I made an AI, and I gave it a voice. Someone trained a voice text to speech model off of recordings of HAL speaking in the movie. And so, I gave HAL access to my file system and now you can run the entire computer with HAL. And I made this wallpaper. I plan on building a physical model based off of this.
Hello Everyone, I'm a high school student interested in Neural Networks. I've been doing quite a bit of research on the subject and I'm working now on creating a Neural Network AI which can be trained to do any number of tasks such as multiplication or addition. I have the basic principle of a neuron already coded and I have 1000 neurons, each neuron processes a different part of the training data. On the Interface sheet you input X1 and X2 and you can input the actual value but it's not necessary. The goal is to have it output the answer to whatever your input values are based on the training data. In the Neural_Net Sheet the first neuron (row under the top two label rows) handles the input you can change, the rest loads the training data from the Interface sheet. If I'm right, it should be able to accomplish this if I create more iterations of the weight/bias updates? And is there any way I can condense the number of iterations necessary to complete the problem provided in the input? I thought maybe I could increase delta in the gradient calculations; I had delta set to 0.01 but I changed it to 1 to see what happened and the value of Loss decreased more in the next iteration. I'd appreciate any help, and please remember that I have limited knowledge on this topic and I have not taken math past algebra. Also, I'm highly skilled in spreadsheets, if you are wondering why I am using a spreadsheet over some other means.
This is a link to my project, please feel free to comment inside and leave tips on how to fix any problems I may have that I do not see.
