Neural Network Learning Dynamics

import matplotlib.pyplot as plt
import numpy as np
from scipy import stats

np.random.seed(123)

# Training data
epochs = np.arange(1, 101)
loss = 2.5 * np.exp(-0.05 * epochs) + 0.15 + np.random.normal(0, 0.08, 100)

# Polynomial fit
z = np.polyfit(epochs, loss, 4)
p = np.poly1d(z)
x_smooth = np.linspace(1, 100, 200)

fig, ax = plt.subplots(figsize=(10, 7), facecolor='#0a0a0f')
ax.set_facecolor('#0a0a0f')

# Confidence region
ax.fill_between(x_smooth, p(x_smooth) - 0.12, p(x_smooth) + 0.12,
                color='#6CF527', alpha=0.12, linewidth=0)

# Fit line
ax.plot(x_smooth, p(x_smooth), color='#6CF527', linewidth=2.5, zorder=3)

# Data points
ax.scatter(epochs, loss, c='#F5B027', s=45, alpha=0.7, edgecolors='none', zorder=4)

# Convergence annotation
ax.axhline(y=0.15, color='#F5276C', linestyle='--', linewidth=1.5, alpha=0.6)
ax.text(95, 0.18, 'convergence', fontsize=9, color='#F5276C', ha='right')

for spine in ['top', 'right']:
    ax.spines[spine].set_visible(False)
for spine in ['bottom', 'left']:
    ax.spines[spine].set_color('#333333')

ax.set_xlabel('Training Epoch', fontsize=12, color='white', fontweight='500', labelpad=10)
ax.set_ylabel('Validation Loss', fontsize=12, color='white', fontweight='500', labelpad=10)
ax.set_title('Model Convergence Analysis', fontsize=15, color='white', fontweight='bold', pad=20, loc='left')
ax.tick_params(colors='#666666', labelsize=10, length=0)

plt.tight_layout()
plt.show()