NeuralVizVR

Features

Multi-Modal Access

Available on VR, desktop, and web platforms for accessibility across devices.

Immersive Visualization

Explore neural network layers, activations, and data flow in a 3D environment.

Interactive Construction

Construct, modify, and analyze neural networks in real time through intuitive 3D interactions.

Interpretability Tools

Gain insights into feature influences on predictions with integrated tools.

Robustness Analysis

Test model resilience to various attack types within the immersive environment.

AI-Assisted Analysis

Receive real-time explanations and guidance from integrated LLM-based AI assistance.

About NeuralVizVR

NeuralVizVR is an innovative, open-source platform developed at Maynooth University, funded by the SFI Frontiers for the Future Program (MAIVV, 2022–2025). Our mission is to improve neural network interpretability and analysis by leveraging immersive 3D visualization, interactive exploration, and advanced AI-assisted techniques, making complex models more accessible and understandable to a wide range of users.

Central Research Question

How can immersive 3D visualization, interaction, and AI-assisted analysis enhance the interpretability, accessibility, and validation of neural networks across diverse modalities and applications?

Objectives

Develop a Multi-Modal Platform: Create a platform accessible via web, desktop, and VR modes, ensuring usability across expertise levels and hardware configurations.
Enhance Visualization and Interaction: Provide tools for constructing, modifying, and analyzing neural networks as 3D representations, enabling real-time interaction and experimentation.
Integrate Formal Verification Methods: Validate model robustness, reliability, and adherence to criteria through formal verification methods.
Integrate AI Assistance: Enhance LLM capabilities for advanced model interpretation, personalized guidance, and intuitive support across expertise levels.
Develop Autonomous Systems Simulation: Introduce a virtual autonomous vehicle simulation for real-world-inspired scenarios for data collection, model training, and testing.

Medet Inkarbekov

Postdoctoral Researcher

Barak A. Pearlmutter

Professor

Rosemary Monahan