Please enable JavaScript.
Coggle requires JavaScript to display documents.
Mac_Lea (Reinforcement (Dynamic programming (without knowing the exact …
Mac_Lea
Reinforcement
learn from mistake
software agent/algo
sw take action
maximize the reward
target var not available:
control (driver-less car)
categorical target var:
classification (optimized marketing)
environment: MDP
Markov Decision Process
Dynamic programming
without knowing the exact
math model of MDP
target large MDP
exact method is infeasible
unsupervised
prob(X)
cluster analysis of unlabeled data
data is not labeled, categorized or classified
identifies commonalities
clustering
Customer segmentation
association
Market Basket Analysis
data driven(identify clusters)
Semi-supervised
labeled data
by humen/expert
small amount
unlabeled data
large amount of data
a model of human learning
classification
Text classification
Clustering
Lane finding on GPS data
Supervised
task driven
regression
Housing price prediction
classification
Medical Imaging
predict next value
labeled training data
prob(X/Y=labeled-data)
infer a func to map I/O