Q1. what is AI (Artificial Intelligence)?
Read the following statement and choose the option that best fits the blank.

What people imagine when they hear the words "AI" or "artificial intelligence" varies from person to person. (A)-type artificial intelligence is also called "(B)AI," and is defined as AI that is not limited to specific tasks and has generalization capabilities similar to or better than humans. The humanoid robots that appear in many science fiction works are flexible enough to perform more than programmed intellectual tasks. On the other hand, (c)-type AI, also called (d)AI, is defined as AI that performs certain tasks as well as or better than humans. For example, (e) AI, which learned the tactics of Go, won a victory over the South Korean professional Go player Lee Sedol (2016). However, it is difficult to apply Go capabilities to chess. For this reason, it is distinguished from (a)-type AI, which has a high generalization capability, and is called "(c)-type AI.

(a) Options that fall under (b)

(a) Generalization (b) Weak
(a) Generalization (b) Strong
(a) Specialization (b) Weak
(a) Specialization (b) Strong

(c) Options that fall under (d)

(c) Generalization (d) Weak
(c) Generalization (d) Strong
(c) Specialization (d) Weak
(c) Specialization (d) Strong

Options corresponding to (e)

elmo
Stockfish
Watson
AlphaGo

Q2. perceptron
The simple perceptron, known as the origin of deep learning, has the structure shown in the figure below. Choose the most appropriate assignment for this simple perceptron from the following options.

1:The state-action space is huge
2: Inability to represent nonlinear transformations
3: Difficult to compute at high speed on GPUs
4: Curse of dimensionality is likely to occur.

Q3. supervised/unsupervised learning and reinforcement learning
AI is classified as shown in the figure below.
Focusing on the type of task (medium classification) here, choose one appropriate example from the following sentences, especially as an example of clustering in unsupervised learning.

1: A car manufacturer analyzed the attributes of past hit products by k-means in order to come up with its next product. The results were grouped into three categories: minivans for home use, one-boxes that carry a lot of luggage, and SUVs that are popular.

2: In order to determine the academic characteristics of the students, principal component analysis was performed on the test scores of 9 subjects. The results showed that there were two useful principal components, and that the students' academic ability could be measured by their logical thinking skills developed in Japanese, mathematics, and English, and their creativity developed in technology, art, and music.

3: In the credit screening, credit scores were learned with a generalized linear regression model using each person's attributes as explanatory variables. The regression coefficients were observed, and it was found that whether the occupation provided a stable income contributed to the prediction.

4: Using a CNN model, we learned an image classification task for dogs and cats. visualizing the filters in the CNN, we found that the filter detecting the presence or absence of whiskers contributed to the classification.

Q4: Bias, Variance and Regularization
Choose the incorrect statement about bias/variance below.

1: Bias and variance are a trade-off (as one becomes smaller, the other becomes larger).
2: Variance is the variance (scatter) of the model's predictions, and if this value is large, we can say that the model is over-trained.
3: Bias is the noise between the data, which cannot be suppressed by machine learning methods.
4: Regularization has the effect of suppressing variance and preventing overlearning by applying constraints to the parameters (weights) and the model during training.

ref
https://avilen.co.jp/test/g-certificate/