Enabling Distributed Applications with Online Machine Learning

  • * Register (or log in) to the Neural Network to add this session to your agenda or watch the replay

  • Date
    17 July 2024
    Timeframe
    17:00 - 18:00 CEST Geneva
    Duration
    60 minutes
    Share this session

    With the recent proliferation of the Internet of Things (IoT), more and more devices now have computing capabilities and Internet connections. While these newfound capabilities have enabled a multitude of emerging applications, e.g., large-scale machine learning that takes as input data collected by many IoT sensors, they also raise new challenges for ensuring that such applications receive the data, computing, and communication resources that they need. Some data analytics tasks, for example, may require significantly more processing capabilities than others, and these capabilities may not always be available depending on the status of devices in the network. Optimizing such resource allocations across heterogeneous applications is in general NP-hard and becomes even more challenging in a dynamic and uncertain environment in which user demands and resource availability may change in unknown ways over time. In this talk, I will present our recent work using online and reinforcement learning techniques to provision both computing and communication resources for heterogeneous applications. By incorporating prior knowledge of the problem structure and application requirements, we can significantly accelerate our ability to learn how to allocate resources without requiring prior models of the environment. Our experiments on distributed machine learning and autonomous vehicle applications indicate that we can improve application performance and utilize fewer resources compared to static and naive learning-based baselines. 

  • Share this session
    Discover more from programme stream
    Integrated AI and wireless for sustainable mobile network evolutions

    Integrated AI and wireless for sustainable mobile network evolutions

    11 March 2025 - 13:00 to 14:00