Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Glossary

ML/AI/MLOps sohasidagi muhim terminlarning English-Uzbek lug’ati. Har termin for qisqacha izohand kontekst.

A

  • Activation function(faollik funksiyasi) — neural network in nonlinearity qo’shadigan funksiya (ReLU, Sigmoid, Tanh).
  • AdamW — Adam optimizatori + better weight decay; modern default.
  • Agent (AI Agent) — LLM + tools + memory; goal to erishish for sequential harakatlar.
  • Anchor box — object detection in predefined bounding box shape.
  • ANN (Approximate Nearest Neighbor) — quickly finding nearby vectors (HNSW, IVF).
  • API (Application Programming Interface) — dasturlash interfeysi.
  • Async / await — Python in concurrent operations.
  • Attention mechanism — sequence’dagi muhim qismlarga “diqqat” qaratish.
  • AUC (Area Under Curve) — ROC curve ostidagi maydon (classification metric).
  • AutoGrad — PyTorch’ning avtomatik gradient hisoblash mexanizmi.

B

  • Backpropagation — gradients orqaga tarqatish; neural network teach algoritmi.
  • Bagging (Bootstrap Aggregating) — parallel ensemble (Random Forest asosi).
  • Batch — bir vaqtda model to uzatilgan samples to’plami.
  • Batch Normalization (BN) — activations batch ichida normallashtirish.
  • Bayesian Optimization — smart hyperparameter qidiruv (Optuna).
  • Bias (matematik) — model output’iga qo’shiladigan constant.
  • Bias (xulosa) — algoritmda noto’g’ri predictions to moyillik.
  • Boosting — sequential ensemble (XGBoost, LightGBM).
  • BPE (Byte-Pair Encoding) — subword tokenization (GPT, Llama in).
  • Broadcasting — NumPy/PyTorch in turli shape’dagi tensors to operatsiya.

C

  • Calibration — model probability’larini ishonchli .
  • Canary deployment — testing a new version on a small portion of traffic.
  • Categorical feature — diskret qiymatli feature (city, color).
  • Chain-of-Thought (CoT) — LLM in step-by-step reasoning prompt.
  • Checkpoint — model state saqlash (resume training for).
  • Classification — sample diskret classs to ajratish.
  • Clustering — o’xshashlarni guruhlash (unsupervised).
  • CNN (Convolutional NN) — image processing for neural network.
  • Cold start — the problem of no data about a new user/item.
  • Concept drift — input → output relationship vaqt o’tishi with o’zgarishi.
  • Confusion Matrix — TP, FP, TN, FN ko’rsatadigan jadval.
  • Context window — LLM bir vaqtda ko’ra that gets token soni.
  • Cosine similarity — ikki vektor orasidagi cos burchak.
  • CRD (Custom Resource Definition) — Kubernetes custom obyekt.
  • Cross-encoder — sentence pairs for classifier (reranking in).
  • Cross-validation (CV) — model bir necha baholash.
  • CUDA — NVIDIA GPUs in parallel computation.

D

  • DAG (Directed Acyclic Graph) — Airflow in workflow ko’rinishi.
  • Data augmentation — sun’iy ravishda training data kengaytirish.
  • Data drift — input distribution vaqt o’tishi with o’zgarishi.
  • Data leakage — test/validation data training to “sizib o’tishi” (xato).
  • DataFrame — Pandas in tabular data strukturasi.
  • DataLoader — PyTorch in batch yuklash.
  • Decision Tree — qoidalar daraxtidan iborat klassik ML algoritmi.
  • Deep Learning (DL) — chuqur (ko’p qatlamli) neural networks.
  • DevOps — software development + operations integratsiyasi.
  • Diffusion model — image generation (Stable Diffusion, DALL-E).
  • Dimensionality reduction — features sonini kamaytirish (PCA, t-SNE).
  • Docker — application containerization.
  • Dropout — randomly “turning off” neurons to reduce overfitting.
  • DVC (Data Version Control) — Git for data.

E

  • EDA (Exploratory Data Analysis) — ma’lumotlarni tahlil bosqichi.
  • Embedding — diskret obyektni dense vektorga aylantirish.
  • Encoder-Decoder — translation/summarization arxitekturasi.
  • Ensemble — bir nechta model birgalikda.
  • Epoch — butun dataset bo’yicha bir martalik training.
  • Evaluation — model sifatini o’lchash.
  • Evidently AI — drift detection and monitoring tool.

F

  • F1 Score — precision and recall’ning harmonic mean.
  • FastAPI — modern Python web framework (Pydantic asosida).
  • Feature — model input’idagi each o’lchov.
  • Feature engineering — creating new features.
  • Feature store — features saqlash and serve (Feast).
  • Few-shot learning — teaching with few examples.
  • Fine-tuning — pretrained modelni o’z task to moslashtirish.
  • Flask — micro web framework (FastAPI from oldingi standard).
  • F-score — F1 ning umumiy holati (beta parametri with).
  • Function calling / Tool use — LLM to tashqi functions chaqirishga ruxsat.

