Glossary
AI and data science terms in English and Vietnamese.
- Accuracy · Độ chính xác
- The share of predictions that are correct.
- Artificial intelligence · Trí tuệ nhân tạo
- Systems that perform tasks that normally need human intelligence.
- Batch · Lô dữ liệu
- A subset of examples processed together in one update.
- Bias–variance trade-off · Đánh đổi độ chệch – phương sai
- Simpler models err by bias, flexible ones by variance.
- Classification · Phân loại
- Predicting a category.
- Clustering · Phân cụm
- Grouping similar examples without labels.
- Computer vision · Thị giác máy tính
- Computing with images and video.
- Confidence interval · Khoảng tin cậy
- A range likely to contain the true value of a parameter.
- Confusion matrix · Ma trận nhầm lẫn
- A table of predicted versus actual classes.
- Correlation · Tương quan
- A statistical relationship between two variables; not proof of causation.
- Cross-validation · Kiểm chứng chéo
- Evaluating a model by repeatedly splitting the data into training and validation parts.
- Data leakage · Rò rỉ dữ liệu
- Information from outside the training data sneaks into the model and inflates results.
- Data pipeline · Quy trình xử lý dữ liệu
- Automated steps that collect, clean and deliver data.
- Data visualization · Trực quan hoá dữ liệu
- Showing data with charts and graphics.
- Dataset · Tập dữ liệu
- A collection of examples used to train or evaluate a model.
- Deep learning · Học sâu
- Machine learning with many-layered neural networks.
- Dimensionality reduction · Giảm chiều dữ liệu
- Representing data with fewer variables while keeping the important structure.
- Embedding · Biểu diễn nhúng
- A vector of numbers that represents the meaning of text, images or other data.
- Epoch · Epoch (lượt huấn luyện)
- One full pass over the training set.
- Feature · Đặc trưng
- An input variable the model uses to make a prediction.
- Fine-tuning · Tinh chỉnh
- Continuing to train a pre-trained model on your own data.
- Gradient descent · Hạ gradient
- Repeatedly adjusting parameters in the direction that reduces the loss.
- Hallucination · Ảo giác
- A model produces confident but false or unsupported output.
- Hyperparameter · Siêu tham số
- A setting chosen before training, such as the learning rate.
- Hypothesis testing · Kiểm định giả thuyết
- A statistical method to judge whether an observed effect is likely due to chance.
- Inference · Suy luận
- Running a trained model to make predictions.
- Label · Nhãn
- The correct output for an example in supervised learning.
- Large language model · Mô hình ngôn ngữ lớn
- A very large model trained on text to understand and generate language.
- Learning rate · Tốc độ học
- The step size used in gradient descent.
- Loss function · Hàm mất mát
- A measure of how wrong the predictions are; training minimises it.
- Machine learning · Học máy
- Algorithms that learn patterns from data instead of following hand-written rules.
- Missing value · Giá trị thiếu
- An entry with no recorded value.
- Natural language processing · Xử lý ngôn ngữ tự nhiên
- Computing with human language.
- Neural network · Mạng nơ-ron
- A model of connected layers of simple units whose weights are learned.
- Normalization · Chuẩn hoá
- Rescaling values to a common range or distribution.
- Outlier · Giá trị ngoại lai
- A data point far from the others.
- Overfitting · Quá khớp
- The model memorises the training data and generalises poorly.
- p-value · Giá trị p
- The probability of a result at least as extreme as observed if the null hypothesis were true.
- Precision · Độ chính xác dương (precision)
- Of the items predicted positive, how many really are.
- Prompt · Câu lệnh gợi ý (prompt)
- The input text that instructs a language model.
- Recall · Độ phủ (recall)
- Of the truly positive items, how many were found.
- Regression · Hồi quy
- Predicting a continuous number.
- Regularization · Điều chuẩn
- A technique that discourages overly complex models to reduce overfitting.
- Reinforcement learning · Học tăng cường
- An agent learns by trial and error to maximise reward.
- Retrieval-augmented generation · Sinh văn bản có truy xuất (RAG)
- A language model answers using documents retrieved for the question.
- Supervised learning · Học có giám sát
- Learning from examples that come with the correct answer (labels).
- Test set · Tập kiểm thử
- Held-out data used once to estimate final performance.
- Token · Token
- A chunk of text (word or word piece) a language model reads and writes.
- Training set · Tập huấn luyện
- Data used to fit the model.
- Transfer learning · Học chuyển giao
- Reusing a model trained on one task as the starting point for another.
- Underfitting · Chưa khớp
- The model is too simple to capture the pattern in the data.
- Unsupervised learning · Học không giám sát
- Finding structure in data that has no labels.
- Validation set · Tập kiểm định
- Data used to tune choices such as hyperparameters.