4x4 Image Recognizer

This circuit is a full-blown AI that can recognize a drawn square or circle on a 4x4 binary image. To use, activate the switches corresponding to the pixels on the display by holding right mouse button over them (which should look like blue lines to the left of the screen as the face of the switches are hidden) and hovering over the "ON" or "OFF" state. Pay attention to the "confidence meters" shown above a GSO battery (circle) and a Space Junkers one-block (square) for the AI's predictions. This circuit uses large amounts of blocks and circuitry, so it may cause lag on low-end CPUs. It also utilizes transmitters and receivers, so it may interfere with or be interfered by other techs with receivers and/or transmitters. If unsure, remove all other techs before deploying. The weights and biases were obtained from a model trained using TensorFlow's Keras API, using the binary cross-entropy loss function and the Adam optimizer, which were then rounded to the nearest thousandths for ease of implementation. It is powered by a feed-forward neural network comprised of 2 ReLU neurons in the first/input layer and 2 hard sigmoid neurons in the second/output layer, with 36 weights and 4 biases adding up to a total of 40 parameters. Hard sigmoid was used due to its simplicity to implement in TerraTech's C&S system. It has been trained off a dataset with 80 samples, ensuring high accuracy on most reasonable inputs.