The 5th edition of the ITU AI/ML in 5G Challenge: A year of competitions in review

How can Artificial Intelligence and Machine Learning (AI/ML) be used to solve problems in the creation, training, and deployment of communication networks?

The ITU AI/ML in 5G Challenge is a competition platform designed to identify solutions in various real-world use cases with AI/ML. Participating teams must collaboratively identify innovative solutions. Anyone is free to partake in these global challenges that advance the achievement of the Sustainable Development Goals (SDGs) using AI/ML.

Organized by AI for Good with several partners, the challenges set up realistic scenarios with a mix of real-world and simulated data. Teams are encouraged to showcase their talents and develop innovative concepts to address the curated problem statements. ITU provides a free compute platform to ensure inclusivity in the competitions, ensuring access to free GPUs and CPUs, pre-installed ML packages, and more. Please see the entire list of ITU AI/ML competitions in 2024 at the end of the article.

In 2024, there have been a total of 13 challenges and a total of 4196 participants joined the various problem statements in 2024, contributing more than 30’000 submissions. A total of more than 50’000 CHF was disbursed as prizes for the various problem statements.

ITU AI/ML in 5G Challenges in 2024

The topics of Green Energy and Traffic Forecasting had a joint challenge launch in August 2024. The problem statement on Smart Energy Supply Scheduling for Green Telecom addressed the need to optimize energy consumption of mobile networks. With more than 700 enrolments, participants came from 86 countries, 33 of which were African countries.

Read more about the challenge conclusion at COP29 in Baku in the corresponding here.

Another challenge introduced in May 2024 was on Large Language Models (LLMs) for Telecom Networks, hosted on the Zindi platform. As a sector, the telecommunications industry has not enjoyed the widespread adoption of LLMs as much as other industries. This competition invited researchers to present use cases of telecom GPTs. From over 400 participants, team 3musketeers emerged victorious, followed by team AsiaLab and SciPhi.

The Challenge on Spatio-Temporal Beam-Level Traffic Forecasting Challenge was hosted on Zindi, and had 660 participants, with most coming from India and Nigeria. The goal was to develop a multivariate time series forecasting model for traffic volume (DLThpVol) at the beam level which would be trained on high-resolution data to capture the intricacies of real-world traffic scenarios. These models could potentially revolutionize traffic management practices. The winners of this challenge are teams Koleshjr, cobusburger, and sandrohu.

The challenge on Estimation of Site specific radio propagation loss, hosted on the ITU platform, asked to construct an AI/ML model to accurately estimate propagation loss. The problem statement was proposed and hosted by KDDI Research, Inc.by and managed in collaboration with ?RISING Japan. The winner was Team Mlab at University of Tokyo, followed by teams from China and Singapore.

Watch the Challenge finale here.

The ITU WTSA24 Hackathon – AI Bharat 5G/6G Sandbox was hosted by ITU in collaboration with the Department of Telecom (DoT) in India as well as the Telecom Centres of Excellence (TCoE, India). The hackathon consisted of an online and an in-person phase at the World Telecommunication Standardization Assembly 2024 (WTSA-24). Both problem statements were anchored in ITU’s recommendations ITU-T Y.3172 and Y.3061, which provide frameworks for deploying AI in future network infrastructure, showcasing their real-world implications.

Read more for further details on the hackathon.

Optimal Multi-user scheduling in massive MIMO mobile channels tasked participants to use ML-based algorithms to solve the multi-user beamforming scheduling problem. This challenge was won by team UofTW, consisting of Sara Al-Kokhon, Elvino Sousa, Hossein Bijanrostami, Elaheh Bassak, and Brad Stimpson.

‘AIntuition’: Retrieval Augmented Generation (RAG) for Public Services and Administration Tasks, organized on the Zindi platform, was launched in April 2024. RAG optimizes the capabilities of LLMs by enabling efficient querying of vast volumes of administrative documents and delivering contextually relevant responses. As a nascent field in public administration, the technology faces several hurdles before deployment, including document complexity and data privacy concerns. This challenge fosters innovation in open-source RAG tools that aim to revolutionize the delivery of public services. The top award was given to Victor Olufemi from Nigeria, followed by AdeptSchneider 22 in Kenya and Berexia Digital from France.

GeoAI Challenges in 2024

Five GeoAI challenges were hosted in 2024, marking its third edition, and focused on use cases in geospatial analysis, using location-based data to solve real-world problems.

Two of them were supported by the Instituto Nacional de Estadística y Geografía (INEGI), an autonomous public organization that collects, analyzes, and disseminates statistical and geographical data about Mexico.

The first one was the Human Settlement Detection Challenge which focused on the challenge of rapidly expanding human settlements. Participants developed a robust and accurate ML model that can detect human settlements in satellite imagery to track and manage this growth. The challenge was won by team Click Click Boom from India.

This was followed by the INEGI UN-GGIM Vegetation Mapping Challenge which tasked participants to map the spatial distribution and condition of vegetation types. Natural vegetation is important for biodiversity, carbon capture and storage, soil formation, and more. The objective of this challenge was to propose and develop an ML model that can help clean and improve training data for INEGI and other national mapping agencies. Team bit_guber from India won this challenge, followed by karam-elhaj from Sudan.

Three more GeoAI Challenges focused on various topics: The GeoAI Amazon Basin Secret Runway Detection Challenge looked at the automated detection of clandestine runways which often serve for illicit activities such as narcotics trafficking and awarded first prize to Dami Akinniyi from Nigeria. Another highlighted global cropland mapping for identifying the use of plastics in agriculture; the GeoAI Agricultural Plastic Cover Mapping with Satellite Imagery’s top prize went to Stephen Kolesh in Kenya . The third GeoAI challenge on Ground-level NO2 Estimation asked for ML models to estimate surface nitrogen dioxide (NO2) using only public remote sensing data as predictor variables, and was won by Engin Sancak from Türkiye.

Read more.

Want to join the next AI for Good Global Summit in July 2025?

Free registration is now open, so don’t miss your chance to attend the leading UN event on AI and hear from the top experts, high-level global lineup, demos and exhibitors!

40% discount on the Leaders Pass. Act Fast—Secure Your Spot at the AI for Good Global Summit! With our Gold or Leaders Pass, you'll not only secure your place but also unlock exclusive Summit programming and networking experiences.

Book your pass now!

Our Sponsors

Enjoyed this newsletter?

Subscribe here to our newsletter AI for Good Insider to receive the latest AI insights.

This edition of AI for Good Insider was written and curated by AI for Good, AI & Robotics Programme Assistant Cindy Zheng.