Convolutional Neural Networks: A Cool Way To Do AI

Beyond the hype, AI is a complex evolving field. No article could ever do justice to this field; this post is focused on convolutional neural networks.

But first a quick primer on AI -- it is a spectrum (source: Justin Gage):

And the many applications are leveraging many different techniques (source: Hubspot):

Amidst all these techniques, neural networks (NNs or artificial neural networks aka ANNs) hold much promise since they are computing systems that learn by considering examples, generally without being programmed with any task-specific rules. NNs are vaguely inspired by biology, namely by the neural networks in animal brains. The challenge is they need a lot of training data otherwise they will just come up with erroneous conclusions.

Convolutional neural networks (CNNs) is a variation of neural network that is often applied to visual imagery. CNN is also inspired by biology, most specifically by how individual neurons respond to stimuli and by overlapping cover the entire visual field. The diagram below illustrates how a trained CNN can be more efficient in getting to an output (source: Julien Despois):

CNN uses little pre-processing and thus can learn many things that have to be otherwise hard-wired. Some major disadvantages are that it is computationally intensive and, being a type of NN, needs a lot of training data otherwise you may not get the full picture (no pun intended).

CNNs have also been mostly used for image recognition and video analysis, but also for natural language processing, drug discovery, health risk assessment, and discovering biomarkers of aging. Popular media has showcased CNN especially through AlphaGo, created by Google DeepMind team to play and win the board game Go against the top human player, which used CNNs in several flavors.

Have an opinion on CNNs -- where it is, where it is going, how it gets there? Comment away.

These are purposely short articles focused on practical insights (I call it gl;dr -- good length; did read). I would be stoked if they get people interested enough in a topic to explore in further depth. I work for Samsung’s innovation unit called NEXT, focused on early-stage venture investments in software and services in deep tech, and all opinions expressed here are my own.