Deep Learning Book Flashcards
What are the three ingredients to machine learning?
1) Data (a set of historical examples)
2) A set of functions that the algorithm will search through to find the best match with the data.
3) Some measure of fitness that can be used to evaluate how well each candidate function matched the data.
Why is machine learning difficult?
1) Most data sets will include noise.
2) It’s often the case that the set of possible functions is larger than the set of examples in the dataset.
How do you overcome an ill-posed machine learning problem?
ill-posed- set of possible functions is greater than the number of examples
1) Gather more examples
2) use inductive bias’s
What is an inductive bias?
An assumption about the characteristics of the best function.
Example: if all the swans that you see in your life are white, you might assume that a general rule of thumb is that “all swans are white”
What is a function?
A deterministic mapping from a set of input values to one more more output values.
What is supervised learning? ( My definition)
Each example in the dataset is labeled with the expected output (or target) value. The algorithm uses the output values by comparing the outputs a function gives with the output values in the data set and using the difference (or error) to evaluate the fitness of the candidate function.
The algorithm compares info given from the labeled output and the functions output to determine how good the function is and how to make adjustments.
What is supervised learning? ( Books definition)
A form of machine learning where the goal is to learn a function that maps from a set of input attributes for an instance to an accurate estimate of the missing value for the target attribute of the same instance.
Target Attribute
In supervised machine learning, a target attribute is the attribute that the mode is trained to estimate the value of.
What is unsupervised learning?
A form of machine learning where the goal is to identify regularities, such as clusters of similar instances, in the data. Unlike supervised learning, there is no target attribute in an unsupervised learning task.
What is the measure of fitness?
It is a function that takes the outputs from a candidate function, generated when the machine learning algorithm applies the candidate function to the data, and compares these outputs with the data in some way.
What is reinforcement learning?
The goal is to enable an agent to learn a policy (function) on how it should act in a given environment. The agent learns in the environment and updates based on the reward of the result.
Example: should to program or the pawn or take the queen
What is a Neuron?
Is a simple model in its own right that maps from a set of inputs to an output.
