Like other autoencoders, contractive autoencoders are trained to represent unlabelled data types through encoding dimensionality reduction. Additionally, the contractive autoencoders, which use unsupervised learning, add analytic penalty while reconstructing the error function. This autoencoder is less sensitive to small variations in the dataset. It is also used for monitoring and analyzing the activity of daily living (ADL). Such autoeoncoders are difficult to standardize.

CNN makes use of the concept of the visual cortex of the biological system and is used for image recognition. Besides, CNN has been used in the medical system for automatic pain recognition during sports, analysis of the relationship between sleep patterns with physical exercise, tracking of personal activities, etc. CNN is a type of artificial neural network that is used immensely in image analysis through deep neural networks. It requires a large dataset to analyze through the deep learning principle.

It’s a class of “generative graphical model“ characterized by multiple layers of hidden layers of latent variables, with a directed connection at the lower layer however, the connection at the two topmost layers is undirected. DBN has been successfully implemented in posture or hand gesture detection and communication with Alzheimer's patients through analyzing the activity of daily living (ADM). The system trains itself through unsupervised learning from the sensor and IoT data, although the process is complex, resources and time consuming

DBM is characterized by a large number of hidden layers with undirected connections between the nodes of two layers. It implies unsupervised learning with feature extraction through a feedback mechanism. DBM has been successfully used in medical devices, like, irregularity in heart beat during strenuous and exhaustive exercise. The challenge in using DBM is joint optimization which is very difficult for large datasets

It’s a kind of deep neural network, featured by effective reconstructing “tainted input values” through training the system on how to utilize specific hidden layers for model reconstruction depending on the obtained inputs. Autoencoders are deep networks that can reconstruct a model through backpropagationand are used for dimensionality reduction. Denoising encoder is again time-consuming and resource-intensive process, although very useful to correct the “corrupt sensor data”

Recurrent neural networks (RNNs) are used to identify the sequential features of input data so that the next pattern can be predicted. It has been used for natural language processing through modeling sequential time series data. The major challenge here is vanishing or exploding gradients.

It’s an artificial neural network (ANN) based autoencoder that trains the algorithm through unsupervised machine learning. These are successfully used for health monitoring. The name “spare” is used for the characterisitics of enforcing sparsity against the loss function to make the modelinvariant to learning applications. This autoencoder is also time and resource intensive.