Sie sind auf Seite 1von 1

• Not to replicate the

training data
What are the inputs to
Instances space the problem? What are
• Generalization: How the attributes?
well a model trained on Binary classification
the training set predicts
Classification
the correct output for
new instances? Multi-class
Given labeled data,
Aim of ML What is the predictive classification
learn a model that can Label space
predict labels on task?
• For best unseen data
generalization, Output is numerical
Regression
complexity of the (real numbers)
learned function, g(x),
must Choosing between
possible hypothesis or Formalizing Supervised What functions should A hypothesis space is a
match the complexity
choosing the right
Model Selection CHAPTER 2 Learning Hypothesis space the learning algorithm set of possible functions
of the function
bias search over? we consider
underlying the data, f(x)

Learning Job is to output a


Underfitting: g(x) is less Size: number of
Algorithm function
complex than f(x) hypotheses to choose
A model that can
from
Classification predict the output
based on label data Job of hypothesis
Overfitting: g(x) is too
is a function that There are 3 concerns for
complex Randomness: stochastic
Hypothesis (h) takes a new input a choosing a hypothesis
or deterministic
and tries to map space space
output
Parameter: the
number and type of
Inductive bias Set of assumptions parameters

Das könnte Ihnen auch gefallen