CDL LICENSING EXAM SCRIPT FINAL PAPER
2026 COMPLETE QUESTIONS AND ANSWERS
REVIEW
▶ Which of the following statements about different optimizers is false?
A. RMSprop solves the problem of the Adagrad optimizer ending too early.
B. Compared with RMSprop, Adagrad is more sensitive to gradient
changes.
C. Both Adagrad and RMSprop can set adaptive learning rate for
parameters.
D. Adam requires bias correction for initial iterations. - Answer: B
▶ Which of the following is NOT an Al deep learning framework?
A. TensorFlow
B. Scikit-learn
C. Pytorch
D. Caffe - Answer: B
▶ Which of the following statements is true about classification models and
regression models in machine learning?
A. For regression problems, the output variables are discrete values. For
classification problems, the output variables are continuous.
B. The most commonly used indicators for evaluating regression and
classification problems are the accuracy and the recall rate.
C. Overfitting may occur in both regression and classification problems.
D. Logistic regression is a typical regression model. - Answer: C
▶ Which of the following are used in TensorFlow to describe the
computation process?
A. Parameter
B. Session
C. Data flow
D. Tensor - Answer: C
▶ Which of the following is NOT a characteristic of PyTorch?
A. Python first
B. Easy to debug
, C. Extensive ecosystem
D. Default use of static maps - Answer: D
▶ Which of the following statements is true about the training of large
models?
A. Foundation models cannot be directly deployed.
B. The pre-training of large models requires a large amount of high-quality
data.
C. Unlabeled data used in the pre-training phase of large models must be
used in the fine-tuning phase.
D. Large models have emergent abilities, and therefore they do not need to
be aligned with human values. - Answer: B
▶ Which of the following is not an advanced Al application?
A. Knowledge graph
B. rget Quantum computing
C. SDN
D. Intelligent driving - Answer: C
▶ What are the main schools of thought in Al? (Multiple choice)
A. Symbolism
B. Illusionism
C. Actionism
D. Connectionism - Answer: ACD
▶ What are the layers used in Transformer to capture time dependencies
within a sequence? (Multiple choice)
A. Residual layer
B. Self-attention layer
C. Location code
D. Feed-forward network layer - Answer: BC
▶ Which of the following hyperparameters may be defined during model
training? (Multiple choice)
A. Batch size
B. Learning rate
C. Optimizer type
D. Loss value - Answer: ABC
2026 COMPLETE QUESTIONS AND ANSWERS
REVIEW
▶ Which of the following statements about different optimizers is false?
A. RMSprop solves the problem of the Adagrad optimizer ending too early.
B. Compared with RMSprop, Adagrad is more sensitive to gradient
changes.
C. Both Adagrad and RMSprop can set adaptive learning rate for
parameters.
D. Adam requires bias correction for initial iterations. - Answer: B
▶ Which of the following is NOT an Al deep learning framework?
A. TensorFlow
B. Scikit-learn
C. Pytorch
D. Caffe - Answer: B
▶ Which of the following statements is true about classification models and
regression models in machine learning?
A. For regression problems, the output variables are discrete values. For
classification problems, the output variables are continuous.
B. The most commonly used indicators for evaluating regression and
classification problems are the accuracy and the recall rate.
C. Overfitting may occur in both regression and classification problems.
D. Logistic regression is a typical regression model. - Answer: C
▶ Which of the following are used in TensorFlow to describe the
computation process?
A. Parameter
B. Session
C. Data flow
D. Tensor - Answer: C
▶ Which of the following is NOT a characteristic of PyTorch?
A. Python first
B. Easy to debug
, C. Extensive ecosystem
D. Default use of static maps - Answer: D
▶ Which of the following statements is true about the training of large
models?
A. Foundation models cannot be directly deployed.
B. The pre-training of large models requires a large amount of high-quality
data.
C. Unlabeled data used in the pre-training phase of large models must be
used in the fine-tuning phase.
D. Large models have emergent abilities, and therefore they do not need to
be aligned with human values. - Answer: B
▶ Which of the following is not an advanced Al application?
A. Knowledge graph
B. rget Quantum computing
C. SDN
D. Intelligent driving - Answer: C
▶ What are the main schools of thought in Al? (Multiple choice)
A. Symbolism
B. Illusionism
C. Actionism
D. Connectionism - Answer: ACD
▶ What are the layers used in Transformer to capture time dependencies
within a sequence? (Multiple choice)
A. Residual layer
B. Self-attention layer
C. Location code
D. Feed-forward network layer - Answer: BC
▶ Which of the following hyperparameters may be defined during model
training? (Multiple choice)
A. Batch size
B. Learning rate
C. Optimizer type
D. Loss value - Answer: ABC
