Van Hans commited on
Update README.md
Browse files
README.md
CHANGED
|
@@ -16,23 +16,23 @@ The core of the network is a Deep Residual Network (ResNet), a structure well-su
|
|
| 16 |
|
| 17 |
The network employs a robust torso composed of 20 Residual Blocks. Each block contains convolutional layers with skip connections, allowing for the effective learning of deep, hierarchical features and maintaining stable training.
|
| 18 |
|
| 19 |
-
|
| 20 |
|
| 21 |
The entire network processes data using a high number of channels, specifically 512 filters in its main convolutional layers.
|
| 22 |
|
| 23 |
-
|
| 24 |
|
| 25 |
The current board state and history are encoded into a multi-plane tensor with 128 input channels, which typically includes information about piece locations, the player to move, castling rights, and repetition history, a common input format for state-of-the-art chess AI.
|
| 26 |
|
| 27 |
-
|
| 28 |
|
| 29 |
The shared ResNet torso branches into two specialized heads:
|
| 30 |
|
| 31 |
-
|
| 32 |
|
| 33 |
Predicts a probability distribution over the 4672 possible moves (actions) that can be taken. This output is crucial for guiding the Monte Carlo Tree Search (MCTS).
|
| 34 |
|
| 35 |
-
|
| 36 |
|
| 37 |
Outputs a single scalar value, typically between -1.0 (Black is winning) and +1.0 (White is winning). This score represents the network's prediction of the final game outcome from the current position.
|
| 38 |
|
|
|
|
| 16 |
|
| 17 |
The network employs a robust torso composed of 20 Residual Blocks. Each block contains convolutional layers with skip connections, allowing for the effective learning of deep, hierarchical features and maintaining stable training.
|
| 18 |
|
| 19 |
+
* Feature Processing:
|
| 20 |
|
| 21 |
The entire network processes data using a high number of channels, specifically 512 filters in its main convolutional layers.
|
| 22 |
|
| 23 |
+
* Input Representation:
|
| 24 |
|
| 25 |
The current board state and history are encoded into a multi-plane tensor with 128 input channels, which typically includes information about piece locations, the player to move, castling rights, and repetition history, a common input format for state-of-the-art chess AI.
|
| 26 |
|
| 27 |
+
* Dual Output Heads:
|
| 28 |
|
| 29 |
The shared ResNet torso branches into two specialized heads:
|
| 30 |
|
| 31 |
+
* Policy Head (p_logits):
|
| 32 |
|
| 33 |
Predicts a probability distribution over the 4672 possible moves (actions) that can be taken. This output is crucial for guiding the Monte Carlo Tree Search (MCTS).
|
| 34 |
|
| 35 |
+
* Value Head (v):
|
| 36 |
|
| 37 |
Outputs a single scalar value, typically between -1.0 (Black is winning) and +1.0 (White is winning). This score represents the network's prediction of the final game outcome from the current position.
|
| 38 |
|