Рет қаралды 21,407
We built the world first redstonic convolutional neural network, the task being the recognition of 15×15 hand-written digits. LeNet-5 as its architecture, the network can achieve an accuracy up to 80%. We used an unconventional computational method, the stochastic computing, to realize the network, making it much simpler in design and layout compared to the traditional full-precision computing. The recognition time is 5 minutes per figure theoretically. However, limited by the computational capacity of Minecraft, the real running time exceeds 20 minutes. Nevertheless, it is a breakthrough in redstonic digital circuits, and it may inspire real-world physical neural networks.
Co-authors: Cohomology0(me), lemoon, enadixxoOxoxO, NKID00, 爱红石的小章鱼
First published on bilibili
Access to the savings (java edition):
1.16.4 CNN release drive.google.com/file/d/1yV3-...
1.19 Sculk touchscreen drive.google.com/file/d/1EIyD...
Codes on GitHub: github.com/leamoon/StochasticNet
BGM: Stellaris, and my own piece.