The most popular and comprehensive Open Source ECM platform
Neural Network techniques like deep learning are currently very popular. Frameworks to support neural net analysis include Theano, TensorFlow and Caffe. Those frameworks work by analyzing large numbers of example data iteratively, identifying characteristics of the data in one layer, and then passing results from one layer to the next one for further analysis. Each layer is able to resolve progressively more complex patterns from the data.
Some of the types of problems that neural networks have been successfully applied to include pattern association, image processing and identification, autonomous vehicles, robot steering, processing of inaccurate or incomplete data.
While there has been a lot of success with Neural Networks,it is only one of other types of AI. Pedro Domingo wrote a book in which he identified five different unique “tribes” of Artificial Intelligence. The Connectionist tribe supports the idea of Neural Networks:
- Symbolists – Inverse deduction
- Connectionists – Neural Networks and Backpropagation
- Evolutionaries – Genetic programming
- Bayesians – Probabilistic inference
- Analogizers – Kernel Machines
One issue with Neural Networks though is that they require huge amounts of data to be used for training the algorithm. Setting up neural networks takes a lot of time to collect the data and also to process it.
The Bayesian probabilistic approach is a potential alternative because it doesn’t require extensive training sets. Some experiments using Bayesian techniques in recognizing characters have had good results. In the future we might expect there to be many different types of AI techniques that are used.