Module- 4 --RNN

1234567891011121314151617181920212223
Across
  1. 1. Representation of an RNN expanded across multiple time steps
  2. 3. A single point or step in a sequence processed by an RNN
  3. 5. Recurrent neural network architecture designed to handle long-term dependencies
  4. 8. Direction in which a bidirectional RNN processes a sequence from end to beginning
  5. 9. Problem in which gradients become extremely small during training
  6. 14. Describes a neural network that uses information from previous time steps
  7. 16. The internal representation maintained by an RNN at each time step
  8. 17. Direction in which an RNN processes a sequence from beginning to end
  9. 19. Relationship between elements of a sequence that may occur across different time steps
  10. 20. A modified form of BPTT that backpropagates through only a limited number of time steps
  11. 21. RNN that processes a sequence in both forward and backward directions
  12. 22. Information maintained by an RNN from previous time steps
  13. 23. Neural network designed to process sequential or time-dependent data
Down
  1. 2. A weakness or drawback of a standard RNN
  2. 4. Structure used to represent computations and dependencies in an RNN
  3. 6. Describes dependencies between events separated by many time steps
  4. 7. Algorithm used to calculate gradients for updating neural-network weights
  5. 10. An ordered set of data elements where the order of elements is important
  6. 11. Method used to train an RNN by propagating errors backward through time
  7. 12. Mathematical quantity used to determine how weights should be updated
  8. 13. Relating to the graph that represents mathematical operations in a neural network
  9. 15. Problem in which gradients become extremely large during training
  10. 18. Basic RNN architecture without advanced memory mechanisms such as LSTM gates
  11. 20. Abbreviation for Truncated Backpropagation Through Time