The latest research on AI in Telecom domain

Source: Deep Learning on Medium

The latest research on AI in Telecom domain

Telecom domain seems to be catching up with AI in recent times. I see a renewed interest of AI in the telecom domain, especially after the 5G deploying picking pace. The number of research articles and papers have considerably increased in IEEE and arxiv. Recently IEEE has published an IEEE network issue focused on AI in the telecom domain. It had an interesting collection of research work that could benefit the feature of telecom especially the wireless domain.

I went through the papers listed in the issue and captured below a brief summary of those papers to give a sense of the use cases and developments happening in this area.

Title: “Label-less Learning for Traffic Control in an Edge Network.”
Link: https://ieeexplore.ieee.org/document/8553648
DOI: 10.1109/MNET.2018.1800110
TL;DR: An approach to minimize the number of features/data sent over the network for AI model training and prediction while maintaining model intelligence in a centralized cloud. Filters and extracts required features at Edge and send it to the core for the AI engine to process.

Title: “AI-Based Malicious Network Traffic Detection in VANETs,”
Link: https://ieeexplore.ieee.org/document/8553649
DOI: 10.1109/MNET.2018.1800074
TL;DR: A hybrid approach for intrusion or DoS detection of Cooperative Intelligent Transportation Systems (C-ITS) applications in a Vehicular ad-hoc network (VANETs).

Title: “An Improved Stacked Auto-Encoder for Network Traffic Flow Classification”
Link: https://ieeexplore.ieee.org/document/8553650
DOI: 10.1109/MNET.2018.1800078
TL;DR: Used modified stacked auto-encoder-decoder together with Bayesian probability model for traffic type classification. Got a better accuracy of about 82–84%, with is about 2% greater than just stacked auto-encoder-decoder model.

Title: “A Deep-Learning-Based Radio Resource Assignment Technique for 5G Ultra Dense Networks”
Link: https://ieeexplore.ieee.org/document/8553651
DOI: 10.1109/MNET.2018.1800085
TL;DR: Uses LSTM to predict the number of packets expected in the future based on the number of packets received so far. This information is used to predict congestion before it could happen and adjust UL/DL ratios to avoid congestion.

Title: Deep Reinforcement Learning for Multimedia Traffic Control in Software Defined Networking”
Link: https://ieeexplore.ieee.org/document/8553652
DOI: 10.1109/MNET.2018.1800097
TL;DR: In an SDN network instead of a rule-based scheduler, it uses an AI scheduler trained using reinforcement learning. The AI scheduler uses current network state and traffic characteristics as input and QoE estimated from QoS is used to calculate rewards.

Title: “Improving Traffic Forecasting for 5G Core Network Scalability: A Machine Learning Approach”
Link: https://ieeexplore.ieee.org/document/8553653
DOI: 10.1109/MNET.2018.1800104
TL;DR: Uses an LSTM and Deep Neural Network (DNN) to forecast traffic in core components and scale-up in advance to handle an increase in traffic.

Title: “Deep Reinforcement Learning for Mobile Edge Caching: Review, New Features, and Open Issues”
Link: https://ieeexplore.ieee.org/document/8553654
DOI: 10.1109/MNET.2018.1800109
TL;DR: Uses Deep-Reinforcement-Learning to learn and predict which data needs to be cached. The algorithm learns on the job by making decisions in live network and cache hit vs misses gives positive and negative reinforcement to the AI model.

Title: “Artificial Intelligence to Manage Network Traffic of 5G Wireless Networks”
Link: https://ieeexplore.ieee.org/document/8553655
DOI: 10.1109/MNET.2018.1800115
TL;DR: Captures high-level challenges that AI could solve in the 5G network and proposes an AI-based approach for predicting what content should be cached at the edge.

Title: “SeDaTiVe: SDN-enabled Deep Learning Architecture for Network Traffic Control in Vehicular Cyber-Physical Systems”
Link: https://ieeexplore.ieee.org/document/8553657
DOI: 10.1109/MNET.2018.1800101
TL;DR: Uses SDN based solution for optimal routing of network traffic in a vehicular cyber-physical system (VCPS) network by integrating an AI model to identify hidden patterns in data packets and choose an optimal route.

Title: “Intelligent Context-Aware Communication Paradigm Design for IoVs Based on Data Analytics”
Link: https://ieeexplore.ieee.org/document/8553658
DOI: 10.1109/MNET.2018.1800067
TL;DR: Uses AI for optimizing routing packets in Internet-of-Vehicle networks. Uses traffic patterns to detect the two vehicles are not-in-line-of-sight and make changes for efficient access medium allocation, route determination and reliable delivery of safety messages.

Title: “Artificial Intelligence Inspired Multi-Dimension Traffic Control for Heterogeneous Networks”
Link: https://ieeexplore.ieee.org/document/8553660
DOI: 10.1109/MNET.2018.1800120
TL;DR: Classify traffic type across multiple access networks. Does a clustering using DBSCAN on traffic across access networks and uses a neural network classify traffic types of each cluster to give a label.

Title: “Living with Artificial Intelligence: A Paradigm Shift toward Future Network Traffic Control”
Link: https://ieeexplore.ieee.org/document/8553661
DOI: 10.1109/MNET.2018.1800119
TL;DR: Provides a state-of-art review of the AI-based approach for network traffic control and highlights the importance of reinforcement learning-based AI model for traffic control.

Title: “Multimedia Data Flow Traffic Classification Using Intelligent Models Based on Traffic Patterns”
Link: https://ieeexplore.ieee.org/document/8553662
DOI: 10.1109/MNET.2018.1800121
TL;DR: Proposes an AI model to predict video quality measures using the video stream and network characteristics as input. It extracts patterns that affect objective quality perceived by the user and predicts video quality as perceived by the end-user.

Title: “DeepTP: An End-to-End Neural Network for Mobile Cellular Traffic Prediction”
Link: https://ieeexplore.ieee.org/document/8553663/
DOI: 10.1109/MNET.2018.1800127
TL;DR: Uses a seq2seq model with an attention mechanism to forecast mobile traffic from spatial-dependent and long-period cellular traffic.