AWS and Microsoft announced Gluon this month, stating that Gluon provides an API for defining machine learning models using pre-built, optimized neural network components to accomplish this. Some developers may be very talented, but lack experience with machine learning. Because the Gluon interface resembles traditional code, it allows itself to be defined and manipulated like any other data structure. Researchers and data scientists will enjoy the ability to make quick prototypes and utilize dynamic neural network graphs for completely new model architectures without losing training speed.
Gluon is ready to use now in Apache MXNet with Microsoft Cognitive Toolkit being released soon. Over time, more frameworks will be available.
Grakn is a database for AI that processes knowledge that is too complex for current databases. Grakn enables machines to manage complex data that serves as a knowledge base for cognitive/AI systems.
How Can We Speed Up The Process?
Machine learning with neural networks has three main components: data for training, a neural network model and the algorithm which trains the neural network. Learning frameworks help to speed up the process, but in order to achieve their optimizations, most frameworks need the developer to do some extra work – like providing a formal definition of the network graph up-front, and then “freezing” the graph, and lastly, adjusting the weights. With millions of connections, the network definition can be massive – and complex, usually having to be constructed by hand. Some learning networks can be unwieldy and difficult to debug. It can also be hard to reuse code between projects. The results can be difficult for beginners and is a time-consuming task for more experienced researchers. Methods established by Gluon or Grakn can be very helpful in streamlining the process.
The Building Blocks Of Gluon and Grakn
Let’s explore these two ways of creating machine learning models further.
There are four components to Gluon.
- The API is friendly. Gluon networks can be defined using simple and clear code.
- The network definition in Gluon is dynamic. It is flexible like any other data structure. This is a stark contrast to the “etched-in-stone” variety of network defining in deep learning frameworks necessary for optimization during training.
- The algorithm defines the network. That means developers can use standard programming loops and conditionals to create these networks and researchers can now define even more sophisticated algorithms and models which were not possible before.
- Operators for training are high performance. Get friendly and concise API and dynamic graphs without sacrificing training speed - a huge step forward in machine learning.
- Grakn allows you to model the real world and all the hierarchies and hyper-relationships contained in it.
- Grakn’s ontology functions as a data schema constraint that guarantees information consistency.
Graql is Grakn’s declarative reasoning and analytics query language. Graql allows you to derive implicit information that is hidden in your dataset, as well as reducing the complexity of that information.
In conclusion, Gluon and Grakn offer assistance forward into AI and our future. There is, of course, a lot more to it than what has been reported here. This is a lot to take in, and the words written here are the tip of the iceberg. Read more about the Semantic Web here.