Neural Networks

knowledge skill

The agent designs neural network architectures by selecting layer types, activation functions, and hyperparameters. It produces a formal model specification including layer dimensions and connectivity maps.

Worked examples

• {"input": {"input_dimensions": [28, 28, 1], "target_output": "multiclass_classification", "constraints": "low_latency"}, "output": {"architecture_layers": [{"type": "Conv2D", "filters": 32, "kernel": 3, "activation": "relu"}, {"type": "MaxPooling2D", "pool_size": 2}, {"type": "Flatten"}, {"type": "Dense", "units": 10, "activation": "softmax"}], "hyperparameters": {"learning_rate": 0.001, "optimizer": "Adam", "batch_size": 64}, "expected_complexity": 45000}}
• {"input": {"input_dimensions": [100], "target_output": "regression", "constraints": "deep_stack"}, "output": {"architecture_layers": [{"type": "Dense", "units": 64, "activation": "relu"}, {"type": "Dense", "units": 32, "activation": "relu"}, {"type": "Dense", "units": 1, "activation": "linear"}], "hyperparameters": {"learning_rate": 0.01, "optimizer": "SGD", "batch_size": 32}, "expected_complexity": 8401}}

Input

• input_dimensions: integer array (e.g., [224, 224, 3])
• target_output: string (e.g., 'multiclass_classification')
• constraints: string (e.g., 'max_params_1M')

Output

• architecture_layers: array of layer objects (type, units, activation)
• hyperparameters: object (learning_rate, optimizer, batch_size)
• expected_complexity: integer (total_trainable_parameters)

Details

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