Coming Up: https://aiforgood.itu.int/event/itu-ai-ml-in-5g-grand-challenge-finale-2/


Collaboration Options

  1. Hosts
    1. Propose a problem →  achieve better solution
  2. Advisory for Problem(s)
    1. Relevance, ML-Technology, etc.
  3. Evaluation Panel
    1. Solution evaluation.
  4. Competitor
    1. Research-group/Researchers participating.
  5. Webinars/Talks
    1. Relevant to the problems.
    2. Present or attend.
  6. Engage in Programming Challenges
    1. BuildAThon
    2. Mentoring in this challenge.

Contact Vishnu Ram:  Vishnu Ram OV <vishnu.n@ieee.org>

ITU-T Secretariat Contacts.

Problems, Hosts, Competitors and Datasets

Host-Based Categorization

Sl. no

Host

2020

2021

1

Federal University of Pará (UFPA), Brazil

beam-selection: Machine learning applied to the physical layer of millimeter-wave MIMO systems

Reinforcement learning: scheduling and resource allocation

2

UPF, Spain

Improving the capacity of IEEE 802.11 WLANs through machine learning

1.Finding groups of BSSs

2.Federated Learning for Spatial Reuse in WLANs

3

BNN‑UPC, Spain

Graph Neural Networking (GNN) Challenge 2020

GNN Challenge. 2021: Creating a Scalable Network Digital Twin

4

North Carolina State University, US

channel estimation: Machine learning applied to the physical layer of millimeter-wave MIMO systems

Localization: Multidevice localization with mmWave signals in a factory environment

5

NEC

Network state estimation by analysing raw video data

Location estimation using RSSI of wireless LAN

6

KDDI

Analysis on route information failure in IP core networks by NFV-based test environment

Network failure detection and root cause analysis in 5GC by NFV-based test environment

7

Turkcell

Using weather info for radio link failure (RLF) prediction

Radio Link Failure Prediction

8

St. Petersburg State Univ. of Telecom.

Traffic recognition and Long-term traffic forecasting based on AI algorithms and metadata

Forecasting Model for Service Allocation Network Using Traffic Recognition

9

China Unicom

1.Fault localization of loop network devices based on MEC platform

2.Energy-saving prediction of base station cells in mobile comm. network

3.Core network KPI index anomaly detection

4.Alarm and prevention for public health emergency based on telecom data

5.Configuration knowledge graph construction of loop network devices based on MEC architecture

Network anomaly detection based on logs

10

China Mobile

1.Network topology optimization

2.Out of service (OoS) alarm prediction of 4/5G network base station

Cross Layer user experience optimization – Radio link performance prediction

11

NIST


WALDO (Wireless Artificial intelligence Location DetectiOn): sensing using mmWave communications and ML

12

Xilinx


Hardware-Efficient Modulation Classification with RadioML

13

Univ. of Alabama


RF-Sensor Based Human Activity Recognition

14

ITU: FG-AN


Network resource allocation for emergency management based on Closed loop analysis

15

ZTE


Delivery route optimization


Problem-Domain based Categorization

The categories of the problems are as follows:
RAN: Radio Access N/W

PHY: Physical Layer

NFV: Network Functions Virtualization

WLAN: Wireless LAN(802.11)

TE: Traffic Engineering

AR: Artificial Reality


Sl. NoCategory of the ProblemsProblemsHost#of CompetitorsDataset
1

RAN

ITU-ML5G-PS-016: Location estimation using RSSI of wireless LAN

RISING Japan38The training data includes the location of AP, RSSI information within the coverage of the AP and its corresponding location. Link is not available.
2NFVNetwork failure detection and root cause analysis in 5GC by NFV-based test environment. Link KDDI Japan35Did not find the public link.
The data sets used for this challenge were created in the NFV-based test environment simulated for 5GC. In this sense, they are synthetic data, but as similar as the real data, resulting from our NFV-based test environment.

3


Federated Learning for Spatial Reuse in a multi-BSS (Basic Service Set) scenario Link

UPF22 A dataset generated with the Komondor simulator is provided to train ML models. Link
4RAN

Radio Link Failure Prediction Link

Turkcell35Zip file contains the following tab separated files (tsv):
 distances.tsv: pair-wise distances
 met-forecast.tsv: meteorology 5-day forecasts
 met-real.tsv: meteorology historic realizations
 met-stations.tsv: meteorology station information
 rl-kpis.tsv: radio link KPIs and configuration parameters
 rl-sites.tsv: radio link site information
GitHub: Link
Training Dataset: Link
Exemplary Colab Project by Turkcell: Link

Build-a-thon(PoC) Network resource allocation for emergency management based on closed-loop analysis Link

ITU Focus Group on Autonomous - Networks26

Powerpoint slides for the progress of teams during the challenge

Link

6TE

ITU-ML5G-PS-011: Combinatorial Optimization Challenge:- Delivery route optimization Link

ZTE19

The dataset is provided in the Dataset module.
Each sample in the dataset is composed of 2 elements:

  1. A json object that describes the graph of the transportation network, see the below example:
    {
    nodes:[0,1,2] ,# id list of the transportation nodes
    edges:[
    (0,1):{ # (starting point, end point)
    lanes: [0,1,2], # id list of the lanes, the length of the array is the number of the lanes between the two adjacent nodes
    lane_weights:[10, 10, 20] # The weight that each lane can carry
    }
    ]
    }
  2. A json object that describes the starting point, endpoint, weight information of each delivery, like:[{“a_end”: 138, “z_end”: 206, “weight”: 1}, {“a_end”: 74, “z_end”: 156, “weight”: 1}]
7TE

