AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
RoBERTa (A Robustly Optimized BERT Pretraining Approach)
Optimized version of BERT that modifies training hyperparameters and removes the Next Sentence Prediction task, resulting in better performance on many NLU tasks.
Downstream Task
Specific machine learning task (such as NER, text classification, etc.) for which a pre-trained model is adapted via fine-tuning.
Classification Head
Final layer added to a pre-trained model during fine-tuning, specifically designed to map contextual representations to outputs of a classification or sequence labeling task.
Sequence Labeling
Type of NLP task where each token in an input sequence is assigned a label, as in NER where tokens are labeled as being part of an entity or not.
BIO Labeling Scheme
Labeling convention for NER where each token is marked as Beginning (start of an entity), Inside (inside an entity), or Outside (outside an entity), to handle multi-token entities.
Masked Language Modeling
Pre-training objective used for models like BERT, where a certain percentage of input tokens are masked and the model learns to predict them based on context.
Hugging Face Transformers
Open-source library that provides thousands of pre-trained models (BERT, RoBERTa, etc.) for natural language processing, facilitating their use and fine-tuning.
F1-Score
Evaluation metric that represents the harmonic mean of precision and recall, providing a balance between the two to measure the performance of a classification model.