Deep Learning

 

Deep Learning is a part of Machine Learning where the learning / model training mimics human brain, this is possible because of the Artificial Network that resembles the human neural network which is thinking pattern of human brain.

Deep Learning can be applied to both Structured and Unstructured data.

Deep Learning applications are again classified to 3 categories same as Machine Learning

  1. Deep Supervised Learning
  2. Deep Unsupervised Learning
  3. Deep Reinforcement Learning

Artificial Neural Networks (ANN):

ANN is the basic structure for any model training through Deep Learning


Any ANN model contains an input layer, hidden layer/s and an output layer, based on the problem statement, the complexity (nodes,hidden layers) of the model will be increased.

Node:

Similar to the Neuron of human brain, Node is a basic computational unit of the ANN.


Activation Function decides the range of output of the node, below are the types of activation functions
  1. Sigmoid
  2. Tanh
  3. ReLU (Rectified Linear Unit)
  4. Leaky ReLU
activation function takes linear combination of inputs of a node multiplied with its respective weights

Please refer to the Single Neuron Implementation Code from the below Git Hub URL

Deep Supervised Learning Architectures:

ANN will be applied for structured data
CNN will be applied for Image/Video data - Computer Vision
RNN will be applied for text/speech data - Natural Language Processing

Comments

Post a Comment

Popular posts from this blog

Supervised Machine Learning

Image
 In Supervised Machine Learning output prediction will be done based on either mathematical or probabilistic approaches , based on output data type the supervised learning problems are classified in to two types Regression (Numerical data) Classification (Categorical data) Algorithms: below are the algorithms available for Supervised Machine Learning: Metrics: Metrics are measure of how good our model is performing during training and testing phases. below are the metrics available for Regression and Classification problems
Read more

What is AI ?

 Artificial Intelligence is introduced to mimic the human intelligence for solving complex tasks humans encounter. Key tools in building AI system: 1. Brain - Experiencing or Learning something from the signals it received 2. Sense Organs - provides several type of inputs to the brains 3. Emotions below are some of the fields in which AI is already operational - Natural Language Processing products: These products are build to use speech or text or voice for getting meaningful insights (applications of NLP can be easily found over the internet) Data generators for NLP : Mouth - speech / audio (artificial equivalent) Ears - speech / audio sensors (artificial equivalent) Computer Vision Products: Computer Vision tools help in getting insights from images / videos (applications of CV can be easily found over the internet) Data generators for CV: Eyes - image / video (artificial equivalent) Structured Data Products: These products are helpful to gain insights / solve tasks easily from...
Read more

Artificial Neural Network

  How training happens in ANN? Training in ML Algorithms: Consider Linear / Logistic Regression models where the training aimed at finding optimum values of coefficients / weights of the linear model (y =WX+B where W denotes the coefficient and C denotes the intercept of the line) Gradient Descent is one of the optimization technique for finding optimum weights of a model using iterative approach          W := W - Learning rate * derivative of Loss w.r.t W          B :=  B   - Learning rate * derivative of Loss w.r.t B W = coefficients of the linear model B = Intercept of the linear model Learning rate = Controllable parameter for getting optimum solution Loss function = Vector difference between predicted and actual outputs once we find optimum W,C value, we can able to perform the predictions in ML through newly trained model. In case of Neural networks, the above coefficients needs to be calculated for every node ...
Read more