targets and analyst ratings are made with several masters in mind, none of whom are the individual investor. In a similar fashion, sell-side stock analysts are generally compensated based upon the overall profitability of their firms, not the quality or accuracy of their analysis. In the end, analysts have minimal structural incentive to be accurate in their predictions; rather their built-in incentive is to be as favorable to their corporate clients as possible. It is a...

Gurus' Results Stay Consistently Bad  — Forbes
Investment gurus make their money selling market predictions, not following them. Their overall performance has been historically and consistently dismal. Why people pay for market predictions is a one of Wall Street's biggest mysteries.

Springer Neural Information Processing: 13th International Conference, ICONIP 2006, Hong Kong, China, October 3-6, 2006, Proceedings, Part III (Lecture Notes ... Computer Science and General Issues)
Book (Springer)
Curse of Dimensionality 1
Curse of Dimensionality 1

Related posts:

  1. Neural Networks of attention
  2. Neural Networks for Machine Learning
  3. Neural Networks for regression Analysis
  4. Neural Networks used in Robotics