Differences

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

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
Next revision Both sides next revision
en:projects:details:neuralsemproject [2011/09/23 14:57]
amir
en:projects:details:neuralsemproject [2013/06/12 13:52]
amir
Line 11: Line 11:
 */ */
 ---- dataentry project ---- ---- dataentry project ----
-title : Simple neural networks with error correcting abilities+title : Implementing Some Feature Extracting Techniques to Model Human Visual System
 contactname:​ Amir Hesam Salavati contactname:​ Amir Hesam Salavati
 contactmail_mail:​ hesam.salavati@epfl.ch contactmail_mail:​ hesam.salavati@epfl.ch
Line 17: Line 17:
 contactroom:​ BC 160 contactroom:​ BC 160
 type : master semester type : master semester
-status : available+status : completed
 created_dt : 2010-11-15 created_dt : 2010-11-15
-taken_dt : YYYY-MM-DD +taken_dt : 2013-01-19 
-completed_dt : YYYY-MM-DD +completed_dt : 2013-06-12 
-by : the full name of the student +by : Diego Marcos Gonzalez ​ 
-output_media : en:​projects:​neural_storage_capacity.pdf|Download ​Abstract ​in PDF Format+output_media : en:​projects:​master_semester:​marcos_salavati_semester_project_report_2013.pdf|Download ​Project Report ​in PDF Format
 table : projects table : projects
 ====== ======
 template:​datatemplates:​project template:​datatemplates:​project
 ---- ----
 +\\ 
 +\\
 /* Description of the project */ /* Description of the project */
-====== Background ======+===== Project Description ​===== 
 +\\ 
 +In computer vision, there are various different techniques to extract important features from images. These features are then later used in patten recognition,​ image classification,​ etc. Some of these techniques are comparable to some models of specific parts in human visual system. ​
  
-Memorizing patterns and correctly recalling them later is an essential ingredient of neural activity. ​In past 25 years, a number of neural networks ​ has been invented memorize and recall patterns. Interestingly, some of these networks are able to recall ​the correct pattern even if the input pattern contains error and is partially corrupted because ​of noiseIn this regard, these artificial networks resemble error correcting codes, i.e. they are able to recognize ​the correct pattern in presence of noise.+In this projectwe are interested in implementing ​some of the widely used techniques in feature learning (extraction) ​and applying them to a dataset ​of natural imagesThis usually corresponds to solving some optimization problem ​to find the features that represent the data more accurately.
  
-However, the storage capacity of these networks are quite small compared to their counterparts ​in coding theory. Given the fact that modern codes use the same basic structure to do error correction and the one used by neural networks, i.e. a bipartite graph with local message passing, it seems interesting to consider applications of modern coding theory to increase the storage capacity of neural networks by finding the appropriate weights for the neural graph+The implementation can be either done in C or MATLAB (MATLAB is preferred)
  
-Up to this point, some weighting schemes ​(including the Hebbian rulewere tested without much successConsidering more weighting schemes, ​such as the BCM rulewould be the next step toward ​the goal of increasing the storage capacity which is the main objective ​of this project+And here are some lines to give you an idea about why we are interested in this project:  
 +Once the feature extraction techniques are implementedthe learned features will then be used as inputs to a neural network which mimics ​some parts of human memory ​(neural associative memory). The ultimate goal would be to see if one will get better information storage capacities in artificial neural memories when the inputs are natural stimuli (such as images) and pre-processed before being stored. HErepre-processing refers to the feature extraction procedure. 
 +\\ 
 +\\ 
 +This project ​is suitable for students interested in computer vision, neural networks and mathematics who prefer a combination ​of theoretical and empirical works
  
 +The prerequisites are:
 +1)Basic knowledge of linear algebra. ​
 +2)Being familiar with a suitable programming language (MATLAB is preferred. But C/C++ is acceptable.)
  
-====== Project Goals ====== +Some knowledge about feature learning models in computer vision is not necessary but is deeply encouraged
-The objectives of this project are: +
-1)To understand the principles of neural networks and modern codes. +
-2)Applying different weighting scheme to neural networks and evaluate the performance for storing codewords of a linear code+
  
-This project is suitable for students who prefer doing simulations to find the appropriate framework for doing theoretical analysis. ​+\\ 
 +\\ 
 +\\
  
-The prerequisites are+====== Report ====== 
-1)Basic knowledge of coding theory. +The report ​ is available via the following link
-2)Being familiar with a suitable programming language (C/​C++,​MATLAB)+{{:​en:​projects:​master_semester:​marcos_salavati_semester_project_report_2013.pdf|Implementing Some Feature Extracting Techniques to Model Human Visual System}}