comparison of machine learning and deep learning. Finally, the paper ends with the conclusion and scope for the future.
Keywords: Deep learning, Architectures, Machine learning, Neural Networks.
1 INTRODUCTION
Machine learning is the subset of Artificial Intelligence (AI) that trains the system, automatically learn without programming it. It learns with observations like preparing patterns from data that produce results and outputs for future. Deep learning depends on machine learning algorithms. Deep learning works on the artificial neural network-ANN.
ANNs use learning algorithms and increase the amounts of data, and in turn the training process gets better. We call it deep learning as number of neural networks increases with time. The learning process here are training and inferring, training includes large data labels and matching their characteristics and inferring makes conclusion and label new unknown data with the help of previous knowledge. It is also called as deep structured learning and hierarchical learning that has layers which includes nonlinear processing units used for conversions and feature extraction.
Fig.1 source google.com
Every subsequent layer takes the results from the previous layer as the input. The learning process is either supervised or unsupervised process. Deep learning uses neuron as functional computational unit, the neuron that takes multiple signals as input. The signals combine with the weight and transfers the combined signals to generate outputs. In deep learning, deep means the numerous layers through which the data is transformed. It has a credit assignment path- CAP depth that show conversions from input to output and also connection between them. It also describes that a machine can transforms its internal attributes that compute the details in each layer, by accepting the abstractions and representations from the previous layer. This learning approach is used in the fields of social network analysis, autonomous driving, natural language processing, sentiment classification, visual data processing, computer vision, biomedicine, disaster management, speech recognition and information retrieval systems.
Fig. 2 source google.com
Deep learning model uses true data to find the uniqueness, and then constructs a new model to obtain a variety of applications. The main aspect of deep learning practice is supervised or unsupervised learning and multilayer non-linear processing. Nonlinear processing in multiple layers is a hierarchical method in where the present layer accepts the results from the previous layer and passes its output as input to the next layer.
Hierarchy is calculated among layers to organize the importance of the data. Supervised and unsupervised learning are linked to the class target label. Its presence is a supervised system and absence is an unsupervised system. Soniya et al. [2] discussed current trends, architecture, models, and the limitations of deep learning. They focussed on the characteristics like learning techniques. They elaborated the challenges for the deep learning.
2 ARCHITECTURES OF DEEP LEARNING
There are different references to deep learning architectures that include recurrent neural networks, convolution neural network, deep belief networks and deep stacked network, gated recurrent unit. Deep learning also referred as Deep Neural Network-DNN is constructed by adding multiple layers that are hidden layers in between the input layers and the output layers of Artificial Neural Network-ANN with different topologies. There are various architectures that are needed to implementing the concept.
2.1 Recurrent Neural Network
Recurrent neural networks[3] have been used for modelling variable-length sequences.
RNNs have also been used for various tasks like language modelling, learning word embeddings, online handwritten recognition and speech recognition. The two novel architectures of a deep RNN which are orthogonal to an earlier attempt of stacking multiple recurrent layers to build a deep RNN[4] (Schmidhuber, 1992; El Hihi and Bengio, 1996). We provide an alternative interpretation of these deep RNNs using a novel framework based on neural operators.
2.3 Deep Belief Network
Deep Belief Network is an algorithm among deep learning. A method of solving the problems from neural network with deep layers, like low velocity and the over fitting phenomenon in learning. Neural network has developed into a great subject since its creation and there emerging a lot of kinds of neural networks, for example BP-networks, SOMF-networks and RBM-networks and so on. These neural networks are suitable for different situations. How to process [6] a Deep Belief Network by using Restricted Boltzmann Machines. What is more, we will combine the Deep Belief Network together with soft max classifier, and use it in the recognition of handwritten numbers. Neural networks have played an important role in many fields such as object classification and data fitting due to its powerful self-learning and adaption.
2.4 Deep Stacked Network, Gated Recurrent Unit
A stacked auto-encoder - SAE model is constructed by stacking numerous auto-encoders that are the typical feed-forward neural networks [7]. A basic auto-encoder has two stages, encoding and decoding stage, as shown in figure
Fig. 3 Source: google.com
3 CHARACTERISTICS OF DEEP LEARNING
Deep learning is a wide-ranging term used for the machine learning-ML and artificial intelligence- AI. Deep learning techniques are becoming successful in lot of applications as they have varied characteristics [8];
Solves high computational tasks.
Has great learning ability.
Data set creation is more effective.
Useful tools in varied sectors.
Based on neural networks.
It ensures optimized results.
Learn feature extraction methods from the data.
Surpass human ability to solve highly computational tasks.
Great prediction performance.
Depends on network structure, activation function, data representation.
Extract features from sensory data.
It has great feature representation than a machine learning model.
Great approaches in the big data era[7].
Prior knowledge not required by deep learning networks.
Can extract complicated features, with high level abstraction.
4 WHY USE DEEP LEARNING
Deep learning methods and applications [9] provide different signal and information processing tasks. Most of the applications are executed based on few cadres like knowledge of researchers, applications involved in successful use of technology like computer vision, social network analytics etc, and scope in the growth of research. It involves multitask deep learning with multimodal information. The motivation to use deep learning includes; deep learning is driving Artificial Intelligence-AI to the enterprise level, deep learning techniques that are based on ANN has increased functionality. Deep learning technology has challenged to enhance the performance, as an example we consider handwriting recognition of the machines needs human level of performance, face recognition and the object recognition metrics also require the same. Convolutional neural network- CNN’s architecture has read hand written postal codes. NVIDIA [9] has influenced the space in 2017 as it involves deep Learning ecosystem. Intel Xeon Phi solutions are buried on influx with respect to deep learning.
5 COMPARISON OF DEEP LEARNING WITH MACHINE LEARNING
Deep learning architecture is built with many hidden layers and multiple neurons in each layer. The multilayer architecture helps with the mapping of the input to higher level representation. The differences between two learning techniques;
Deep learning needs hardware for resulting in high performance.
The problem solving in deep learning is done on end-to-end basis but in machine learning it is done by decomposing a bigger task into smaller tasks and then combining the results.
Deep learning needs huge amount of data whereas machine learning requires a small amount of data to work and conclude.
In deep learning time required to train is more when compared to machine learning.
Graphical processing units needed by deep learning are expensive.
Transparency goes with machine learning methods rather than the deep learning methods.
Deep learning creates new features by its own processes and techniques, whereas machine learning, features are accurately and precisely recognized by the users.
6 APPLICATIONS
6.1 Social Network Analysis
Machine learning with social network analysis-SNA [10] can control social media data to better respond to critical situations. Using deep learning, misuse of twitter data was monitored. The effect on cost also was noticed. We use deep learning (DL)- a subset of machine learning, to classify text content of some tweets, and we integrate that with SNA to
inertial sensors. These observations are used by the car’s computer to make driving decisions. The basic block diagrams of an AI powered autonomous car are shown in Figure.
Fig. 4 Source: google.com
The driving decisions are calculated in a modular perception-planning-action pipeline -Fig. 4(a), or in an End2End learning fashion-Fig. 4(b), where sensory information is directly mapped to control outputs. The components of the modular pipeline can be designed based on AI and deep learning methods, or by classical non-learning processes. A deep learning-based object detector gives input to a classical A-star path planning algorithm. A safety monitor is designed to assure the safety of each module. The modular pipeline in Fig. 4(a) is hierarchically decomposed into four components which can be designed using either deep learning and AI approaches, or classical methods. These components are: Perception and Localization, High-Level Path Planning, Behaviour Arbitration, or low-level path planning, Motion Controllers.
6.3 Natural Language Processing
Natural language processing (NLP) is a subfield of linguistics, computer science, and artificial intelligence concerned with the interactions between computers and human language, in particular how to program computers to process and analyse large amounts of natural language data. The result is a computer capable of "understanding" the contents of documents, including the contextual nuances of the language within them. The technology can then accurately extract information and insights contained in the documents as well as categorize and organize the documents themselves.
