Machine Unlearning Goes Backwards

Folks this is the first neural network learning backwards, I just invented machine unlearning 😭 for(int i = 0; i < MINI_BATCH_SIZE; i++) backward(&net->output, hidden_outp } PROBLEMS OUTPUT DEBUG CONSOLE TERMINAL Epoch 1, Accuracy: 86.82%, Time: 3.30 seconds Epoch 2, Accuracy: 87.07%, Time: 3.27 seconds Epoch 3, Accuracy: 85.54%, Time: 3.27 seconds Epoch 4, Accuracy: 83.59%, Time: 3.27 seconds Epoch 5, Accuracy: 81.48%, Time: 3.27 seconds Epoch 6, Accuracy: 77.88%, Time: 3.27 seconds Epoch 7, Accuracy: 76.62%, Time: 3.27 seconds Epoch 8, Accuracy: 73.50%, Time: 3.27 seconds Epoch 9, Accuracy: 72.48%, Time: 3.33 seconds Epoch 10, Accuracy: 71.82%, Time: 3.30 seconds Epoch 11, Accuracy: 70.26%, Time: 3.27 seconds Epoch 12, Accuracy: 68.77%, Time: 3.27 seconds Epoch 13, Accuracy: 65.82%, Time: 3.27 seconds Epoch 14, Accuracy: 66.50%, Time: 3.28 seconds Epoch 15, Accuracy: 66.03%, Time: 3.27 seconds Epoch 16, Accuracy: 62.49%, Time: 3.27 seconds Epoch 17, Accuracy: 61.58%, Time: 3.27 seconds Epoch 18, Accuracy: 60.02%, Time: 3.27 seconds Epoch 19, Accuracy: 59.62%, Time: 3.27 seconds Epoch 20, Accuracy: 57.97%, Time: 3.28 seconds

A social media post claims to show the first neural network learning backwards and jokingly calls the process “machine unlearning.” The code editor below shows training accuracy steadily decreasing over successive epochs, making ordinary deterioration look like a deliberate reverse-learning breakthrough. It is used as a reaction to jokingly describe forgetting, making progress in reverse, or confidently presenting a bug as an invention.

• @ithueti / 14809 2024-12-14