Deep Learning: Learning from data
Deep learning is a machine learning technique that allows systems to learn by example. All present day AI systems are based on deep learning.
In continuation of my blog post on Artificial Intelligence, I would like to elaborate on how deep learning works and how it is different from machine learning. I have incorporated illustrations to make the reading more interactive.
What is Deep learning:
Deep learning is a subset of machine learning that imitates humans in processing data and creating patterns for decision making.
Deep learning has the capability to read both structured and unstructured data. Deep learning is the key technology behind driverless cars and voice recognition consumer electronics.
How does it work ?
In deep learning, a model learns to perform classification tasks directly from images, text, video or audio. Models are trained by using a large set of labeled data and neural network architectures that contain many layers. Labeled data is the set of data that is tagged for the system to recognize.
Example: A voice control assistant on a consumer device will be trained with sound/text/videos of different settings and features of the device to perform tasks dictated by the user.
Deep learning models are trained by using large sets of labeled data and neural network architectures that learn features directly from the data without the need for manual feature extraction.
The majority of the deep learning models use neural network architecture. Neural Network is an adaptive system that learns by using interconnected nodes in a layered structure.
A neural network can learn from data and can be trained to recognize patterns, forecast events and classify data. Convolutional Neural Network (CNN), is the most popular neural network.
Deep learning VS Machine learning:
In the machine learning workflow, features are manually extracted and then used to create a model that classifies the objects, whereas in deep learning workflow, feature extraction and classification of models are done automatically. The deep learning model improves as the size of the data increases.
Deep learning requires a huge amount of data including labeled data and high performing GPUs (Graphical Processing Units) to train a model, whereas a machine learning model can be trained based on different techniques that vary from amount of data, application type etc.
Deep learning algorithms:
There are 4 key deep learning algorithms that are widely used depending on their data analysis, data processing and industry use cases.
Algorithms are classified based on the Supervised and Unsupervised learning models part of deep learning architecture.
Supervised learning model:
- Makes predictions iteratively by adjusting data to correlate with the solution intended
- Requires human intervention to label the data
- Learns from the training data set
Unsupervised learning model:
- Works independently to identify relation between unlabeled data
- Data analyst needs to analyze output and recommendations
- Convolutional Neural Network (CNN): CNN is a deep learning algorithm that analyzes the input image/object and categorizes it. This helps to differentiate the object from others during the tests.
CNN is mainly used for image processing, segmentation and classification of data
2. Recurrent Neural Network (RNN): RNN uses time series or sequential data for analysis. These algorithms are widely used in ordinal searches, such as voice recognition and language translations.
3. Deep Belief Network (DBN): DBN is a novel training algorithm which comprises of multiple hidden layers in which each connected layer is an RBM (a two layered neural network).
DBN is extensively used for collaborative filtering, feature learning and applications of quantum mechanics
4. Deep Boltzmann Machine (DBM): DBM is comprised of multiple layers of hidden layers with no inter-connections. DBM is an unsupervised and probabilistic model with unrelated connection between different layers.
To conclude, in this blog post we learned about what deep learning is and how it works in brief, followed by deep learning algorithms and their uses.
I hope you enjoyed reading this article and you know the drill — clap, comment and share.
Cheers, Karthik
