Tarjetas de este mazo (10)

• What is the exploration vs. exploitation trade-off in RL?

  • Exploration: Discover new strategies
  • Exploitation: Maximize known rewards
  • Balance: Both improve learning over time
  reinforcement_learning trade_off
• What is backpropagation in neural networks?

  • Algorithm for adjusting weights
  • Uses the chain rule
  • Calculates gradient of the loss function
  neural_networks backpropagation
• What is agglomerative clustering?

  • Bottom-up hierarchical clustering
  • Each data point starts in its own cluster
  • Similar clusters are merged step-by-step
  clustering machine_learning
• What is Ridge regression?

  • Uses L2 regularization
  • Shrinks all feature weights
  • Reduces overfitting while keeping variables
  regression machine_learning
• What is Lasso regression?

  • Uses L1 regularization
  • Shrinks some feature weights to zero
  • Enables automatic feature selection
  regression machine_learning
• How do Ridge and Lasso differ?

  • Ridge: Reduces all weights, keeps them non-zero
  • Lasso: Can set some weights to zero, removes unimportant features
  regression machine_learning
• Why is a dictionary not machine learning?

  • Provides 1-to-1 mappings
  • Does not learn patterns
  • ML generalizes to unseen data
  machine_learning comparison
• Do neural network layers represent literal image slices?

  • No: Layers abstract patterns statistically
  • Features extracted based on weight optimization
  neural_networks features
• What is the role of a cost function in ML?

  • Measures distance between predictions and actual values
  • Guides model learning via optimization techniques
  machine_learning cost_function
• What is a criticism of quantitative obsession in ML research?

  • Minor tweaks (e.g., Ridge to Lasso) published for better metrics
  • Ignores innovation or qualitative insight
  research criticism