In a paper published in 2019, Nature, artificial intelligence (AI) researchers demonstrated a self-driving bicycle navigating around different obstacles, responding to voice commands, and following a person. Although the self-driving bicycle was of little use, the underlying technology behind the concept was remarkable. These researchers powered the bicycle using a neuromorphic chip, a unique AI concept.
If you go down in history, you will get to know that the concept of neuromorphic computing is not new. This concept had been proposed in the 1980s. However, it is only now that its interest is being developed in the industry.
The accelerated growth of neural networks and deep learning prompted the development of AI hardware specialized at large. Owing to this, one of the growing trends is neuromorphic computing. Let us refer to a case study to showcase our advancement in this technology.
Case study:
Computer-science research was driven by Intel Labs whose focus was on third-generation AI. The research covered major areas in neuromorphic computing – how to emulate the neural structure of the brain, probabilistic computing to further create an algorithmic approach in dealing with uncertainties or contradictions taking place in the natural world.
Despite the advent of this technology in the real world, organizations still faced challenges.
One of the major challenges includes to match the human’s flexibility with the ability to learn from unstructured stimuli equal to the amount of efficiency that is required of a human brain.
However, using computational building blocks within a neuromorphic computing system is logically analogous to the neuron. Therefore, running through Spiking neural networks (SNNs) could be a novel approach in arranging such elements to imitate neural networks present in the human brain.
Here’s how it works:
Each neuron present in the SNN can be processed independently à this further sends signals to the other neurons within the network to change the electrical state of those neurons à information gets encoded within the signals à this is when SNNs imitate NLP by remapping the synapses in response to the stimuli.
Neuromorphic computing leading to analog revolution
The neuromorphic chip or neuromorphic device are physical structures representing a form of an artificial neural network. Every neuromorphic chip available composes of multiple small computing units corresponding to an artificial neuron.
In comparison to the CPUs, these chips cannot perform multiple operations. They can only perform a single mathematical function of a single neuron.
Given that, the physical connection between artificial neurons makes it more like an organic brain, it creates a formation of physically connected artificial neurons providing neuromorphic computers their core strength.
At its core, the human brain can function differently, some of which we’re not even aware of. But with a neuromorphic device, it can only be developed to carry out one single function. Therefore, for this device to become ‘artificially intelligent’ it requires to a perform task that otherwise requires intelligence (like a human being).
Perhaps your next question might be, can neuromorphic computing function like a brain?
An AI algorithm undergoes an analysis as a result of the simulation which is smart until another algorithm tags along. Whereas, with a neuromorphic device, it can only function in a way that corresponds to some part of the brain. These parts could either be calculation, memory, optimization, or logic of a method.
You cannot specifically say that this device is qualified as an AI. This is because, with intelligence, these functions need to work in tandem, whereas the underlying mechanism in neuromorphic helps it mimic just a single function.
This is where the term neuromorphism steps in, which means taking the form of a brain. Although it doesn’t explain how equipped will the brain function, but it does help study the mechanism of a human brain. For instance, the kind of information to be remembered, or how many neurons need to be fired before any decision is made.
The point is, neuromorphic is yet to move beyond the question of whether our analysis of reasoning can be represented by the value stored in chips. It is still a long shot for artificial intelligence engineers. However, they might still be content in simply making a simple discovery as to why brains require less energy to maintain while producing so much information and analysis.
Though we’re yet to fill the bits and pieces around the technology – it has shown quite a promising AI future.
Here are a few examples of recent developments in the field:
- Loihi, the fifth-generation self-learning neuromorphic learning chip designed by Intel Labs in 2017
- The brain-inspired chip by TrueNorth – IBM Labs
- Brain on a chip – MIT
The future
By 2025, neuromorphic chips will be entrenched in every smartphone while prediction also says, the business of neuromorphic chip will rise to USD 1.78 billion.
Are you ready to compete with near-human levels of data perception?
