| No. | Item | Definition |
|---|---|---|
| 1. | ablation | testing by removing components |
| 2. | accuracy | share of correct predictions |
| 3. | activation | function adding nonlinearity |
| 4. | algorithm | step-by-step problem-solving procedure |
| 5. | alignment | matching model behavior to goals |
| 6. | attention | mechanism for focusing on inputs |
| 7. | backpropagation | error signal passed backward |
| 8. | baseline | simple reference performance level |
| 9. | batch | small group of training examples |
| 10. | benchmark | standard test for comparison |
| 11. | bias | systematic error or unfair skew |
| 12. | boosting | sequentially improving weak learners |
| 13. | classification | assigning items to categories |
| 14. | clustering | grouping similar items together |
| 15. | compression | reducing model size or cost |
| 16. | compute | processing power used |
| 17. | corpus | large structured text collection |
| 18. | correlation | statistical association between variables |
| 19. | cross-validation | repeated train-test splitting method |
| 20. | data augmentation | creating varied training examples |
| 21. | dataset | collection of data for analysis |
| 22. | decision tree | rule-based branching model |
| 23. | deployment | putting a model into use |
| 24. | distillation | teaching a smaller model |
| 25. | distribution | pattern of values in data |
| 26. | drift | change in data or behavior |
| 27. | dropout | randomly removing units during training |
| 28. | embedding | dense numeric representation of meaning |
| 29. | ensemble | combination of multiple models |
| 30. | entropy | measure of uncertainty or disorder |
| 31. | epoch | one full pass through data |
| 32. | explainability | ability to explain model decisions |
| 33. | F1 score | balance of precision and recall |
| 34. | fairness | equitable treatment across groups |
| 35. | feature | input attribute used for learning |
| 36. | federated learning | training across separate devices |
| 37. | fine-tuning | adapting a pretrained model |
| 38. | generalization | performance on unseen data |
| 39. | GPU | processor suited for parallel math |
| 40. | gradient | direction of steepest change |
| 41. | ground truth | trusted correct answer |
| 42. | hallucination | confidently generated false content |
| 43. | heuristic | practical shortcut method |
| 44. | hyperparameter | setting chosen before training |
| 45. | inference | using a trained model to predict |
| 46. | initialization | starting parameter values |
| 47. | interpretability | how understandable a model is |
| 48. | k-means | common clustering algorithm |
| 49. | kernel | similarity function in some models |
| 50. | latency | delay before a response |
| 51. | learning rate | step size during optimization |
| 52. | likelihood | probability of data under model |
| 53. | logit | raw score before probability conversion |
| 54. | loss | measure of prediction error |
| 55. | memory | storage available during computation |
| 56. | metric | measure used to judge performance |
| 57. | model | system that learns patterns from data |
| 58. | monitoring | tracking system behavior over time |
| 59. | multimodal | using multiple data types |
| 60. | nearest neighbor | prediction by similar examples |
| 61. | neural network | layered system inspired by neurons |
| 62. | NLP | computer processing of human language |
| 63. | noise | random or irrelevant variation |
| 64. | normalization | rescaling values for stability |
| 65. | optimizer | method for updating parameters |
| 66. | overfitting | memorizing training data too closely |
| 67. | parameter | learned value inside a model |
| 68. | PCA | method reducing data dimensions |
| 69. | perplexity | uncertainty measure for language models |
| 70. | posterior | updated probability after evidence |
| 71. | precision | correct positives among predicted positives |
| 72. | pretraining | initial broad training stage |
| 73. | prior | belief before seeing evidence |
| 74. | prompt | input instruction for a model |
| 75. | quantization | using lower-precision numbers |
| 76. | random forest | ensemble of decision trees |
| 77. | recall | found positives among actual positives |
| 78. | regression | predicting continuous numeric values |
| 79. | regularization | method to reduce overfitting |
| 80. | reinforcement learning | learning through rewards and actions |
| 81. | ReLU | common rectifying activation function |
| 82. | robustness | stability under changes or noise |
| 83. | safety | reducing harmful model behavior |
| 84. | sampling | selecting examples or outputs |
| 85. | scalability | ability to handle growth |
| 86. | sigmoid | S-shaped squashing function |
| 87. | signal | useful pattern in data |
| 88. | softmax | turns scores into probabilities |
| 89. | SVM | margin-based classification method |
| 90. | test set | held-out data for evaluation |
| 91. | throughput | amount processed per time |
| 92. | token | basic unit of text |
| 93. | tokenization | splitting text into tokens |
| 94. | TPU | specialized chip for machine learning |
| 95. | training | process of teaching a model |
| 96. | transformer | attention-based neural architecture |
| 97. | underfitting | failing to capture patterns |
| 98. | variance | sensitivity to training fluctuations |
| 99. | vector database | store for embedding search |
| 100. | weights | connection strengths in a model |