Modular Neural Network

-
Modular Neural Network

A modular neural network is one that is composed of more than one neural network model connected by some intermediary. Modular neural networks can allow for the sophisticated use of more basic neural network systems managed and handled in conjunction.

• MNN is a collective group of different networks working independently and contributing towards the desired outcome.

• It enhances a unique quality of constructing networks and performing sub-tasks by breaking them.

• The efficiency of this network is high compared to other networks because there are no inter-connected links between them.

• Robustness is quite an advantageous factor in these networks.

• Modular neural networks reduce a single large, unwieldy neural network to smaller, potentially more manageable components.

• It reduces the complexity of the network into smaller components

Issues Leading to Modular Networks:

-> Reducing Model Complexity

-> Incorporating Knowledge

-> Data Fusion and Prediction Averaging

-> Combination of Techniques

-> Learning Different Tasks Simultaneously

-> Robustness and Incrementality

Constructive algorithms for modular networks:

Cascade correlation: This algorithm cascade correlation was proposed by Fahlman and Lebiere to address the issues regarding residual approximation error.

· Cascade Correlation Characteristics :

Ø A supervised learning algorithm is evaluated on its performance via an external source.

Ø A network that regulates its own size and topology.

Ø It consists of input/output layers simultaneously.

Ø It is used to enhance a minimal multi-layer network by creating a hidden layer.

Ø recruits new units according to the residual approximation error.

Cascade algorithm:

1. Starts with a minimal network consisting of an input and output layer.

2. Train the network with a learning algorithm (ie. Gradient Descent, Simulating Annealing).

3. Train until no significant error reduction can be measured.

4. Add the new hidden unit to reduce the residual error.

5. Hidden units are added one by one to the network which is connected by all input units and to every pre-existing hidden unit. 

6. Freeze all incoming weights of the hidden unit.

7. Repeat until the desired performance is reached.

Advantages of cascade correlation :

1. Reduces learning time.

2. Transparent.

3. Creates a structured network.

Disadvantages :
  • Can lead to specialization of just the training sets.
Hybrid networks:

ART architecture: (Adaptive Resonance Theory)

Conclusion :

We finally conclude that:
Modular neural networks can overcome any kind of current non-modular neural network and their drawbacks with ease. However, they are considered to be sensitive and efficiency is based on task decomposition. In this way modular neural networks have numerous ways to approach.











Comments

Post a Comment

Popular posts from this blog

Our First Workshop Went Like...!

-
Image
Here's the story of the QuAIT's first workshop with its peaks and valleys. It took a team of 6 members and active participation from the students of SRI INDU College Of Engineering and Technology(Hyderabad), to make this workshop a success! Wanna know the full story? Read to know more!  THE TEAM: First of all! the team that has toiled day and night for the workshop. T. RISHWI (CEO) T.NITHYA (AI Researcher) A. SAI KUMAR (APP Developer) RAGHAVENDRA(Software Developer) M. SREENIDHI IYENGAR (AI & ML Engineer) V HARIKA (Quantum Researcher) THE PURPOSE: We know for a fact that Engineering students always want to learn something new but also are a bit lazy when it comes to the traditional way of learning things. So we came up with an idea to conduct a 2-day workshop, where we can share our knowledge on the basics of ML/AI to the passionate students who are willing to put an all-out effort in learning new technologies.  And that's when we got to know that SRI INDU College'...
Read more

Convolutional Neural Networks

-
Image
  Hello, everyone! Welcome back to another interesting series on Applied Deep Learning. Last week we came across Artificial neural networks and its implementations. Today we are going to learn about CNN (convolutional Neural Networks). So let's get started… Convolutional Neural Networks(CNN) are seen and implemented everywhere in our day to day life. CNN is probably considered the most desired deep learning architecture. These networks are quite familiar to feed-forward neural networks which we discussed in the last session. CNN undergoes the concepts of weights and biases. Neuron's arrangements are represented in the form of 3-dimensional arrays. The limelight to CNN in recent years is all because of its immense popularity in deep learning and its effectiveness towards convents.  CNN is computationally efficient and it has a unique ability that it surpassed its predecessors in terms of detecting important components without the involvement of human supervision. The spatial ...
Read more

Classification of Animals Using CNN Model

-
Image
Classification of animals using CNN Model In this model, we try to predict whether the animal is a dog or a cat. #We will use the Keras library to create our model. Keras: is an open - source library that provides a python interface for artificial neural networks. from keras.models import Sequential from keras.layers import Conv2D,Activation,MaxPooling2D,Dense,Flatten,Dropout import numpy as np #Let’s initialize a convolutional neural network using the sequential model of Keras. classifier = Sequential() # Adding Convolution Layer classifier.add(Conv2D(32,(3,3),input_shape=(64,64,3))) #Conv2D stands for a 2-dimensional convolutional layer. Here, 32 is the number of filters needed. A filter is an array of numeric values. (3,3) is the size of the filter, which means 3 rows and 3 columns. The input image is 64 64 3 in dimensions, that is, 64 height, 64 widths, and 3 refer to RGB values. Each of the numbers in this array (64,64,3) is given values from 0 to 255, which describes the pixel in...
Read more