NEXTorch

NEXTorch implements Bayesian optimization in Python/PyTorch to guide experimental design and select experiments that minimize time, materials, and computational resources for automating and optimizing chemical systems.

Key Features:

• State-of-the-Art Bayesian Optimization Algorithms: Implements Bayesian optimization as a sequential global optimization method suitable for automated and human-in-the-loop optimization.
• Data-driven Active Learning: Supports data-driven active learning algorithms to iteratively select informative experiments.
• Fast Predictive Modeling: Provides rapid predictive modeling capabilities for iterative experimental design workflows.
• Flexible Optimization Loops: Enables customizable optimization loops to adapt optimization strategies to specific research needs.
• Parameter and Data Type Conversions: Supports multiple parameter types and versatile data type conversion options for diverse experimental setups.
• GPU Acceleration and Parallelization: Utilizes GPU acceleration and parallel processing to reduce computation time for large-scale problems.
• Visualization Capabilities: Includes visualization tools to aid interpretation of complex data and optimization results.
• Interfacing with Legacy Software: Provides interfaces to integrate with existing software systems.

Scientific Applications:

• Catalyst Synthesis: Guides experimental selection and optimization in catalyst development workflows.
• Reaction Condition Optimization: Optimizes reaction conditions to improve yields, selectivity, or efficiency.
• Parameter Estimation: Supports parameter estimation tasks by selecting experiments that reduce uncertainty in model parameters.
• Reactor Geometry Optimization: Applies optimization methods to the design and tuning of reactor geometries.

Methodology:

Implements Bayesian optimization (sequential global optimization) and data-driven active learning using Python and PyTorch; supports predictive modeling, customizable optimization loops, parameter/data type conversions, visualization, GPU acceleration, parallel processing, and interfaces to legacy software.