AI

ESIA

School of Artificial Intelligence

Worldwide cohort
Students across time zones
HomeMaster’sTrainingsProjectsResearchBlogAboutContact
← Back to Trainings

Deep Learning & Computer Vision (Practice)

Train CNNs with strong discipline: augmentation, imbalance, transfer learning, and reliable training/debugging patterns.

Deep LearningComputer VisionCNNPyTorch/KerasAugmentation

Duration

4 days

Format

Remote

Level

Intermediate

Key outcomes

  • Build and train CNN models with proper evaluation
  • Use transfer learning efficiently for business datasets
  • Handle class imbalance and noisy labels
  • Improve robustness with augmentation and monitoring

Syllabus

Day 1 — Training fundamentals

  • Loss, optimizers, schedules, early stopping
  • Regularization: weight decay, dropout
  • Debugging training and monitoring curves

Day 2 — CNNs & transfer learning

  • Convolutions, pooling, normalization
  • Pretrained backbones and fine-tuning strategy
  • Feature extraction vs full fine-tuning

Day 3 — Real-world issues

  • Augmentation strategies
  • Imbalance: sampling, focal loss (concept), thresholds
  • Error analysis: confusion matrix, failure clusters

Day 4 — Delivery patterns

  • Exporting models and inference basics
  • Batch inference vs real-time constraints
  • Reproducibility, seeds, and experiment tracking