Every MNIST story collected by The AI Daily, 2 so far, newest first, refreshed hourly.
MLP built from scratch in pure NumPy with GUI showing training internalsA developer released an educational tool implementing a small MLP with manual backprop in pure NumPy, reaching about 98.5% on MNIST. During training it shows per-layer gradient norms, inactive neuron percentages, weight distributions, layer-by-layer t-SNE, and supports neuron ablation and pruning experiments.
Framework-free prototype learner Jayce trains 1.6x-4x faster than backpropA developer released Jayce, an experimental learner using Adaptive Prototype Memory that stores an LLM's raw context vectors in 4,096 prototype slots and shifts the nearest prototype to learn or correct facts without touching model weights. It reports training updates 1.6x-4x faster than Adam backprop, higher accuracy than backprop at equal sample counts on tasks like MNIST, and runs offline on a local Qwen3-4B GGUF using only NumPy and Java.
The A.I. Daily