EvoVis Dashboard

Overview

EvoVis is a dashboard for visualizing Evolutionary Neural Architecture Search (ENAS) algorithms. It provides a comprehensive visualization tool that helps researchers and practitioners understand and optimize neural architectures through the evolutionary process.

ENAS mimics biological evolution to automatically discover optimal neural network architectures for specific tasks. The process starts with a population of randomly generated architectures and iteratively improves them through selection, crossover, and mutation.

Key Features

EvoVis offers a holistic view of the ENAS process with the following features:

1. Hyperparameter Overview: Efficiently tune hyperparameters with an organized display of settings
2. Gene Pool Graph: Visualize potential neural architecture topologies and connectivity patterns across generations
3. Family Tree Graph: Navigate the family tree of architectures, analyze performance metrics, and explore structures
4. Performance Plots: Monitor performance trends in real-time with evolutionary metrics
5. Data Structure Interface: Support for visualizing ENAS results from various frameworks and sources

Integration with EdgeVolution

EvoVis can be used in combination with EdgeVolution to visualize the evolutionary processes and results. This integration allows users to gain deeper insights into the neural architecture search process, better understand the evolution of model architectures, and make more informed decisions when optimizing neural networks for edge devices.

Getting Started

To use EvoVis with your ENAS results:

1. Clone the EvoVis repository
2. Install dependencies
3. Run EvoVis by specifying your run results directory path
4. Access the dashboard via localhost

For detailed instructions and compatibility information, refer to the EvoVis GitHub repository.