Numenta Research: Key Discoveries in Understanding How the Brain Works
This is your brain. More specifically, it’s your neocortex, one of the most fascinating parts of a mammalian brain. In humans, it makes up about 70% of the brain’s volume.
This wrinkly sheet of cells, about the size of a dinner napkin, and 2.5 mm thick, is responsible for intelligence. It’s the only structure that knows things, the only structure that discovers things. It defines us as a species. And if we want to understand ourselves, if we want to understand intelligence, then we have to understand the neocortex.
Scientists have spent more than a hundred years trying to decipher its complex circuitry. What’s most remarkable about the neocortex is the consistency of the microarchitecture. Everywhere you look, you find the same common elements. Neuroscientist Vernon Mountcastle was the first to propose that because every part of the neocortex has the same complex circuitry, then every part of the neocortex is doing the same thing. He argued that the way we see, hear, feel, speak – even the way we do high level planning runs on the same circuitry. Given that a column contains all the essential elements of this circuit, Mountcastle proposed that if we can understand a column, we will understand the cortex.
The cortical column is an amazing thing. It has roughly 100,000 neurons. Most of them are called pyramidal neurons. Each has thousands of connections to other neurons. You may have heard there are six layers, but scientists now know there are dozens of different cell types. They form a very complex circuit with many cells in each layer that connect to other layers and other parts of the brain.
Our goal at Numenta is to understand the function and operation of this complex circuit: What are the neurons doing? What are the layers doing? How does a cortical column work?
So what have we learned? Our research team has been working on this for more than a decade. We’ve made some important discoveries that lay the foundation for understanding the neocortex:
Discovery #1: How the neuron works First is a new understanding of how the neuron works and the function of dendritic spikes. This discovery is documented in a March 2016 paper called “Why Neurons Have Thousands of Synapses, a Theory of Sequence Memory in Neocortex.” The main idea is that every pyramidal cell is a prediction machine.
To understand this discovery, first you must understand that of the thousands of synapses on a typical pyramidal neuron, roughly 90% create dendritic spikes rather than action potential spikes. Action potential spikes occur when a neuron communicates to another neuron. The spike starts in the soma and travels down the axon.
Dendritic spikes depolarize the cell, but don’t create an action potential, and therefore don’t leave the cell. Scientists have known this for many years, but the purpose of dendritic spikes was not understood. We realized that dendritic spikes put the cell into a predictive state. A depolarized or predictive neuron will fire a little sooner than other neurons and will inhibit its neighbors from spiking. This is how the brain knows when its predictions are correct. One pyramidal neuron can recognize hundreds of unique patterns, hundreds of unique contexts in which it can predict its input.
Discovery #2: Cortical columns are more powerful than previously thought Our second major discovery was that cortical columns are more powerful than previously thought. This discovery is documented in an October 2017 paper called “A Theory of How Columns in the Neocortex Enable Learning the Structure of the World.”
We realized that everywhere in the neocortex, there is a location signal. Some of the layers in each column are processing information about locations. When sensory input comes in, the column knows where that sensory information is – not relative to you, but relative to the object you are sensing. The column knows both the sensory input and where that sensory input is. As your sensors move, the inputs change, and therefore a single column can learn the structure of complete objects, making it far more powerful than previously thought!
When you understand that a cortical column can create complete models, it forces you to rethink how the neocortex works as a whole. We have more to do, but have made great progress and are moving closer to having a complete framework for how a cortical column works.
Visit numenta.com to learn more about these discoveries and our current research.