An innovative optimization of a computational model through the analogue of a chess engine, integrating NNUE (Efficiently Updatable Neural Network) evaluation with the Nega-Max search algorithm.
Our developed chess engine demonstrated comparably superior performance to the most powerful open-source chess engine, Stockfish v16.0, through intelligent evaluation and adaptive depth heuristics.
🏛️ Published at the IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT), hosted at the Indian Institute of Science (IISc), Bengaluru.
View the IEEE Paper →
This project focuses on enhancing classical search-based chess engines with a neuro-adaptive evaluation mechanism inspired by NNUE.
Unlike traditional handcrafted evaluation functions, NNUE learns optimal board representations, allowing the engine to assess positions with greater accuracy and computational efficiency.
- Bridge the gap between symbolic (rule-based) and connectionist (neural network) paradigms.
- Leverage Nega-Max for a cleaner, recursive minimax framework.
- Optimize evaluation speed and decision accuracy under real-time constraints.
- Uses incrementally updated features to efficiently re-evaluate board states.
- Reduces redundant computations during move generation.
- Trained using high-quality datasets of chess positions and engine evaluations.
- Simplifies the Minimax approach using symmetry in game theory.
- Incorporates alpha-beta pruning for enhanced search efficiency.
- Integrates dynamic depth adjustment based on position complexity.
- Combines neural evaluation and heuristic-based scoring.
- Implements adaptive weighting to balance speed and precision.
- Language: C++ / Python (depending on implementation layer)
- Frameworks: PyTorch / NumPy
- Dataset: Self-generated and engine-annotated positions
- Evaluation Metrics: Win rate, move accuracy, time per move
- Efficient move ordering and pruning
- Incremental neural evaluation
- Parallelized search with dynamic time management
- Modular design for training new NNUE models
- Configurable engine parameters for experimentation
IEEE Citation:
A. M. Cherian, “Implementing the Chess Engine using NNUE with Nega-Max Algorithm,”
2024 IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT), IISc Bengaluru, India, 2024.
DOI: 10.1109/CONECCT10677087
- Integration of Reinforcement Learning (RL) for self-play training.
- Incorporation of Monte Carlo Tree Search (MCTS) hybridization.
- Implementation of hardware-optimized NNUE inference (FPGA/Edge devices).
Special thanks to the IEEE CONECCT Conference Team and the Indian Institute of Science (IISc) for providing a platform to present this research.
Author: Aaron Mano Cherian
LinkedIn: linkedin.com/in/aaronmanocherian
Email: aaron.cherian@columbia.edu
Publication: IEEE Xplore Link
⭐ If you found this project interesting, consider giving it a star!