Please enable JavaScript.

Coggle requires JavaScript to display documents.

NLP (Reinforcement Learning (Introduction (State (Experience is a sequence…

- - - - Text summarization is the process of automatically generating natural language summaries from an input document while retaining the important points.
      - extractive summarization
      - abstractive summarization
        
        issues
        
        discrepancy between training and test performance, because:
        
        exposure bias
        
        potentially valid summaries
        
        metric difference
  - - - agent’s behavior function
      - function that maps from state to action
    - - how good would be each state and/or action
  - - - Experience is a sequence of observations, actions, rewards
      - The state is a summary of experience
  - - - Make the good trajectories/action more probable
      - Push the actions towards good actions
      - Represent policy by deep neural network
    - - Represent value function by Q-network
- - - - trying to predict the next word given the previous words in a sentence
    - - or similar standard classification or regression tasks
    - - there are many datasets and tutorials for that nowadays
    - - says if pairs of sentences are in agreement or contradict
      - it's not entirely clear whether it's actually useful
    - - area of some recent significant advances
      - it's not entirely clear whether it's actually useful
    - - tons of information extraction tasks can be modeled as sequence labelling
      - POS tagging, entity recognition
  - - - you feed the texts to a NN in an unsupervised approach
      - then fine-tune for specific tasks
        
        summarizing a scientific article
        
        classifying an email as spam
        
        generating a satisfying end to a short story
      - the hope is that those patterns will capture some general aspects of language - a sense of words are, or the basic grammar
    - - word embeddings: maps how words relate to each other based on how they appear in large amount of text
      - new model aims to go deeper, capturing information that scales up from words to higher-level concepts of language