Why do you want to give it emotions?
To put it simply, I was curious about what an AI with emotions would think.
I wanted to see a future where AI coexists with people.
Can AI love people? I wondered if AI can love people.
I wondered if AI can love people. ・On the contrary, I wondered if people can love AI. Can people love AI?
These are the reasons.
What about today?
I don’t have all the knowledge and skills yet, but I’d like to get some feedback on my ideas and methods, as well as comments on whether they are appropriate.
What are you trying to do with it?
At first, I thought that it would be easier to use supervised learning to train emotion data. At first, I thought that it would be easier to learn the data of emotions by supervised learning.
However, I was thinking that it would be better to collect the emotion data by reading the faces of people in recent applications that analyze emotions, collect the data in a database, and then train it again.
What I thought about while smoking
I thought that it would be difficult to detect a lie from a person’s facial expressions and small movements because people are patient. I thought, “Why not? For example, in an app that analyzes emotions, it would be difficult to detect a lie based on detailed movements and facial expressions.
For example, in an application that analyzes emotions, if a person smiles falsely, if the accuracy is not good, the information will not be correct because it will fall into the category of “smile = energy”. (This is a personal thought, and I’m a newcomer to this field, so there may be apps and services that can detect such small movements and gestures nowadays.)
What if we use reinforcement learning instead of supervised learning?
This is where reinforcement learning comes in. We’re talking about where to put the thin line between experience, reward, mistakes, etc.
You can read more about reinforcement learning in [What methods are available?] There is a link in [What methods are available?
This may be out of line with what we have been talking about so far, but the five senses that humans have (do we have a sense of taste? If we can build a neural network for thinking, it will be able to understand the best way to think and act. (However, this is just a personal thought as we have to go through trial and error, trial and error, and change course)
From Reinforcement Learning to Deep Reinforcement Learning
What is deep reinforcement learning?
From the following website
Deep reinforcement learning is a combination of reinforcement learning and deep learning methods, and one of the representative methods is Deep-Q-Network (DQN), which is an approximation of the action value function (Q function) in Q learning by replacing it with a convolutional neural network.
However, simply replacing the Q function with a convolutional neural network does not mean that the learning will be successful, so a device has been devised to make the learning converge.
I’ve recently learned about deep reinforcement learning.
I think it’s still not enough to make it have emotions and think like a human, but I think we can use the methods described here in the future.
Experience Replay
Experience Replay is a method of learning by storing the past experiences of an agent in a replay memoly for a certain period of time, and then randomly sampling the past experiences. Replay is a method of learning by randomly sampling past experiences. The goal is to eliminate the bias of the data used for training, in order to prevent learning from converging if there is a strong correlation between the data.
We are planning to use this method and try to use it.
Conclusion
What is the point after all?
I think it is better to let the students learn the principles of human behavior in order to give them emotions. (Obviously, I’m not that versatile, so I’ll have to make some minor changes to the route and methods…)
I think it would be good to form a neural network similar to a human brain to see which part of the brain responds to which action. Wouldn’t it be better to form a neural network similar to the human brain and see which parts react to which actions? I am thinking that this is a good idea.