Progressive Learning
Chess Engine
A hybrid Bayesian-LSTM architecture with curriculum learning,
spaced repetition, and Pavlovian conditioning
Key Features
Hybrid Architecture
Combines Bayesian networks for probabilistic reasoning with LSTM networks for sequential pattern recognition
Curriculum Learning
Progressive difficulty from preschool basics to infinite chess variants with mastery-based advancement
Spaced Repetition
SM-2 algorithm for long-term memory retention with adaptive review intervals
Pavlovian Conditioning
Rescorla-Wagner model for reward-based learning and move evaluation
Chess Engine
Full position evaluation, move prediction, and minimax search capabilities
Multi-Agent Framework
Extensible to football, basketball, baseball, hockey, soccer, and tennis
Architecture
System Components
- Neural Network: Hybrid Bayesian + LSTM layers Processes 8×8×12 board matrices
- Curriculum System: 10 difficulty levels 85% accuracy threshold for advancement
- Spaced Repetition: Adaptive review scheduling Exponential interval spacing
- Pavlovian Learner: CS-US associations Rescorla-Wagner model updates
- Training Engine: Orchestrates all learning Supports SGD, Adam, Adagrad, RMSprop
- Inference Engine: Position evaluation & move prediction Minimax search with configurable depth
About
The Progressive Learning Chess Engine is a research project that combines multiple learning paradigms to create a chess engine that learns progressively, similar to how humans learn chess—starting with basic concepts and gradually advancing to complex strategies.
The system implements a hybrid neural network architecture combining Bayesian networks for probabilistic reasoning with LSTM networks for sequential pattern recognition. Curriculum learning guides the training process through 10 difficulty levels, while spaced repetition ensures long-term memory retention. Pavlovian conditioning enables reward-based learning for move evaluation.
Technology Stack
Copyright (C) 2025, Shyamal Suhana Chandra. All rights reserved.