G

  • GAN (Generative Adversarial Network) — generator + discriminator.
  • Gemini — Google’ning LLM oilasi.
  • Generative AI — content yaratuvchi AI (matn, rasm, audio).
  • Gini index — Decision Tree in split quality.
  • GitHub Actions — CI/CD platform.
  • GPT (Generative Pretrained Transformer) — OpenAI LLM oilasi.
  • GPU (Graphics Processing Unit) — parallel computation for.
  • Gradient — the direction of fastest growth of a function.
  • Gradient Boosting — sequential boosting algoritm.
  • Gradient Descent — loss minimize algoritmi.
  • Grafana — monitoring dashboard.
  • GridSearch — hyperparameter exhaustive qidiruv.

H

  • Hallucination — LLM’ning ishonchli ko’rinishda noto’g’ri javob berishi.
  • Helm — Kubernetes package manager.
  • HNSW (Hierarchical Navigable Small Worlds) — fast ANN algorithm.
  • HPA (Horizontal Pod Autoscaler) — Kubernetes auto-scaling.
  • HuggingFace — ML modellar and datasetlar for platform.
  • Hybrid search — vector + keyword (BM25) qidiruv.
  • HyDE (Hypothetical Document Embeddings) — RAG texnikasi.
  • Hyperparameter — training from oldin belgilangan parametr (lr, batch).

I

  • Image segmentation — pixel-level classification.
  • Imbalanced data — classs soni teng emas.
  • Inference — model with prediction .
  • Ingress — Kubernetes external HTTP routing.
  • Instance segmentation — har object to alohida mask.
  • Instruction tuning — instructions with fine-tuning.
  • IoU (Intersection over Union) — object detection metric.

J

  • Jupyter Notebook — interactive Python environment.

K

  • Keras — high-level NN API (TensorFlow in).
  • K-Fold Cross-validation — dataset K ta foldga .
  • K-Means — clustering algoritmi.
  • KNN (K-Nearest Neighbors) — yaqin K ta sample asosida classification.
  • Kubernetes (K8s) — container orchestration.
  • Kubeflow — Kubernetes-native ML platform.

L

  • L1, L2 regularization — Lasso (L1), Ridge (L2).
  • LangChain — LLM application framework.
  • LangGraph — stateful multi-agent workflows.
  • Langfuse — LLM observability platform.
  • LayerNorm — Layer normalization (Transformers in).
  • Learning rate (lr) — gradient descent qadam kattaligi.
  • LightGBM — fast gradient boosting (Microsoft).
  • Linear Regression — eng oddiy regression algoritmi.
  • LLM (Large Language Model) — large language model.
  • LlamaIndex — RAG framework.
  • LoRA (Low-Rank Adaptation) — efficient fine-tuning.
  • Loss function — model xatosini o’lchaydigan funksiya.

M

  • MAE (Mean Absolute Error) — regression metric.
  • MAP (mean Average Precision) — object detection metric.
  • MAPE (Mean Absolute Percentage Error) — % ko’rinishidagi xato.
  • MCP (Model Context Protocol) — Anthropic’ning agent tool standarti.
  • MinMaxScaler — scaling features to [0, 1].
  • MLflow — experiment tracking platform.
  • MLOps — ML + DevOps integratsiyasi.
  • Model registry — versionlangan modellar saqlash.
  • MSE (Mean Squared Error) — regression loss.
  • Multi-class classification — 3+ classs orasida choose.
  • Multi-label classification — bir sample to bir nechta label.
  • Multi-task learning — bir model bir nechta task.

N

  • N-gram — N ta consecutive so’zlar.
  • Naive Bayes — probabilistic classifier (text for mashhur).
  • NER (Named Entity Recognition) — matnda nomlangan obyektlar.
  • Neural Network (NN) — bir-biriga bog’langan neuronlar tarmog’i.
  • NLP (Natural Language Processing) — matn with work.
  • NMS (Non-Maximum Suppression) — overlapping detections filter.
  • Normalization — bringing features to the same scale.
  • NumPy — numerical computation library.

O

  • One-Hot Encoding — categorical → binary vektor.
  • ONNX (Open Neural Network Exchange) — cross-framework model format.
  • OpenAI — GPT yaratuvchi kompaniya.
  • Optimizer — gradient qanday qo’llash (SGD, Adam, AdamW).
  • Optuna — Bayesian hyperparameter tuning.
  • Overfitting — model train in yaxshi, test in yomon.

