Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Gradual Drift Detection
Technique for identifying progressive changes in data or concepts, enabling anticipatory adaptation before the model's performance degrades significantly.
Unsupervised Change Detection
Methods identifying changes in data distributions without using predefined labels, based on statistical metrics or divergence between time periods.
Stability-Plasticity
Fundamental dilemma in incremental learning seeking to balance the model's ability to retain prior knowledge (stability) while adapting to new information (plasticity).
Catastrophic Forgetting
Phenomenon where a model learning new information completely or partially loses previously acquired knowledge, a major problem in incremental learning.
Short-Term Memory
Temporary storage mechanism for recent observations used to rapidly detect changes and adapt the model before their permanent integration into long-term memory.
Active Instance Selection
Strategy selectively choosing the most informative instances for model update, optimizing computational efficiency while maintaining relevance in the face of changes.
Abrupt Drift Detection
Technique specialized in identifying sudden and significant changes in data, requiring rapid model adaptation to avoid catastrophic performance degradation.
Dynamic Re-evaluation
Continuous process of evaluating model performance on new data to determine when adaptation is necessary, based on dynamic thresholds or degradation metrics.
DDM Test
Drift Detection Method, statistical algorithm monitoring the model error rate to detect significant changes, based on statistical controls of means and variances.
Block Adaptation
Model update strategy using batches of data rather than individual instances, providing a trade-off between reactivity to changes and computational efficiency.