In this blog, I will consider the difference between learning in the human brain and learning in machine learning.
- Learning in the human brain
- Learning in machine learning
- in conclusion
How does learning take place in the human brain?
Our memories are thought to be stored in the cerebral cortex. Figure 1 shows the process of creating a new memory in the cerebral cortex. A part of the brain called the hippocampus plays an important role in making memories. The hippocampus stores sensory information such as vision and smell. The collected sensory information is written by the hippocampus into the cerebral cortex. This is memory. The hippocampus is also required for memory retrieval.
However, after a certain period of time (up to several months), it becomes possible to read out memories in the cerebral cortex without the help of the hippocampus. In addition, it is thought that memories that cannot be expressed in words, such as how to move the body, are sometimes stored in the cerebellum.
In areas of the brain that control memory, such as the hippocampus and cerebral cortex, neurons form a complex network, as shown in Figure 2. The important thing here is that specific memories are not stored in individual neurons. In modern science, the network structure itself, such as how neurons are connected and the size of their joints, is considered to be memory.
The connections between neurons are called synapses. When the brain memorizes a certain thing, some synapses become larger or new synapses are formed. Conversely, synapses may become smaller or disappear. An example of synaptic changes is shown in Figure 3.
The synaptic state is basically not fixed. Repeated learning of the same thing enlarges synapses and strengthens connections between neurons. When this happens, it becomes difficult to forget memories. Conversely, if you do not learn, the synapses become smaller and the connections between nerve cells weaken. When this happens, you forget what you have learned.
Synapses have the property of being flexible in this way, and this property is called synaptic plasticity.
How does machine learning work? Here, we introduce the learning principle of neural networks, which is a representative method of machine learning.
A neural network is a technique that was devised with reference to neurons in the human brain. An example of a neural network is shown in Figure 4. The circles in this diagram are called nodes, which correspond to nerve cells in the human brain. Nodes are connected by lines. These lines are called edges, and each edge usually has a variable called a weight.
During the learning process, the weights are iteratively adjusted to their optimal values. The optimized weights correspond to memory in the human brain, but they don’t change like synapses in the human brain. In other words, the weights in neural networks are fixed, not strengthened and faded like human memory. Of course, by devising a way to make it learn, it is possible to artificially achieve something equivalent to that, but even if it does, it will not be the same as the mechanism of the human brain.
In the human brain, it happens that we forget what we have learned. Forgetting is not necessarily a negative thing, it leads to remembering only the necessary information. From this, it can be said that the human brain is doing information processing with very good cost performance.
On the other hand, neural networks basically do not forget what they have learned once. If you want to forget what you learned, you need to do additional learning with new data.
This is the big difference between learning in the human brain and learning in machine learning. The human brain has a mechanism for naturally forgetting things that are not important, but current neural networks do not have such a mechanism. In the future, if a mechanism for “naturally forgetting” is realized in neural networks, it may be possible to learn more like humans.
At Skill Up AI, we are currently offering a related course, ” Basic course for machine learning and data analysis that can be used in the field .”
It is designed to start with an introduction to machine learning and gradually learn the core of each algorithm. Please consider taking the course.
In addition, we hold a practical AI study session ” Skill Up AI Camp ” every Wednesday. At the study sessions, we will cover various practical themes and provide hints that will lead to improved practical skills in data analysis and AI development. There is also a corner where the instructor answers questions and concerns from the participants.