열렬히.뛰기

질문 : NLP with BERT and GPT

취업 준비 > 기술면접 : 예상 > 질문 : NLP with BERT and GPT

머신러닝 / 딥러닝

  • 딥러닝은 무엇인가요? 딥러닝과 머신러닝의 차이는?
    • deep learning is a mimic of the human brain.
    • There are a lot of types of machine learning methods such as Logistic regression, SVM, tree model, random forest. Deep Learning is one of them.
  • 왜 갑자기 딥러닝이 부흥했을까요?
    • The quantity of data has increased so much after the era of smartphone arrive.
    • But the Data we faced these days are different from the traditional situation.
    • Structure of the Data has been changed such as video, audio, and photos
  • Cost Function과 Activation Function은 무엇인가요?
    • Cost function, aka loss function, quantifies the error between the predicted values and actual value. ex. MSE
    • Activation function tells whether a neuron should be activated or not. They allow the model to learn more complex pattern. ex. ReLU, Sigmoid, Tan-hyperbolic..
  • 오버피팅일 경우 어떻게 대처해야 할까요?
    • There are 3 ways to handle overfitting.
    • First, simplify the model by using L1 or L2 regularization technique
    • Second, use cross-validation techniques
    • Last, pruning or dropout can be used. We can achieve a balance between bias and variance
  • 하이퍼파라미터는 무엇인가요?
    • parameters, that are not learned from the data.
    • play a crucial role in determining the behavior of the training algo.

자연어처리

  • One-Hot 인코딩에 대해 설명해주세요
    • It is the method to resolve categorical data.
    • Each categories are
  • POS 태깅은 무엇인가요?
    • a process in NLP, clarify the meaning of a word in its specific content
    • the accuracy and reliability of POS tagging can significantly impact the performance of NLP.
  • RNN / LSTM 설명
    • RNN is a type of Neural Network that is used for sequential data. Unlike feed-forward NN, they have loops within so that the info can persist. Key feature : captures info about previous state.
    • LSTM is a special kind of RNN, specifically designed to overcome the problem of vanishing gradient. They achieve it by components called gates.
github.com/zzsza/Datascience-Interview-Questions