The month of March has seen some very interesting advances in the AI and Tech space. So in case you’ve missed anything, we’ve got you covered.
With new findings surrounding neural network hardware, universal memory and the development of advanced prosthetics, it’s probably safe to assume that the future is looking pretty amazing.
That said, let’s take a look at some of the latest progress that we’ve made in the growing field of AI.
“Shrinking” The Hardware Needed For Neural Networking
Self-driving cars, financial forecasting, image and voice recognition. Those are only a few of the amazing features that as well as making sense of the petabytes involved in big data are only some of the amazing benefits that Artificial Neural Networks (ANN) can bring to us.
But with the computing power and hardware required to have these neural networks function effectively, it becomes increasingly difficult to scale and advance the technology.
That’s exactly why researchers at the University of California San Diego have developed an artificial neuron device capable of running neural network computations by using “100 to 1000 times less energy and area than traditional CMOS-based hardware”.
Essentially, neural networks are trained to act in the same way that the human brain does. It takes different information in (as an input) and then outputs insights and
patterns that can be used for more inputs and further processing. In order for this to work properly, a mathematical calculation (known as a non-linear activation function) needs to be applied.
This non-linear activation function is critical to the operation of a neural network. The only issue is that it tends to take up a lot of computing power and circuitry, as a transfer of data between memory and an external processor needs to constantly take place.
So by focusing on the activation function, researchers have developed a nanometer-sized device to handle the process more effectively. For those of you that don’t know, 1mm consists of around 1,000,000 nanometers – it’s essentially microscopic.
“We developed a single nanoscale artificial neuron device that implements these computations in hardware in a very area- and energy-efficient way,” says Duygu Kuzum, an electrical and computer engineering professor at UC San Diego.
Ultimately, this development can help in dealing with highly complex and high demanding tasks, such as facial and object recognition (required in self-driving cars) and could be an absolute game-changer for machine learning and our world as we know it.
“Universal Memory” and ULTRARAM Taking It’s First Step
While we’re all used to the two main types of memory (DRAM and Flash), it will probably come as a surprise to you that a whole new form of RAM is being developed. And the aim is to combine all of the best features that both of them have to offer.
And the first signs of progress are being made by Lancaster University physicists.
We all know that standard dynamic RAM (DRAM) is fast and efficient, but unable to store information when power is removed; while flash is slow, but great for storage (and can’t be used for active memory). Now with ‘universal memory’, both storing and changing data can be achieved – something that was considered impossible, until now.
By using a quantum mechanical effect called resonant tunneling, these physicists are able to allow barriers to switch from opaque to transparent using a small voltage.
This new, non-volatile RAM is being used to progress the implementation of ‘universal memory’, which is said to have all of the benefits of using DRAM and flash without any of the downsides.
Through modifying the design of their device to make proper use of resonant tunneling, they’ve achieved speed increases that are up to 2000 times faster than the original prototypes and are said to be at least 10 times better than flash – with no compromise in data retention.
Robotics and AI Working on Smart Exoskeletons
Using computer vision and deep-learning AI, researchers are developing prosthetic legs and exoskeletons that are capable of both thinking and making calculated decisions on their own.
This technology is being used to mimic the way that people walk and identify their surroundings, while adjusting their movements.
While exoskeletons are currently being operated by motors through the use of smartphone software or joysticks, researchers at the University of Waterloo are attempting to give vision to these exoskeletons so that they are able to control themselves.
Unfortunately, using applications and joysticks can be considered a hassle, as users are forced to make frequent use of them for most changes in movement.
“That can be inconvenient and cognitively demanding,” said Brokoslaw Laschowski, a PhD candidate in systems design engineering and leader of University of Waterloo research project, ExoNet. “Every time you want to perform a new locomotor activity, you have to stop, take out your smartphone and select the desired mode.”
To combat this, the researchers fit exoskeleton users with wearable cameras and make use of AI-enabled software to recognise and process external objects like doors and stairs through the video feed.
The next step in developing this technology will see that these exoskeletons are capable of taking instructions in order to avoid obstacles, use stairs and engage in the right actions according to user movement and the surrounding terrain/environment.
“Similar to autonomous cars that drive themselves, we’re designing autonomous exoskeletons and prosthetic legs that walk for themselves,” said Laschowski.
Another interesting addition that is still being explored and developed will see human motion self-charge batteries for greater energy efficiency in the motors.
The world of technology and AI is on a rapid growth trajectory. And it’s not slowing down any time soon. Want to know more on AI? Click Here.