P

  • Pandas — tabular data manipulation.
  • Parameter — a value learned in the model (weight).
  • PCA (Principal Component Analysis) — dimensionality reduction.
  • PEFT (Parameter-Efficient Fine-Tuning) — LoRA, QLoRA and h.k.
  • Perceptron — eng oddiy neuron.
  • Pipeline — sklearn in preprocessing + model.
  • Pod — the smallest unit in Kubernetes.
  • Pooling — CNN in downsampling (MaxPool, AvgPool).
  • POS tagging (Part-Of-Speech) — gap aniqlash.
  • Postgres / PostgreSQL — relational database.
  • Precision — TP / (TP + FP).
  • Prefect — modern workflow orchestrator.
  • Pretrained model — a model pre-trained on a large corpus.
  • Prompt — LLM to beriladigan input matn.
  • Prompt engineering — yaxshi prompt yozish san’ati.
  • Prometheus — metrics monitoring system.
  • PSI (Population Stability Index) — drift detection metric.
  • Pydantic — Python data validation.
  • PyTorch — deep learning framework.

Q

  • QLoRA — 4-bit quantization + LoRA.
  • Qdrant — vector database (Rust).
  • Quantization — model precision’ini kamaytirish (8-bit, 4-bit).
  • Query — LLM/search to beriladigan savol.

R

  • R² — coefficient of determination (regression).
  • RAG (Retrieval Augmented Generation) — LLM + knowledge retrieval.
  • RAGAS — RAG evaluation framework.
  • Random Forest — bagging Decision Trees.
  • RandomizedSearch — random hyperparameter qidiruv.
  • Recall — TP / (TP + FN).
  • Recommender system — tavsiya sistemasi.
  • ReAct (Reasoning + Acting) — agent pattern.
  • Recurrent Neural Network (RNN) — sequence for NN.
  • Redis — in-memory database.
  • Regex (Regular Expression) — pattern matching.
  • Regression — uzluksiz qiymat bashorat.
  • Regularization — overfitting kamaytirish (L1, L2, Dropout).
  • Reranking — search natijalarini qayta tartibga solish.
  • REST API — HTTP-based API standard.
  • ResNet — skip connection’lari CNN.
  • RLHF (Reinforcement Learning from Human Feedback) — LLM alignment.
  • RMSE (Root Mean Squared Error) — sqrt(MSE).
  • ROC-AUC — Receiver Operating Characteristic Area Under Curve.

S

  • SageMaker — AWS ML platform.
  • Scaler — feature normalization (Standard, MinMax).
  • scikit-learn — Python ML library.
  • Self-attention — sequence ichidagi tokens orasidagi attention.
  • Self-supervised learning — labels without pretraining.
  • Semantic search — meaning-based qidiruv (vector search).
  • Sentence Transformer — sentence embeddings.
  • SFT (Supervised Fine-Tuning) — instructions with fine-tune.
  • SGD (Stochastic Gradient Descent) — klassik optimizer.
  • SHAP (SHapley Additive exPlanations) — model interpretation.
  • Shadow deployment — testing a new model without traffic.
  • Sigmoid — activation function (binary class for).
  • Softmax — multi-class output activation.
  • spaCy — NLP library.
  • Standardization — (x - mean) / std.
  • Streaming — real-time response (SSE, WebSocket).
  • Supervised learning — learning with labels.
  • SVM (Support Vector Machine) — klassik classifier.

T

  • Tensor — multi-dimensional array (NumPy ndarray’ning generalizatsiyasi).
  • TensorFlow — Google’ning DL framework’i.
  • Test set — yakuniy baholash for ajratilgan data.
  • TF-IDF — text feature representation.
  • Threshold — classification decision chegarasi.
  • Token — tokenization from keyingi atomic unit.
  • Tokenizer — matnni tokens to ajratish.
  • TorchServe — PyTorch production serving.
  • Train set — model learn data.
  • Transfer learning — pretrained model o’z task to qo’llash.
  • Transformer — attention-based arxitektura (BERT, GPT).
  • Triton — NVIDIA inference server.

U

  • Underfitting — model juda oddiy, train in ham yomon.
  • Unicode — character encoding standard.
  • Unsupervised learning — learning without labels.

V

  • Validation set — hyperparameter tuning for data.
  • Variance — data tarqoqlik darajasi.
  • Vector — 1-D array.
  • Vector Database — embeddings saqlash and search.
  • ViT (Vision Transformer) — rasm for Transformer.
  • vLLM — fastest LLM inference server.

W

  • WandB (Weights & Biases) — experiment tracking.
  • Weight — neuron coefficient.
  • WebSocket — bidirectional connection.
  • Word2Vec — word embedding model.
  • Workflow orchestration — tasks ketma-ketligini boshqarish (Airflow).

X

  • XGBoost — popular gradient boosting library.
  • XLM-R — multilingual RoBERTa.

Y

  • YAML — config fayl formati.
  • YOLO (You Only Look Once) — fast object detection.

Z

  • Zero-shot learning — task without examples using pre-existing knowledge.

Return to Main page or go to Resources.