Forecasting Model for Service Allocation Network Using Traffic Recognition Link

The Bonch-Bruevich Saint Petersburg State University of Telecommunications16The training data in pcap format can be downloaded here.
8PHY

RF-Sensor Based Human Activity Recognition Link

The University of Alabama24

Github with Data Download:
https://github.com/ci4r/CI4R-Activity-Recognition-datasets

Watch the webinar here: https://aiforgood.itu.int/events/understanding-how-people-move-using-modern-civilian-radar/


ML5G-PHY-Localization: Multidevice localization with mmWave signals in a factory environment Link

North Carolina State University17

The training data set consists of a collection of channels associated to the links between devices and access points in a factory environment and also their associated positions in the room. The factory environment has been simulated by ray tracing, with 12 access points (AP) located at the ceiling. The devices are randomly placed at the room and assigned to the AP with highest gain during the sector level sweep. For simplification, we assume perfect synchronization, this means that the first path arrives to the AP at time 0. The beam-training is measured. Then, with probabilities 50%, 25% and 25%, 0, 1 or 2 other devices interfere with the beam-training through random 4-QAM OFDM symbols using 512 frequency carriers.

Two datasets are provided, namely “train” and “test”. The dataset “train” consists on all the information required for the training of your algorithm, this is UE location and the AP it is associated to, while the dataset “test” lacks the UE localization.

Each participant can get access to each of the 12000 samples measurement through the provided executable file by passing as flags the name of the dataset and the indices [0-11999] of the samples to generate.
Example: “gen_channel.exe train 0 1 2” generates the first 3 samples of the “train” dataset.

We also provide Python code to more easily understand the dataset through visualization.

The provided Python code generates the measurements regarding the sample indexes 0, 4 and 8, which happen to have 0, 1 and 2 interferers and then prints the information regarding these samples interferer and plots the power spectrums in time, direction of departure and direction of arrival.

These files can either be downloaded from the dataset tab once the team is registered or from the NCSU webpage of the problem.

10AR

ML5G-PHY-Reinforcement learning: scheduling and resource allocation Link

UFPA37The dataset is composed of “traces” from simulations executed with Unreal Engine + AirSim, and will be available on June 30 at this folder
11WLAN

WALDO (Wireless Artificial intelligence Location DetectiOn): sensing using mmWave communications and ML. Link

National Institute of Standards and Technology31

For software and dataset resources, please refer to https://github.com/usnistgov/PS-002-WALDO.

The challenge dataset relies on two open-source software:

  1. IEEE 802.11ay packet generator available at: https://github.com/usnistgov/PS-002-WALDO.
  2. NIST Q-D channel realization software: https://github.com/wigig-tools/qd-realization used to generate the millimeter wireless channel.

Training Dataset
The training datasets consist of a collection of received packets at different SNR. The IEEE 802.11ay preamble is filtered through the NIST Q-D channel realizations, which model the millimeter wave wireless signal propagation in a room with multiple people moving.

12RAN

Lightning-Fast Modulation Classification with Hardware-Efficient Neural Networks Link

Xilinx68

This challenge uses the RadioML 2018.01A dataset provided by DeepSig. You will be required to provide your contact information to download the dataset.
Please note that this dataset is released by DeepSig under the Creative Commons Attribution - NonCommercial - ShareAlike 4.0 License and you must abide by the terms of this license to participate in our challenge.\

Train/test split

Please note that we use a fixed train/test split on the dataset for a fair comparison, and you must abide by this in order to provide a valid submission. This split is given by the fixed random seed and other parameters in the Jupyter notebook we provide in the sandbox.

13

Network anomaly detection based on logs Link

China Unicom45The data is the log data of network equipment, including time stamp, log information, etc. The log is in text format. The data is divided into training set and test set. The training set contains tag information, that is, with exception and without exception; it is a Boolean tag representing a true or false result. The test set comes from the same network equipment as the training set. The test set does not contain tag information. After modelling, the contestants predict the test set and get its tag. The data could not be found.
14

Graph Neural Networking Challenge 2021 - Creating a Scalable Network Digital Twin Link

Barcelona Neural Networking Center47

We provide a dataset generated with the OMNet++ network simulator, which is a discrete event packet-level network simulator. The dataset contains samples simulated in several topologies and includes hundreds of routing configurations and traffic matrices. Each sample is labelled with network performance metrics obtained by the simulator: per-flow performance statistics (mean per-packet delay, jitter and loss), and port statistics (e.g. queue utilization, size).


BNN_UPC_fig2.png

Data is divided in three different sets for training, validation and test. As the challenge is focused on scalability, the validation dataset contains samples of networks considerably larger (51-300 nodes) than those of the training dataset (25-50 nodes). Likewise, the test dataset will be released at the end of the challenge, just before the evaluation phase starts, and it will contain samples following the same distribution as in the validation dataset.

Please, find a detailed description of the datasets and the links to download them here:
https://bnn.upc.edu/challenge/gnnet2021/dataset

Thoth as a Competitor

  1. NFV Related Problems
  2. Start with 2021 Problems. 
    1. KDDI
    2. China Unicom*

Thoth/Anuket/LFN as a Host - ?

LFN.

Tentative Problems :

  1. Loss Characterization
  2. Synthetic Observability (logs and metrics) Data Generation GANs.