Commit 088f91c9 authored by pacalet's avatar pacalet
Browse files


parent 76fbfa62
......@@ -150,5 +150,29 @@ Students will become experienced with languages dedicated to digital audio proce
- Evaluation: We decide the "evaluation day" and we display some images on a big screen. Then, given a metric (like F1 score), each group will be evaluated on both the image classification of their image processing program, and the classification of their consensus algorithm.
- Target platform: Raspberry Pi 4 and camera
### High Level Idea and Application
- Proposer: Derya
Distributed agents (RPIs) in a wireless network collectively try to achieve a common goal, e.g., reconstructing a very large IMAGE from distributed pixels at a central agent.
- Each RPI holds a subset of pixels. However, it does not know the information of its coordinates
- Each RPI will be given a data vector, containing the ID and the coordinate information of only a subset of RPIs (local neighborhood). A collection of RPIs might have shared (repeated) information about IDs and the coordinates.
- Each RPI can communicate only with its local neighborhood and exchange information pertaining to the ID and the coordinates.
- There is a central agent that collects the coordinate information from the RPIs. Once the pixel coordinates are collected, it can reconstruct the IMAGE.
_Communication inefficient._ One trivial approach with no cooperation is the following. Each RPI transmits directly to the central agent its vector in its entirety. Since RPIs have redundant data, this is not the most efficient way of reconstructing the IMAGE.
_Storage inefficient._ Another trivial approach could be full cooperation, where each RPI transmits its whole data vector with its neighbors and each RPI learns the IDs and the coordinates of the rest of the agents. Then, it suffices if one designated RPI communicates the coordinates to the central agent.
_Balance._ An improved approach could be to have a constraint on cooperation such that each agent can only gather a restricted amount of information, e.g., 1 bit from each neighbor, and use this information to decide what to transmit to the central agent.
_Your solution._ Is there a better approach that provides a low storage utilization as well as low communication complexity?
- How does cooperation help? More specifically, what is the tradeoff between the number of bits exchanged among the agents and the total number of bits to be communicated to the central agent to reconstruct the IMAGE?
- How does the central agent know whether the constructed IMAGE is accurate? What is a good measure of accuracy?
- Can we incorporate learning into this problem?
<!-- vim: set tabstop=4 softtabstop=4 shiftwidth=4 noexpandtab textwidth=0: -->
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment