Please enable JavaScript.
Coggle requires JavaScript to display documents.
Deep Learning (Supervised (Concerns (Feature learning (General Algorithm…
Deep Learning
Supervised
inferring a function from labeled training data
each individual example consists of an input object and and an output value
algorithm produces an inferred function
to generalize from the training data to unseen situations in a "reasonable" way
Concerns
Bias-variance tradeoff
Function complexity and amount of training data - more complexity means larger training set
Attempting to fit the data too carefully leads to
overfitting
.
Heterogeneity of the data - solved by
decision trees
Redundancy in the data - solved by
regularization
)
Feature interactions and non-linearities
Feature learning
learned function
g
scoring function
f
loss function
L
may be conditional probabilistic functions
General Algorithm Strategies
Empirical risk minimization
Maximize fit
Structural risk minimization
Minimize penalty, loss
Learning Models
Discriminative model
Generative model
Unsupervised
Producing inferred function from unlabelled data
no evaluation of the accuracy to the objective reality
Approaches
Clustering
k-means
mixture models
hierarchical clustering
Anomaly detection
Neural Networks
Hebbian Learning
Generative Adversarial Networks