The most popular and comprehensive Open Source ECM platform
Let us know more
Neural Algorithmic Reasoning: An Approach for Solving Messy Real-World Problems with Algorithmic Elegance
The use of neural networks in AI research have led to very impressive results which include:
- Image classification and identification – figuring out what objects are in an image
- Generation of text that can be written in a certain style and topic
- Language translation
- Image generation of objects or people which do not exist
Neural networks themselves are relatively simple. They consist of nodes that accept input data, apply simple processing to the input, and then produce the new output. The magic comes when a network is created that combines and layers millions of nodes. While the work done by a single node isn’t very interesting, the results produced from a large network of interacting nodes can be amazing.
Researchers are now trying to improve and make the internals of neural networks smarter by integrating them with standard algorithms or known principles. Improvements may mean needing less data to train and enabling of even more inciteful insights.
Petar Veličković, Charles Blundell, researchers at Google’s DeepMind, wrote that “algorithms possess fundamentally different qualities to deep learning methods, and this strongly suggests that, were deep learning methods better able to mimic algorithms, generalization of the sort seen with algorithms would become possible with deep learning — something far out of the reach of current machine learning methods. Furthermore, by representing elements in a continuous space of learnt algorithms, neural networks are able to adapt known algorithms more closely to real-world problems, potentially finding more efficient and pragmatic solutions than those proposed by human computer scientists.”
While algorithms are often elegant in their solutions, they often require assumptions and approximations to be made so that the problem can be tractably solved. Combining algorithms with neural networks allows for there to still be elegance but it also allows messier kinds of problems to be solved which more accurately simulate reality.
