Deep Learning

knowledge skill

The agent designs and implements neural network architectures to solve pattern recognition tasks. It produces executable model definition code and hyperparameter configurations for training on specific datasets.

Worked examples

• {"input": {"dataset_description": "28x28 grayscale images of handwritten digits", "model_type": "CNN", "performance_metric": "Accuracy"}, "output": {"model_code": "class Net(nn.Module): ... conv1 = nn.Conv2d(1, 32, 3) ...", "hyperparameters": {"learning_rate": 0.001, "optimizer": "Adam", "batch_size": 64}, "layer_architecture": ["Conv2d(1, 32, 3)", "MaxPool2d(2)", "Linear(128, 10)"]}}
• {"input": {"dataset_description": "Sequential stock price time-series data", "model_type": "LSTM", "performance_metric": "RMSE"}, "output": {"model_code": "model = Sequential([LSTM(50, input_shape=(seq_len, n_features)), Dense(1)])", "hyperparameters": {"learning_rate": 0.01, "optimizer": "RMSprop", "batch_size": 32}, "layer_architecture": ["LSTM(50)", "Dropout(0.2)", "Dense(1)"]}}

Input

• dataset_description: string: features and target labels
• model_type: string: e.g., CNN, Transformer, LSTM
• performance_metric: string: e.g., F1-score, MSE

Output

• model_code: string: Python/PyTorch/TensorFlow implementation
• hyperparameters: object: learning rate, batch size, optimizer
• layer_architecture: array: sequence of layer types and dimensions

Details

• Skill type: knowledge skill
• Safety level: safe_public_research
• Version: 1.0.0