AlirezaMahDev.Extensions.Brain.Abstractions 0.0.1-alpha7

AlirezaMahDev.Extensions.Brain.Abstractions

Project Description

This project defines the core abstractions and interfaces for the Brain extension, providing a contract-based approach to neural network operations. It serves as the foundation for the Brain implementation, enabling loose coupling and testability.

Dependencies

Project Dependencies

• AlirezaMahDev.Extensions.DataManager.Abstractions: Provides data management abstractions for neural network persistence

Core Interfaces

INerve<TData>

The central interface representing a neural network, providing methods for training and inference:

• Learn: Trains the network with provided data
• Think: Performs inference on input data
• Sleep: Optimizes the network state
• Name: Identifies the nerve instance
• Root: Provides access to the root connection

INeuron<TData>

Represents a single neuron in the network:

• Offset: Memory offset of the neuron
• RefValue: Direct reference to the neuron's value
• RefData: Direct reference to the neuron's data
• Connection: Gets or creates connections to other neurons

IConnection<TData>

Manages connections between neurons:

• Location: Data location of the connection
• Offset: Memory offset of the connection
• Neuron: Connected neuron
• Previous/Next: Navigation through connection chains

IBrainService

Service interface for managing neural network instances:

• GetOrAdd<TData>: Retrieves or creates a named neural network

Data Structures

NeuronValue<TData>

Structure representing a neuron's state:

• Data: The actual data stored in the neuron
• Score: Activation score
• Weight: Connection weight
• Connection: Reference to connected neurons

ConnectionValue

Structure representing a connection between neurons:

• Next: Reference to the next neuron in the connection chain
• Weight: Strength of the connection

Design Principles

1. Type Safety: Strongly-typed generic interfaces
2. Separation of Concerns: Clear distinction between different neural network components
3. Memory Efficiency: Uses value types and references where appropriate
4. Extensibility: Designed to support various neural network architectures

Usage Example

// Get or create a neural network
var nerve = brainService.GetOrAdd<float[]>(networkName);

// Train the network
var trainingData = GetTrainingData();
nerve.Learn(trainingData);

// Perform inference
var input = new float[] { /* input data */ };
var result = nerve.Think(input);

Architecture

The project follows a clean architecture with:

• Abstractions Layer: Defines the contracts for neural network operations
• Data Structures: Provides the core data types used by the implementation
• Service Contracts: Defines the service interfaces for neural network management

This design allows for different implementations while maintaining a consistent API for consumers.