| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| ------------------------------------------------------------------------------------------------------------------------------ NeuronRain AsFer-KingCobra MAC Electronic Money - Proof-of-Work and Universally Unique ID Hash implementation ------------------------------------------------------------------------------------------------------------------------------ (#) Fictitious Message-As-Currency (MAC) in AsFer-KingCobra has been named "Neuro". (#) This commit implements a non-trivial proof-of-work computation and finds a universally unique hash id for each Neuro currency which has 2 leading "ff"s - Proof of Work is analogous to reCAPTCHA in websites for filtering out robots (#) Unique id is created from Boost UUID random generator and checked in a loop for 2 leading "ff" in stringified hex representation (#) Protocol Buffer version has been upgraded to 3.5/15 and currency.proto has been recompiled with protoc (#) asfercloudmoveclient.cpp has been updated to make a choice between std::move() and std::forward() move semantics: - std::move() just does =operator overload and moves Neuro currency over network - std::forward() + std::move() first deferences rvalue of && for currency uuid and then does =operator overload to move Neuro currency |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| g++ commandline changed for Optimized TileSearch |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| -------------------------------------------------------------------------------------------------------------------------------------- Compressed Sensing Update - Decompression and Error estimation of recovered image bitmap from sketch -------------------------------------------------------------------------------------------------------------------------------------- (*) Sketch Ax=B of an image bitmap has been persisted to a file CompressedSensing.sketch (*) A is a random matrix of dimensions (m,n) m<<n and m is scaled by a sketch ratio variable (*) Original image x is recovered from sketch Ax=B by inverting A and multiplying with sketch: Ainverse.Ax = x_recovered (*) For non-square only approximate pseudo inverse is computable. (*) For inverting non-square matrix A, Moore-Penrose Pseudoinverse function pinv() from NumPy is invoked. (*) Error of the recovered image computed by trace (sum of all entries) of the recovered image. (*) Logs for multiple sketch ratios (row values:100,200,300,400,50) have been committed to testlogs/ which show an increase in error as size of the sketch decreases. |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| ---------------------------------------------------------------------------------------------------- Compressed Sensing - Image Sketch implementation ---------------------------------------------------------------------------------------------------- (*) Sketch B of an image bitmap X is computed by multiplying with a random matrix A: B=AX (*) import ImageToBitMatrix from image_pattern_mining/ for mapping an image to a bitmap matrix (*) Sketch matrix B contains compressed information of the larger image. Original image can be sensed from this sketch. |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| ------------------------------------------------------------------------------------------------------ Support Vector Machines - Mercer Kernel Update ------------------------------------------------------------------------------------------------------ (*) Mercer Kernel Function has been changed to return a tuple of feature mapped points and the dot product (*) Feature mapped points in higher dimension are : [....., square_root(eigenvalue[i])*eigenfunction(x[i]),....] |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
almost 9 years ago
|
| ---------------------------------------------------------------------------------------------------- Support Vector Machines - Mercer Theorem - Kernel Implementation ---------------------------------------------------------------------------------------------------- (*) This commit implements the kernel trick for lifting points in lower dimension to higher dimension by a Feature map so that decision hyperplane in higher dimension separates the points accurately. (*) Mercer Theorem creates a kernel function unifies Feature map lifting and Dot product in higher dimension (*) Feature map phi maps a point in a dimension d to a point in dimension d+k: phi(x) = X. Inner Product (Dot) of two vectors in dimension d+k = phi(x)*phi(y). Mercer Theorem unifies the Feature map and Dot product into a Kernel function defined as series: K(x,y) = Sigma(eigenvalue(i)*eigenfunction(x)*eigenfunction(y)) (*) This implementation randomly instantiates a N*N square matrix, finds its Eigenvalues and Eigenvectors, and computes the series for K(x,y) as per previous identity (neglecting imaginary parts of Eigenvalues and Eigenvectors) (*) logs for this have been added to testlogs/ |
|
More...
|
almost 9 years ago
|
