Neural network - Revision history

Rice at 19:35, 17 May 2024

2024-05-17T19:35:17Z

← Older revision		Revision as of 19:35, 17 May 2024
Line 1:		Line 1:
			[[Category:Machine Learning]]
	A '''neural network''' is a type of machine learning model that utilizes layers of connected '''neurons''' to perform classification and regression.		A '''neural network''' is a type of machine learning model that utilizes layers of connected '''neurons''' to perform classification and regression.

Rice: /* Loss function */

2024-05-01T03:20:49Z

Loss function

← Older revision		Revision as of 03:20, 1 May 2024
Line 19:		Line 19:

	Any layer in between are called '''hidden layers'''.		Any layer in between are called '''hidden layers'''.

			= Training =
			Training of a neural network consists of [[Feed-forward pass\|feed-forward passes]] to compute the network output and [[back propagation]] to calculate the error.

	= Loss function =		= Loss function =

Rice: /* Constraints */

2024-05-01T01:15:03Z

Constraints

← Older revision		Revision as of 01:15, 1 May 2024
Line 26:		Line 26:
	= Constraints =		= Constraints =
	Neural network relies on the input being ''standardized'' and numerical. [[One-hot encoding]] and [[label encoding]] can help encode categorical data into numerical ones.		Neural network relies on the input being ''standardized'' and numerical. [[One-hot encoding]] and [[label encoding]] can help encode categorical data into numerical ones.

			Neural networks are robust and resilient to noise. They are also good at handling non-linear stuff.

			On the other hand, they are somewhat a black box, computationally expensive, and prone to overfitting.

	= Classification =		= Classification =

Rice: /* Classification */

2024-05-01T00:24:43Z

Classification

@@ Line 23: / Line 23: @@
 [[File:Covex vs. Non-convex.png|thumb|Figure 1. Convex vs. non-convex loss function. Non-convex has local minimums that needs to be filtered out.]]
 The loss function in ANN is usually ''non-convex'', making minimization more difficult than setting the gradient to 0.
 = Classification =
@@ Line 30: / Line 33: @@
 * [[Artificial neural network]] has 3 layers
 * Deep neural network has a lot of layers.

Rice at 23:52, 30 April 2024

2024-04-30T23:52:19Z

← Older revision		Revision as of 23:52, 30 April 2024
Line 19:		Line 19:

	Any layer in between are called '''hidden layers'''.		Any layer in between are called '''hidden layers'''.

			= Loss function =
			[[File:Covex vs. Non-convex.png\|thumb\|Figure 1. Convex vs. non-convex loss function. Non-convex has local minimums that needs to be filtered out.]]
			The loss function in ANN is usually ''non-convex'', making minimization more difficult than setting the gradient to 0.

	= Classification =		= Classification =

Rice: Created page with "A '''neural network''' is a type of machine learning model that utilizes layers of connected '''neurons''' to perform classification and regression. = Structure = == Neuron == The '''neuron''' is the data-processing unit that composes a neural network. It combines information from multiple neurons (or input) and become activated or deactivated. The output is determined by the following steps: # A neuron generates a linear combination of input data, ''z''. This is done..."

2024-04-30T23:50:50Z

Created page with "A '''neural network''' is a type of machine learning model that utilizes layers of connected '''neurons''' to perform classification and regression. = Structure = == Neuron == The '''neuron''' is the data-processing unit that composes a neural network. It combines information from multiple neurons (or input) and become activated or deactivated. The output is determined by the following steps: # A neuron generates a linear combination of input data, ''z''. This is done..."

New page

A '''neural network''' is a type of machine learning model that utilizes layers of connected '''neurons''' to perform classification and regression.

= Structure =

== Neuron ==
The '''neuron''' is the data-processing unit that composes a neural network. It combines information from multiple neurons (or input) and become activated or deactivated. The output is determined by the following steps:

# A neuron generates a linear combination of input data, ''z''. This is done with a set of weights.
# ''z'' is fed into an [[activation function]]
# The activation function determines whether the neuron is activated (1) or not (0)

== Layers ==

Many neurons form a '''layer''', and many layers form a neural network. Each neuron takes input from neurons of the previous layer, and send outputs to the next layer.

The first layer of an NN is the '''input layer'''. In contrast to other layers, each neuron in the input layer takes one attribute as their input. As such, the size of the input layer is the same as the number of attributes used to train the model.

The last layer of an NN is the '''output layer'''. Pretty self explanatory.

Any layer in between are called '''hidden layers'''.

= Classification =

The number of layers and the size of each layer determines the complexity of the model.

* [[Artificial neural network]] has 3 layers
* Deep neural network has a lot of layers.