6.4 Sentiment Classification
The social networks, forums, review sites and blogs provides lot of data like the users views, opinions, emotions about different social events, goods, brands, and politics. Sentiments of
users has impact on the readers. The unstructured form of data from the social media must be analyzed and well-structured and so, sentiment analysis has recognized significant attention. Sentiment analysis[12] is referred as text organization that is used to classify the expressed mind-set or feelings in different manners such as negative, positive, favourable, unfavorable, thumbs up, thumbs down, etc. The challenge for sentiment analysis is lack of sufficient labeled data in the field of Natural Language Processing (NLP). And to solve this issue, the sentiment analysis and deep learning techniques have been merged because deep learning models are effective due to their automatic learning capability. This Review Paper highlights latest studies regarding the implementation of deep learning models such as deep neural networks, convolution neural networks and many more for solving different problems of sentiment analysis such as sentiment classification, cross lingual problems, textual and visual analysis and product review analysis, etc.
6.5 Computer Vision
Computer vision solutions combined with artificial intelligence algorithms that achieved important results in the detection of patterns in images. This work presents a systematic review that aims to identify the applicability of computer vision in precision agriculture for the production of the five most produced grains in the world: maize, rice, wheat, soybean, and barley. In this sense, we present 25 papers selected in the last five years with different approaches to treat aspects related to disease detection, grain quality, and phenotyping.
From the results of the systematic review, it is possible to identify great opportunities, such as the exploitation of GPU (Graphics Processing Unit) and advanced artificial intelligence techniques, such as DBN (Deep Belief Networks) in the construction of robust methods of computer vision applied to precision agriculture.
7 CHALLENGES OF DEEP LEARNING
Deep learning techniques have been solving various complex applications with multiple layers and high level of abstraction. In today’s scenario, the technology has to accept many challenges.
• Deep learning algorithms have to focus on input data.
• Transparency of algorithms.
• Storage requirements.
• Improved methods for big data analytics. Deep networks are called black box networks.
• High computation power.
• Requires large amount of data.
• Expensive for the complex problems and processing.
• No strong theoretical foundation.
• Difficult to find the topology, training parameters for the deep learning.
• Deep learning provides new tools and infrastructures for the computation of the data and enables computers to learn objects and representations.
8 CONCLUSION AND FUTURE SCOPE
Deep learning is important application of machine learning. The uses of the algorithms of deep learning in different fields have shown success. Deep learning depends on the optimization of applications in machine learning and its creation on hierarchical layer process. Deep learning gives effective results for different applications such as speech recognition. Deep learning is a current subject in artificial intelligence. We conclude that with the increased availability of data and computational resources, the use of deep learning in many applications is openly taking to the acceptance. The technology is in boom presently and the advancement of deep learning in more in natural language processing, remote sensing, agriculture and healthcare.
REFERENCES
1. Razvan Pascanu1, Caglar Gulcehre1, Kyunghyun Cho2 and Yoshua Bengio1, How to Construct Deep Recurrent Neural Networks. https://arxiv.org/pdf/1312.6026.pdf?source=post_page--- -deep.
Engineering , https://doi.org/10.1007/s11831-019-09344
9. Li Deng, Dong Yu, Deep Learning: Methods and Applications, Foundations and Trends in Signal Processing-June2014 https://doi.org/10.1561/2000000039.
10. Qurat Tul Ain, Mubashir Ali, Amna Riazy, Amna Noureenz, Muhammad Kamranz, Babar Hayat and A.
Rehman, Sentiment Analysis Using Deep Learning Techniques: A Review, (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 8, No. 6, 2017.
11. Sorin Grigoresen, Bogdan Trasnea, Tiberiu Cocia, Grigel Macesam, A survey of Deep Learning Techniquies for Autonomous Driving, arXiv:1910.07738v2 [cs.LG] 24 Mar 2020.