Features
ANNHUB key features
Easy to use
ANNHUB does not require deep machine learning knowledge and programming skills to design a neural network for your application. Friendly user interface with auto-suggest features allow an user complete AI design task with ease.
Advanced algorithms
ANNHUB supports advanced training algorithms, including Scaled Conjugate Gradient, Levenberg Marquardt, Quasi Newton, and Bayesian Regularization that help speeding up training speed and improving convergence rate.
Effective Evaluation
ANNHUB supports various evaluation metrics, including ROC curves, confusion matrix, performance metrics, and regression curves to effectively evaluate a trained neural network model. The performance of the trained model is also can be evaluated with new dataset.
Flexibility
ANNHUB supports different types of activations functions and cost functions that allow a designer to customise a neural network structure. Bayesian Neural Network is supported to effectively deals with small dataset applications.
Easy to deploy
Trained neural network model can be easily deployed in real-time applications developed various programming languages such as LabVIEW, LabVIEW NXG, LabWindow CVI, C.C++, C#, iOS, Android and Arduino by using appropriate API provided by ANS Center.
Design assitance
Based on dataset loaded by a designer, ANNHUB suggests a neural network structure to achive optimal performance. During training process, over-fitting issue will be taken care by early stopping technique or Bayesian regularization automactically.
Design process
How to design a neural network
Load dataset into ANNHUB
ANNHUB supports dataset in comma separated values (CSV) format as shown above figure. The ouputs are identified by key words "output, target, class". Each row in the csv file is equivalents to 1 data sample.
Configure a neural network structure
ANNHUB supports various activation types for both hidden layer and output layer. Both pre-processing and post-processing methods are provided to normalise dataset inputs and outputs. Designers can use different cost functions in conjuction with training engines to customise a neural network structure that suits their applications. Suggest neural network structure is also provided based on dataset format.
Train a neural network
ANNHUB seperates dataset into training set, validation set and test set. During training process, training set is used to train a neural network by optimising its cost function while validation set is used to prevent over-fitting issue. If Bayesian Neural Network is used, regularization will help to prevent over-fitting. The training is finished when stopping criteria are met.
Evaluate the trained neural network using ROC curve.
ANNHUB supports ROC curve technique to evaluate the performance of the trained neural network on both training set, validation set and test set. These ROC curves provide important information that helps AI design process.
Evaluate the trained neural network using confusion matrix.
ANNHUB supports confusion matrix technique to evaluate the performance of the trained neural network on both training set, validation set and test set. These confusion matries also include accuracy, sensitity and specificity information that helps classifcation design process.
Evaluate the trained neural network using regression curve
ANNHUB supports regression curve technique to evaluate the performance of the trained neural network on both training set, validation set and test set. These regression curves provide important information that helps regression design process
Evaluate the trained neural network with new dataset.
Completely new dataset can be loaded into ANNHUB to evaluate the performacen of the trained neural network model. This helps to verify the trained model before deciding to deploy it into a real-time applcation.
Export trained neural network model for deploymennt
ANNHUB supports exporting trained neural network model into a weight file. This weight file is then can be loaded into different programming environments using Application Programming Interface (API) provided by ANS Center. By using appropriate programming environment, the trained model can easily be deployed into any real-time applications.
Application Programming Interface
How to deploy a neural network into applications
After the trained neural network has been verified and tested with new dataset, it is ready to be deployed in a real-time application. ANNHUB allows user to export this trained model into a file with "ann" extenstion. This file contains the trained neural network structure as well as its parameters.
ANNHUB supports different API to allow deploying the trained neural network model into a application developed by different programming languages such as LabVIEW, LabVIEW NXG, LabWindow CVI, C/C++/C#, Android, iOS and Arduino. Please login ANSCENTER website and download appropriate API from Customer Portal. Use supported API to import a trained model in “ann” extension text file into a machine learning application developed in a supported programming environment.
The customised application then can be compiled and deployed to an appropriate hardware platform.
Case studies
Learn ANNHUB via case studies
ANS Center provides elegent solutions to simplify machine learning and deep learning design and deployment.
We support mutiple hardware platforms and programming languages, including LabVIEW, LabVIEW NXG, LabWindow CVI, C/C++/C#, and Arduino.
Latest Posts
How to implement Artificial Intelligence in an Arduino Project
3/11/2019
Thought Control System
3/11/2019
Handwriting recognition
3/11/2019