This repository provides a collection of Microsoft Excel workbooks that implement various deep learning concepts from scratch. It targets educators, students, and practitioners who want to understand the inner workings of neural networks through interactive, visual examples without requiring programming expertise. The benefit is a hands-on, intuitive grasp of complex AI algorithms.

How It Works

The project leverages Excel's built-in functions, formulas, and VBA scripting to simulate neural network operations. Each workbook focuses on a specific architecture, breaking down concepts like matrix multiplication, activation functions (Softmax, LeakyReLU), backpropagation, and advanced models like Transformers and LSTMs into manageable, transparent Excel sheets. This approach demystifies AI by making the mathematical and logical flow explicit and editable.

Quick Start & Requirements

• Install: Download and open the relevant .xlsx workbook files.
• Prerequisites: Microsoft Excel (version not specified, but newer versions with VBA support are recommended).
• Setup: Minimal; requires only Excel.
• Links: ai-by-hand-excel

Highlighted Details

• Covers a wide range of fundamental to state-of-the-art AI models, including MLP, RNN, LSTM, ResNet, Transformer, and Mamba.
• Implements core components like Softmax, LeakyReLU, backpropagation, and attention mechanisms.
• Offers a visual and interactive learning experience directly within Excel.
• Includes a "Coming Soon" section for GANs, VAEs, U-Net, and CLIP.

Maintenance & Community

The project appears to be actively developed, with a "🔥 NEW" tag indicating recent additions like DeepSeek and Mixture of Experts. No specific community links (Discord, Slack) or contributor details are provided in the README.

Licensing & Compatibility

The repository does not explicitly state a license. Users should assume all rights are reserved or contact the maintainers for clarification on usage, especially for commercial purposes.

Limitations & Caveats

Excel's computational limitations may restrict the scale and complexity of models that can be practically simulated. Performance will be significantly slower than code-based implementations. The reliance on specific Excel versions or features could lead to compatibility issues.