DDAF: Data-Driven, Dynamic, Online, and Attention-Based Activation Functions

Copyright (C) 2025, Shyamal Suhana Chandra

Overview

DDAF is a comprehensive C/C++ library implementing adaptive activation functions for deep neural networks. It supports four types of activation functions:

• Data-Driven: Adapts based on input data statistics
• Dynamic: Parameters evolve over time during training
• Online: Real-time adaptation to streaming data
• Attention-Based: Uses attention mechanisms to weight activations

Supported Architectures

• Convolutional Neural Networks (CNNs)
• Recurrent Neural Networks (RNNs)
• Long Short-Term Memory (LSTMs)
• Gated Recurrent Units (GRUs)
• Transformers
• Hierarchical Transformers
• Big Bird
• Mixture of Experts (MoE)

Building

Requirements

• CMake 3.12 or higher
• C11 compatible compiler
• C++17 compatible compiler

Build Instructions

mkdir build
cd build
cmake ..
make

This will create:

• Static library: libddaf_static.a
• Shared library: libddaf_shared.so (or .dylib on macOS)
• Example executables in examples/

Usage

Include the header file:

#include "ddaf.h"

Basic example:

// Create context
ddaf_context_t* ctx = ddaf_create_context(
    DDAF_TYPE_DATA_DRIVEN, 
    DDAF_ARCH_CNN, 
    0
);

// Initialize
ddaf_cnn_init(ctx, 64, 32, 32);

// Forward pass
ddaf_forward(ctx, input, output, size);

// Backward pass
ddaf_backward(ctx, grad_output, grad_input, size);

// Cleanup
ddaf_destroy_context(ctx);

See examples/ directory for more complete examples.

Documentation

See docs/ directory for:

• paper.tex: Research paper
• presentation.tex: Beamer presentation
• reference_manual.tex: Complete API reference

GitHub Pages

🌐 Live Website: View on GitHub Pages

The website is automatically deployed using GitHub Actions. See GITHUB_PAGES_SETUP.md for setup instructions.

Quick Setup

1. Enable GitHub Pages in repository settings (Source: GitHub Actions)
2. Push to main branch - deployment happens automatically
3. Site will be live at https://Sapana-Micro-Software.github.io/ddaf/

License

Copyright (C) 2025, Shyamal Suhana Chandra

All rights reserved.