activation-functions-analysis

Activation Features Empowered: An Efficient Gradient-Free Proxy to Zero-Shot Neural Architecture Search

Implementasi forward pass Multi-Layer Perceptron (MLP) 1 hidden layer dari nol menggunakan numpy murni (tanpa framework deep learning).

TensorFlow/Keras implementation of the trainable Sb-PiPLU activation function and the image-classification experiments associated with the IEEE Access paper.

This project implements the classical LeNet-5 CNN for MNIST digit classification using PyTorch. It covers a complete pipeline from data preprocessing to deployment. The model achieves ~98.8% test accuracy, showing the strong effectiveness of early CNN architectures for image classification.

This folder contains the supporting code and data for arXiv:2607.03664

Systematic study of the Information Bottleneck theory of deep learning; comparing Tanh vs ReLU, SGD vs BGD, and Binning vs KDE mutual information estimation across multiple CNN and feedforward architectures on MNIST.

Ayah Lab is a Deep Learning based Quranic verse theme classification and exploration platform with Streamlit, PyTorch, and Transformer models.

Repository dedicated to analysing and comparing the quality and fidelity of activation function families across wide array of metrics.