So a few days ago I bought a Windows Phone device. So far, I am impressed with the Windows Phone ecosystem.
The nicest thing is that I was finally able to test an app I've been eager to try for months: Point and Call.
Point-and-call in everyday life.
Now, the most interesting part, is that this app uses part of the Accord.NET Framework to do its magic :-)
The app author, Antti Savolainen, was kind enough to share some details about his app. It uses part of the SVMs framelet from Accord.NET to do the digit recognition, mostly based on one of the earlier CodeProject articles I've posted in the past. Needless to say, Antti did an awesome job, as the SVM part was surely just a tiny fraction of all the work in preprocessing, adjusting, locating, and doing the right things at the right times that I would never be able to figure out alone :-)
Surely, he and his company, Sadiga, deserves all the credits for this neat app!
If you would like find more interesting uses of the Accord.NET Framework, don't forget to check the framework's Showcase page for details!
In the meantime, this article is also serving as a hook to a future article, an article about Hidden Conditional Random Fields (HCRFs). The HCRF models can serve the same purpose as the HMMs but can be generalized to arbitrary graph structures and be trained discriminatively, which could be an advantage on classification tasks.
As always, I hope readers can find it a good read :-)
SlimDX.Direct3D9.Direct3DX9NotFoundException was unhandled Message=Direct3DX 9 was not found. Please install the latest DirectX end-user redistributable package from Microsoft. Source=SlimDX StackTrace: at SlimDX.Direct3D9.Direct3D..ctor() InnerException: System.Runtime.InteropServices.SEHException Message=External component has thrown an exception. Source=SlimDX ErrorCode=-2147467259 StackTrace: at D3DXCheckVersion(UInt32 , UInt32 ) at SlimDX.Direct3D9.Direct3D..ctor() InnerException:
when trying to create a SlimDX.Direct3D9.Direct3D object, most likely you will be able to solve this by placing D3DX9_43.dll into your output folder. Please note that the proper solution would be to install the full DirectX 9 runtime into your system, but it may be too much for some quick testing.
A similar error may occur when using SharpDX, but in SharpDX's case, the error message will be much more helpful and actually tell you what is missing.
I've recently started to record videos to demonstrate some capabilities of the Accord.NET Framework. Surprisingly, there were only a few, free, opensource applications to achieve this goal - and none of them had all the features I needed.
It is, until I decided to roll my own.
Screencast Capture Lite is a tool for recording the desktop screen and saving it to a video file, preserving quality as much as possible. However, this does not mean it produces gigantic files which take a long time to be uploaded to the web. The application encodes everything using solely H624 in an almost lossless setting.
As a demonstration, please take a look on the Youtube video sample shown below. However, note that Youtube actually reduced the quality of the video, even if you watch it in HD. The local copy produced by Screencast Capture has an even higher quality than what is being shown, while the generated video file occupied less than 2 megabytes on disk.
And by the way what would be a better approach to demonstrate the capabilities of the AForge.NET and Accord.NET frameworks other than writing this application using them?
Well, actually this application has been created specifically for two things:
to aid in the recording of instructional videos for the Accord.NET Framework, and;
to serve itself as a demonstration of the use and capabilities of the AForge and Accord Frameworks.
This means the application is written entirely in C# making extensive use of both aforementioned frameworks. The application is completely open source and free, distributed under the terms of the GPL, and a suitable project page is already being served on Google Code.
With the beginning of this year, I would like to share a video I wish I had found earlier. It is about the recent breakthrough given by Deep Neural Networks in the field of speech recognition - which, despite I had known was a breakthrough, I didn't know it was already leading to such surprising great results.
Deep neural networks are also available in the Accord.NET Framework. However, they've been a very recent addition - if you find any issues, bugs, or just wish to collaborate on development, please let me know!
Class diagram for Deep Neural Networks in the Accord.Neuro namespace.
Deep neural networks have been listed as a recent breakthrough in signal and image processing applications, such as in speech recognition and visual object detection. However, is not the neural networks which are the new things here; but rather, the learning algorithms. Neural Networks have existed for decades, but previous learning algorithms were unsuitable to learn networks with more than one or two hidden layers.
But why more layers?
The Universal Approximation Theorem (Cybenko 1989; Hornik 1991) states that a standard multi-layer activation neural network with a single hidden layer is already capable of approximating any arbitrary real function with arbitrary precision. Why then create networks with more than one layer?
To reduce complexity. Networks with a single hidden layer may arbitrarily approximate any function, but they may require an exponential number of neurons to do so. We can borrow a more tactile example from the electronics field. Any boolean function can be expressed using only a single layer of AND, OR and NOT gates (or even only NAND gates). However, one would hardly use only this to fully design, let's say, a computer processor. Rather, specific behaviors would be modeled in logic blocks, and those blocks would then be combined to form more complex blocks until we create a all-compassing block implementing the entire CPU.
The use of several hidden layers is no different. By allowing more layers we allow the network to model more complex behavior with less activation neurons; futhermore the first layers of the network may specialize on detecting more specific structures to help in the later classification. Dimensionality reduction and feature extraction could have been performed directly inside the network on its first layers rather than using specific separate algorithms.
Do computers dream of electric sheep?
The key insight in learning deep networks was to apply a pre-training algorithm which could be used to tune individual hidden layers separately. Each layer is learned separately without supervision. This means the layers are able to learn features without knowing their corresponding output label. This is known as a pre-training algorithm because, after all layers have been learned unsupervised, a final supervised algorithm is used to fine-tune the network to perform the specific classification task at hand.
As shown in the class diagram on top of this post, Deep Networks are simply cascades of Restricted Boltzmann Machines (RBMs). Each layer of the final network is created by connecting the hidden layers of each RBM as if they were hidden layers of a single activation neural network.
Now, the most interesting part about this approach will given now. It is about one specific detail on how the RBMs are learned, which in turn allows a very interesting use of the final networks. As each layer is a RBM learned using an unsupervised algorithm, they can be seen as standard generative models. And if they are generative, they can be used to reconstruct what they have learned. And by sequentially alternating computation and reconstruction steps initialized with a random observation vector, the networks may produce patterns which have been created using solely they inner knowledge about the concepts it has learned. This may be seen fantastically close to the concept of a dream.
The new release also includes k-dimensional trees, also known as kd-trees, which can be use to speed up nearest neighbor lookups in algorithms which need it. They are particularly useful in algorithms such as the mean shift algorithm for data clustering, which has been included as well; and in instance classification algorithms such as the k-nearest neighbors.
.png)
