Discovery

Thematic series of technical talks on AI/Machine learning.

The “Discovery-Channel” digs deeper into thematic areas transformed by Artificial Intelligence/Machine Learning as well as into challenges of current AI/ML technology. Each “Discovery” episode dedicates one hour to a researcher to present their latest findings in one of the topic areas below.

ML5G

Trustworthy AI

AI and Health

AI and Climate Science

GeoAI

ML5G

Many stakeholders in the information and communication technology (ICT) domain are exploring how to make best use of AI/ML. But applying AI/ML in communication networks poses different challenges than applying machine learning in, say, image recognition or natural language processing. The reasons are:

  1. Time scales vary a lot in communication networks, from annual (e.g. your subscription to a telecom provider) to millisecond timescales. If your network parameters changes on a millisecond timescale, you need to (re)train your ML model on a similar timescale.
  2. The network environment is noisy.
  3. Computing resources in a network are limited.

5G, combined with AI, will speed up the advancement of the UN Sustainable Development Goals with significant contributions in healthcare, education, agriculture, energy, manufacturing and transportation, among others. The technology will transform these sectors by providing significantly higher speed and lower latency for people, devices and applications.

ITU has been at the forefront to explore how to best apply AI/ML in future networks including 5G networks. Please see the “AI (Pre-)Standardization Section”

Curator

Independent Research Consultant

Trustworthy AI

Artificial Intelligence (AI) systems have steadily grown in complexity, gaining predictivity often at the expense of interpretability, robustness and trustworthiness. Deep neural networks are a prime example of this development. While reaching “superhuman” performances in various complex tasks, these models are susceptible to errors when confronted with tiny (adversarial) variations of the input – variations which are either not noticeable or can be handled reliably by humans. This expert talk series will discuss these challenges of current AI technology and will present new research aiming at overcoming these limitations and developing AI systems which can be certified to be trustworthy and robust.

Curator

Head of Department of Artificial Intelligence
Fraunhofer Heinrich Hertz Institute

AI and Health

The health sector, one of the most important sectors for societies and economies worldwide, is particularly interesting for AI applications, given the ongoing digitalization of health data and the promise for an improved quality of health and healthcare.

Many investigators from the machine learning community are driven to applying their methodological tool kits to improve patient care, inspired by the impressive successes in image analysis (e.g. in radiology, pathology and dermatology).

However, due to the complexity of AI models, it is difficult to distinguish good from bad AI-based solutions and to understand their strengths and weaknesses. ITU and the World Health Organization established the ITU/WHO Focus Group on “AI for Health” to clarify responsibilities and building trust among AI developers, AI regulators and AI users.

Curator

Chair of the Department of Biomedical Informatics
Harvard Medical School

AI and Climate Science

For many years climate scientists have used comparatively simple statistical approaches to try and discern subtle changes in observational datasets, or to interpret abundant climate model data output. The opportunity now presents itself for climate science to exploit advances in Machine Learning to answer some of the most pressing challenges of our time – while they are still relevant for policy makers. This acceleration will be built upon: supervised and un-supervised learning of features and patterns in the vast amounts of Earth observation and climate model data that is now available, transforming our ways to constrain climate models and the detection of climate change; robust emulation of existing climate models and their components; and causal detection and attribution of regional climatic changes.

Climate scientists have begun to enthusiastically explore these possibilities, but scaling these novel approaches to the exabytes of data that will be created over the next decade to answer urgent scientific and policy relevant questions in a timely manner will require a concerted collaboration between academia, industry and policy makers. Industrial partners in particular can play a leading role in bringing in their extensive expertise and know-how but also in ensuring that climate data is accessible and interoperable with the latest ML algorithms and the specialized computational hardware required to run them.

This series provides a forum for leading voices in these fields and across sectors to outline a vision for how we will achieve this – with the aim of Accelerating Climate Science with AI.

Curator

Head of Atmospheric, Oceanic and Planetary Physics
University of Oxford

GeoAI

Everything happens somewhere – applying machine learning to geospatial analysis

Geospatial AI (or GeoAI for short) is the discipline that uses AI to analyze data sets which include a spatial (location) component, i.e., a component that can be located by a coordinate system. Most data sets have location coordinates.

AI for Good Discovery partners

Scroll Up

Your Gateway to #AIforGood in your inbox!

Join our 60K+ global community to receive insights & invites to AI for Good! Just one email each week, and zero spam.