이전 퀴즈 문제들에 대한 설명

퀴즈 문제 및 설명:

이전 퀴즈 (LSTM 모델 관련):

Q1: Which category of deep learning models do LSTMs belong to?

• Correct Answer: RNNs (Recurrent Neural Networks)

  • Explanation: LSTMs (Long Short-Term Memory networks) are a specific type of RNN designed to handle long-term dependencies in sequential data. They are a more advanced form of RNN and address the vanishing gradient problem.

Q2: Which statement(s) are accurate about LSTMs?

• Correct Answers:

  • LSTM has the ability to remove or add information to the cell states using gates

  • LSTMs are a special kind of RNN, capable of learning long-term dependencies

  • Explanation:

    • Gates: LSTMs utilize input, forget, and output gates to control the flow of information in their memory cells, allowing them to selectively add, remove, or retain information.
    • Long-Term Dependencies: LSTMs are designed to learn long-term dependencies in sequential data, which is a significant improvement over traditional RNNs.
    • Sparse Data & Regularization: LSTMs are not specifically designed to work with sparse data, although can. Regularization is often applied externally through techniques such as Dropout.
    • Internal encoder/decoder: LSTMs do not have internal encoder/decoder structures for dimensionality reduction, that is more related to Autoencoders.

Q3: Which statement best describes the "recall" value in the confusion matrix:

• Correct Answer: Indicates the ability of the model to identify all true positive samples of the dataset.

  • Explanation: Recall (also known as sensitivity or true positive rate) measures the proportion of actual positives that are correctly identified by the model. It focuses on the model's ability to find all relevant instances.
    • Recall = TP / (TP + FN)

Q4: Which statement best describes the "precision" value in the confusion matrix:

• Correct Answer: Indicates the ability of the model to identify the relevant samples only.

  • Explanation: Precision measures the proportion of predicted positives that are actually positive. It focuses on the model's ability to avoid making false positive predictions.
    • Precision = TP / (TP + FP)

Q5: What is the purpose of using a max-pooling layer:

• Correct Answer: To down-sample an input representation and reduce its dimensions

  • Explanation: A max-pooling layer is used to reduce the spatial dimensions of an input, making the network more efficient and reducing the number of parameters. It achieves this by selecting the maximum value within a pooling window. Max-Pooling does not address loss or pooling entropy directly.

최근 퀴즈 (Autoencoder 및 활성화 함수 관련):

Q1: Which statement is not true on autoencoders?

• Correct Answer: The decoder network tries to predict the sample class of the input.

  • Explanation: Autoencoders are designed for unsupervised learning tasks, primarily for dimensionality reduction, feature learning, and reconstruction of input data. They consist of two main parts: an encoder and a decoder. The encoder maps the input data to a lower-dimensional representation (latent space), and the decoder tries to reconstruct the original input from this representation. They don't attempt to predict the sample class of input data directly as a classification task does.
    • Other Statements are generally correct
    • The output dimension is the same as the input dimension: This is correct.
    • The encoder encodes its input data into a smaller dimensional space, called the latent space: Correct.
    • Autoencoders could be thought of as a generalization of the PCA method: Correct. They can perform nonlinear dimensionality reduction, unlike PCA which is linear.

Q2: What is ROC Curve?

• Correct Answer: A performance measurement for classification problem at various thresholds settings

  • Explanation: A Receiver Operating Characteristic (ROC) curve is a graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied. It plots the true positive rate (recall) against the false positive rate at various threshold settings.
    • ROC curves are not concerned with precision, but with the tradeoff between true positives and false positives.
    • ROC curves are specifically for model assessment and not to show data imbalance.

Q3: What statement is not accurate on "tanh" and "sigmoid" activation functions?

• Correct Answer: The range of sigmoid function is also between -1 to 1

  • Explanation:
    • tanh: The hyperbolic tangent function (tanh) outputs values in the range of -1 to 1.
    • sigmoid: The sigmoid (or logistic) function outputs values in the range of 0 to 1. It is commonly used for probability predictions because its output can be interpreted as a probability. The tanh function can be thought of as a scaled logistic function.
    • The other statements are correct

Q4: Which category of problems is not applicable to autoencoder models?

• Correct Answer: Timeseries prediction

  • Explanation:
    • Autoencoders are primarily used for:
      • Denoising: Removing noise from images or data.
      • Image reconstruction: Reconstructing an image from a compressed representation.
      • Anomaly detection: Identifying unusual data points that differ significantly from normal data.
      • Feature extraction: Learning compact representations of data that can be used in other tasks.
    • Timeseries Prediction: while autoencoders can be part of time series models, they are not primarily used for timeseries prediction. Instead, RNNs (Recurrent Neural Networks) such as LSTMs are typically used for timeseries prediction.

정리 (모든 퀴즈):

• 이전 퀴즈 (LSTM):
  • Q1: RNNs
  • Q2: LSTMs are a special kind of RNN, capable of learning long-term dependencies; LSTM has the ability to remove or add information to the cell states using gates
  • Q3: Indicates the ability of the model to identify all true positive samples of the dataset
  • Q4: Indicates the ability of the model to identify the relevant samples only
  • Q5: To down-sample an input representation and reduce its dimensions
• 최근 퀴즈 (Autoencoder 및 활성화 함수):
  • Q1: The decoder network tries to predict the sample class of the input
  • Q2: A performance measurement for classification problem at various thresholds settings
  • Q3: The range of sigmoid function is also between -1 to 1
  • Q4: Timeseries prediction

이 설명이 퀴즈 문제와 관련된 개념들을 이해하는 데 도움이 되기를 바랍니다. 혹시 추가적으로 궁금한 점이나 다른 질문이 있으시다면 언제든지 저에게 알려주세요. 제가 최선을 다해 답변 드리겠습니다.