top of page
What is Neuromorphic Computing?
Hey, what's with the retrofuturistic arcade theme?
Neurobotx is retrofuturistic, and so is neuromorphic computing. Also, we are techno-optimistic nostalgics and we like techno. The main advantage of neuromorphic computing stands in it's ANALOG, as opposed to digital nature. That's what generates the huge energy savings and the increase in speed. Computers were meant to be analog and neuromorphic since their very inception. So the retrofuturistic theme of neurobotx is about revisiting the most fundamental principles in computing, and bringing them back into the mainstream again: analog computing, energy efficiency, and going beyond the classical logic of 0s and 1s and get gradients instead, just like neurons when they spike. We are driven and inspired by Isaac Asimov's sci fi novels, flying cars, androids, robots and the 80s techno-optimism. We are looking to bring that vision back into our currently reality.
What does 'neuromorphic' mean?
Neuromorphic means 'shaped like the brain'. In fact, all neural networks are neuro-morphic, as they have the term 'neural' in them. Neural networks aim to mimic brain function through the resemblance between their nodes and the different layers in the network with how neurons are actually arranged in the cortex, also functioning as nodes, and arranged in layers with different logic rules between them (eg. 'input' or 'output' layer).
So if neural networks are already neuromorphic, then what's the point?
Current neural networks are a very crude, or oversimplified version of how our brains work! As you are reading this (you are reading about how your brain works, meaning you are a sentient being) your brain is running billions of computations in parallel, and all are connected to each other. Your eyes know where to look in order to understand the text because your prefrontal cortex told your visual cortex to instruct your eyes to perceive the pixels and wavelenghts on your monitor such as way as to comprehend what is written. Your prefrontal cortex is then further integrating that information, and updating your existing knowledge about the brain and neural networks. Your brain is doing all of this: updating its own knowledge of the brain and AI, by controlling your senses. And here is the fun part: your brain is currently using less energy than a light bulb! Most of what goes on in your brain comes from passive ion movements, with very few exceptions of a few ion pumps. So essentially the revelation you just had is nothing but passive flow of current through some cable and transistors. To explain this even better, we will explain the action potential to you. But first, you should now understand that 'neuromorphic' is a buzzword, just like AI, or AGI, and that emulating brain processes is an ongoing and very ambitious project.
There are so many 'AI' companies out there. What is special about neurobotx?
Neurobotx is the AI company with the highest level of expertise in brain science, and how the brain perceives and navigates in a simulated vs. real environment. More specifically, the founder has focused 15 years of brain research of looking at input-output computation in single neurons, inside the brain of awake mice while they are running in a VR environment. The founder of neurobotx has focused her research in the laboratory of the 2 Nobel laureates that discovered the brain's navigation system and has tested that navigation system in a VR arena to see what is needed for the brain to perecive the VR environment as real. The founding team is also composed of brilliant computer vision engineers, visionary venture capitalists, corporate intrapreneurs, business angels, and SLAM engineers. But what makes neurobotx special is that it has the know-how to emulate parts of the brain, using proprietary neurphysiological data from mouse brain. Yes, you read correctly, we use RAW DATA from the brain to build and run our algorithms.
What is brain emulation?
Glad you asked. The neurobotx founding team has coined the term 'Whole Brain Emulation' (or WBE) as a necesary but not sufficient condition for achieving general AI, or Artificial General Intelligence(AGI). A common question for AGI scientists is up to what level do we need to emulate the brain to achieve sentience. Brain areas? Circuits? Single neurons? Molecules? Atoms? The purpose of WBE is to identify which parts of brain functioning are most needed for improving current neural networks, and making them more powerful, faster and more energy efficient like our brain.
So how do we make neural networks more neuromorphic?
One of the first steps has been the invention of neuromorphic cameras. They are able to perceive only the edges out of the objects in every frame, which allows them to be much faster and more energy efficient than standard RGB cameras. This however, is a very small step in emulating the entire brain. This, in fact, is what the first layer of the human retina does! The next steps are emulating the function of the visual cortex, and a full reconstruction of the eye is still underway. Most of these efforts are spearheaded by neurobox.
At the same time, neuromorphic chips are gaining popularity as the increasing demand for more storage capacity and RAM in the last few years. Most seminconductor products are sold out now up to 2023, and an increase of approximately 1000x in computing power is estimated to be needed to sustain current data demands, according to Intel. 'Neuromorphic' chips are more like the brain in that the physical hardware contains nodes or neurons that can have a 1-to-1 correlation with the neural networks (ie. ithe 100 neurons in the neural network correspond to 100 physical nodes on the neuromorphic chip). When combining the neuromorphic sensors (which detect events instead of pixels), neuromorphic neural networks and neuromorphic chips, we can in principle reach 1000x in speed at 10.000x less energy consumption. This is the aim of neurobotx.
How do we achieve 1000x increase in speed at 10.000x less energy consumption?
Glad you asked. In order to achieve this we need to do a few things:
1. Stop detecting and analyzing every pixel. The future is intelligent perception, that is event-based, just like you are now reading this text. If your brain would allow your eyes to perceive every single pixel, you would run out of energy fast.
2. Bring both memory and computation in the same physical location on the chip. This is the essence of what is called 'edge computing'. As long as we are running memory storage and computation sequentially, we are going to loose a lot of energy.
3. Begin physical and software emulation of the rest of the retina, eye, visual cortex, motor cortex, auditory cortex and hippocampus. This system is one of the oldest in the brain (took millions of years to build) and will require significant collaborative effort between neuroanatomists, neurophysiologists, computational neuroscientists, computer vision and SLAM engineers to understand and integrate into an ongoing emulation project. At each stage, the benefits for each industry are enormous. Neurobotx has ambarked on this monumental task, and we are very excited to share this vision with you.
bottom of page