Main Tasks

• Task 1: Bandlimited Coordinate Descent: Learning-driven Joint Power Allocation and Gradient Estimation.
• Task 2: Bandlimited Gradient Sketching: Power Control and Bandwidth-Accuracy Tradeoffs.
• Task 3: Bandlimited Coordinate Descent based on Zero-order Estimates.
• Task 4: Bandlimited Gradient Sketching based on Zero-order Methods.
• Task 5: Exploiting RF Characteristics in DL-Based CSI Estimation.

Figure 1. Flow of research tasks.

Management Plan

This project will investigate two sets of edge learning algorithms over wireless MAC channels, namely 1) bandlimited coordinate descent algorithms and 2) bandlimited gradient sketching algorithms. We will devise these algorithms using two intimately related approaches, namely 1) first-order stochastic gradient descent methods (which consists of two detailed tasks Tasks 1, 2 and 3) zero-order stochastic optimization methods (which consists of two detailed tasks Tasks 2 and 4. Task 5 will be dedicated to investigating machine learning aided channel estimation. The PIs will evaluate the performance of the proposed learning framework for wireless edge computing, using simulation (Matlab/Python), experiments, and prototype development. The year-by-year research plan for the proposed research effort is presented in Figure 2. The figure refers explicitly to the tasks highlighted throughout the proposal. This proposal has laid out an ambitious agenda. We expect our team to work together on all tasks; however, a task lead has been identified to ensure progress on each task.