Machine Learning Flashcards
What does a supervised learning model learn from?
Labeled training data, where inputs are mapped to known outputs/classes
Give two examples of supervised learning tasks
Classification (image classification, spam detection)
Regression (predicting housing prices, stock prices)
Why is a test set needed in supervised learning?
To evaluate the model’s performance on unseen data and check for overfitting
What is the goal of unsupervised learning?
To discover inherent patterns, groups or structures within unlabeled input data
Give two examples of unsupervised learning tasks
Clustering (customer segmentation, grouping related documents)
Dimensionality reduction (compressing data while preserving patterns)
Why is unsupervised learning useful?
It allows extracting insights and understanding from unlabeled datasets with no predefined outputs
How does a reinforcement learning agent learn?
By taking actions in an environment to maximize a reward signal through trial-and-error
Give an example of a reinforcement learning problem
Game playing (chess, Go), robotics control for navigation/grasping
What distinguishes reinforcement learning from supervised/unsupervised?
It learns from reward signals instead of ground truth labels or inherent patterns
What algorithm allows training deep neural networks?
Backpropagation to propagate errors backwards and update weights
What is a key limitation of current neural networks?
They are not biologically plausible models of human learning and cognition
Why is training data bias an important issue in machine learning?
Biased datasets can lead to harmful model outputs that reinforce societal biases
What is overfitting?
When a model fits the training data too well, capturing even the noise/randomness, such that it fails to generalize well to new unseen data.
What techniques can help mitigate overfitting?
Using a validation/test set to evaluate generalization
Regularization methods that constrain model complexity
Early stopping of training before overfitting occurs
List some key aspects that distinguish human cognition from current machine learning models like neural networks.
Humans learn much faster from fewer examples compared to neural nets
Human learning is embodied and situated in the physical world
Humans exhibit reasoning, abstraction and top-down attentional control
The neural mechanisms underlying human learning are still not well understood
