eli5:What are weights in nueral networks and how do they work
In: 2
Every node in the network has some output depending on what inputs it receives. That output is forwarded to some number of other nodes, and is multiplied by the weight. I.e. if a single node is outputting 1 for example, but the weight for the connection from that node to another node is 0.3, then the input that other node receives from that connection is 0.3. So essentially it changes how important the output of one node is for the effect it has on a specific other node.
Networks are made by nodes connected to one another. The weights of each connection are how “important” those connections are.
For example, let’s say we want to make a network to determine if a person is obese. Our three input nodes (i.e. information we have) are age, height, and weight. Now, height and weight are important to determine obesity, but age isn’t that important (a 300lb 20 year old is just as obese as a 50 year old). Therefore, all of the weights of connections coming from the age node are going to be very small.
On the other hand, height and weight are both quite important, so the weights of those connections will likely be quite high (though height may be a high negative because a taller person at the same weight is going to be skinnier).
Did that make sense or was it confusing?
Latest Answers