NeuralNetworks - 9th Semester NTUA

Lab 1 - Supervised Learning

Part 1

Dataset: Statlog (Vehicle Silhouettes)

• Classifiers: Dummy, GNB, kNN, LR
• Metrics: Accuracy, F1
• 10-fold cross validation
• Out-of-the-box evaluation
• Parameter fine tuning

Part 2

Dataset: 2019 database of road traffic injuries

• Classifiers: MLP, SVM
• Metrics: F1 Macro
• 5-fold cross validation
• Out-of-the-box evaluation
• Parameter fine tuning using Optuna, improving performance using Pipelines

Lab 2 - Unsupervised Learning

Dataset: Carnegie Mellon Movie Summary Corpus

• Movie Recommender using TFIDF/Cosine similarity vs W2Vec/Cosine similarity and evaluation
• SOM mapping and kNN clustering

Lab 3 - Deep Learning

Improving Image captioning with visual attention of TensorFlow

• Using BLEU score evaluation
• Choosing a more suitable CNN
• Better caption preprocessing
• Integrating word embeddings
• Utilizing beam search
• Fine tuning Transformer parameters