Deep Learning for IoT Big Data and Streaming Analytics: A Survey (1712.04301v2)

Abstract: In the era of the Internet of Things (IoT), an enormous amount of sensing devices collect and/or generate various sensory data over time for a wide range of fields and applications. Based on the nature of the application, these devices will result in big or fast/real-time data streams. Applying analytics over such data streams to discover new information, predict future insights, and make control decisions is a crucial process that makes IoT a worthy paradigm for businesses and a quality-of-life improving technology. In this paper, we provide a thorough overview on using a class of advanced machine learning techniques, namely Deep Learning (DL), to facilitate the analytics and learning in the IoT domain. We start by articulating IoT data characteristics and identifying two major treatments for IoT data from a machine learning perspective, namely IoT big data analytics and IoT streaming data analytics. We also discuss why DL is a promising approach to achieve the desired analytics in these types of data and applications. The potential of using emerging DL techniques for IoT data analytics are then discussed, and its promises and challenges are introduced. We present a comprehensive background on different DL architectures and algorithms. We also analyze and summarize major reported research attempts that leveraged DL in the IoT domain. The smart IoT devices that have incorporated DL in their intelligence background are also discussed. DL implementation approaches on the fog and cloud centers in support of IoT applications are also surveyed. Finally, we shed light on some challenges and potential directions for future research. At the end of each section, we highlight the lessons learned based on our experiments and review of the recent literature.

Deep Learning for IoT Big Data and Streaming Analytics: A Survey

The paper "Deep Learning for IoT Big Data and Streaming Analytics: A Survey" by Mehdi Mohammadi et al. provides a comprehensive review of how Deep Learning (DL) techniques can be utilized to enhance data analytics in the Internet of Things (IoT) domain. The authors delve into various characteristics of IoT data and categorize it into two main types from a machine learning perspective: IoT Big Data and IoT Streaming Data. This distinction is critical for understanding the appropriate DL techniques that can be applied in different scenarios.

IoT Data Characteristics

IoT data possesses unique features such as large-scale streaming data, heterogeneity, time and space correlation, and high noise levels. The paper categorizes IoT data into two essential types:

• IoT Big Data, characterized by its vast volume, high velocity of production, significant variety, veracity, variability, and value.
• IoT Fast and Streaming Data, which require real-time or near real-time actions due to their continuous nature. The authors emphasize that advanced analytics on this data are necessary to derive actionable insights promptly.

Deep Learning Architectures

The paper provides a detailed overview of several common and cutting-edge DL models suitable for IoT applications:

• Convolutional Neural Networks (CNNs)
• Recurrent Neural Networks (RNNs)
• Long Short-Term Memory Networks (LSTMs)
• Autoencoders (AEs)
• Variational Autoencoders (VAEs)
• Generative Adversarial Networks (GANs)
• Restricted Boltzmann Machines (RBMs)
• Deep Belief Networks (DBNs)
• Ladder Networks

Each model's architecture is thoroughly examined, including its characteristics, learning models, typical input data, and associated IoT applications.

Applications of DL in IoT

The paper identifies and discusses several foundational services in IoT where DL has shown significant promise:

1. Image Recognition: Utilized in applications such as traffic sign detection and analysis of drone images.
2. Speech/Voice Recognition: Employed in intelligent personal assistants and smart appliances.
3. Indoor Localization: Leveraging DL models like CNN and LSTM to improve location accuracy in indoor environments.
4. Physiological and Psychological State Detection: Used in smart home systems to monitor residents' activities and health.
5. Security and Privacy: Enhancing anomaly detection and preventing FDI attacks.

Additionally, the paper explores the role of DL in different IoT verticals such as smart homes, smart cities, energy management, intelligent transportation systems, health and wellbeing, agriculture, education, industry, government, and sports.

DL on Resource-Constrained IoT Devices

A crucial area tackled by the paper is the deployment of DL algorithms on resource-constrained IoT devices. Several methods and technologies to achieve this include:

• Network Compression: Reducing the storage and computational burden by pruning unnecessary network parameters.
• Approximate Computing: Utilizing approximate methods to lower energy consumption.
• Specialized Hardware Accelerators: Designed to improve DL performance on constrained hardware.

Cloud and Fog-Based DL for IoT

The survey discusses cloud computing as a pivotal solution for IoT big data analytics but recognizes its limitations regarding latency and security. In contrast, fog computing is positioned as a viable alternative allowing for localized, low-latency analytics. The paper also highlights the need for specialized platforms and infrastructural advancements to support distributed DL at the edge and cloud environments effectively.

Future Directions and Challenges

Key challenges identified for the future of DL in IoT include:

• Large IoT Datasets: The need for extensive real-world datasets to train DL models.
• Preprocessing Requirements: The complexity of preparing heterogeneous IoT data.
• Security and Privacy: Ensuring secure and privacy-preserving DL models.
• Handling the 6V’s of IoT Data: Addressing the unique volume, velocity, variety, veracity, variability, and value of IoT data.
• Scalable DL Models: Developing models that operate efficiently on both resource-constrained devices and large-scale cloud environments.

The paper emphasizes the necessity for advancements in semi-supervised learning methods to handle the scarcity of labeled data and calls for research into online resource provisioning algorithms to support dynamic, real-time analytics in IoT environments.

In conclusion, this survey offers an extensive exploration of integrating DL techniques within the IoT landscape, underscoring both the potential and the challenges. It provides a solid foundation for researchers aiming to enhance IoT systems' intelligence through advanced DL analytics, propelling smart applications and services to new heights.