User Tools

Site Tools


projects:current:start

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision
Previous revision
projects:current:start [2015/02/11 01:41] hjprojects:current:start [2015/02/11 01:51] (current) hj
Line 2: Line 2:
  
  
-===== Current Projects ===== +===== Hybrid Orthogonal Projection & Estimation (HOPE) =====
- +
-  - **Hybrid Orthogonal Projection & Estimation (HOPE)** +
  
 {{:projects:current:one-nn-layer.png?0x220 |NN hidden layer}}{{ :projects:current:one-hope-layer.png?0x220| HOPE}} {{:projects:current:one-nn-layer.png?0x220 |NN hidden layer}}{{ :projects:current:one-hope-layer.png?0x220| HOPE}}
-Each hidden layer in neural networks can be formulated as one HOPE model+Each hidden layer in neural networks can be formulated as one HOPE model[1]
 \\ \\
 +  * The HOPE model combines a linear orthogonal projection and a mixture model under a uni ed generative modelling framework; 
 +  * The HOPE model can be used as a novel tool to probe why and how NNs work; 
 +  * The HOPE model provides several new learning algorithms to learn NNs either supervisedly or unsupervisedly. 
 \\ \\
 +**Reference:** 
 \\ \\
-The HOPE model combines a linear orthogonal projection and a mixture model under a uni ed generative modelling framework. Each hidden layer can be reformulated as a HOPE model. As a resultthe HOPE framework +[1] //Shiliang Zhang and Hui Jiang//"Hybrid Orthogonal Projection and Estimation (HOPE): A New Framework to Probe and Learn Neural Networks," [[http://arxiv.org/abs/1502.00702|arXiv:1502.00702]].  
-can be used as a novel tool to probe why and how NNs workmore importantly, it also + 
-provides several new learning algorithms to learn NNs either supervisedly or unsupervisedlyIn +**Software:** 
-this work, we have investigated the HOPE framework in learning NNs for several standard tasks, +\\ 
-including image recognition on MNIST and speech recognition on TIMIT. Experimental results +The matlab codes to reproduce the MNIST results in [1] can be downloaded here. 
-show that the HOPE framework yields signi cant performance gains over the current stateof- +
-the-art methods in various types of NN learning problems, including unsupervised feature +
-learning, supervised or semi-supervised learning.+
projects/current/start.1423618910.txt.gz · Last modified: by hj