maximuspowers/muat-fourier-5-medium-classifier

Pattern Classifier

This model was trained to classify which patterns a subject model was trained on, based on neuron activation signatures.

Dataset

• Training Dataset: maximuspowers/muat-fourier-5-medium
• Input Mode: signature
• Number of Patterns: 14

Patterns

The model predicts which of the following 14 patterns the subject model was trained to classify as positive:

1. palindrome
2. sorted_ascending
3. sorted_descending
4. alternating
5. contains_abc
6. starts_with
7. ends_with
8. no_repeats
9. has_majority
10. increasing_pairs
11. decreasing_pairs
12. vowel_consonant
13. first_last_match
14. mountain_pattern

Model Architecture

• Signature Encoder: [512, 256, 256, 128]
• Activation: relu
• Dropout: 0.2
• Batch Normalization: True

Training Configuration

• Optimizer: adam
• Learning Rate: 0.001
• Batch Size: 16
• Loss Function: BCE with Logits (with pos_weight for training, unweighted for validation)

Test Set Performance

• F1 Macro: 0.2624
• F1 Micro: 0.2737
• Hamming Accuracy: 0.7323
• Exact Match Accuracy: 0.0148
• BCE Loss: 0.4683

Per-Pattern Performance (Test Set)

Pattern	Precision	Recall	F1 Score
palindrome	14.2%	76.4%	24.0%
sorted_ascending	53.8%	46.7%	50.0%
sorted_descending	13.4%	96.2%	23.5%
alternating	18.6%	80.7%	30.2%
contains_abc	17.8%	84.6%	29.4%
starts_with	9.9%	87.1%	17.9%
ends_with	39.7%	78.9%	52.8%
no_repeats	13.7%	63.0%	22.4%
has_majority	0.0%	0.0%	0.0%
increasing_pairs	22.9%	45.8%	30.6%
decreasing_pairs	14.3%	92.9%	24.8%
vowel_consonant	0.0%	0.0%	0.0%
first_last_match	26.7%	45.1%	33.5%
mountain_pattern	16.8%	89.6%	28.3%