IMDEA Networks research on a new AI explainability tool for mobile networks wins award at IEEE INFOCOM 2024

AIChronoLens has several applications in different scientific fields. Among other use cases, these models are very useful for predicting future traffic loads of 5G base stations

11 April 2024

The work “AIChronoLens: Advancing Explainability for Time Series AI Forecasting in Mobile Networks” (by Claudio Fiandrino, Eloy Pérez Gómez, Pablo Fernández Pérez, Hossein Mohammadalizadeh, Marco Fiore, and Joerg Widmer) has been awarded the “Best Paper Award” at the prestigious IEEE INFOCOM 2024 conference, taking place in Vancouver, Canada from May 20-23.

This study introduces “AIChronoLens”, an innovative tool that provides explanations for artificial intelligence (AI) models involved in predicting time series data. These AI models are utilized across various scientific fields, from finance to epidemiology. In the realm of mobile networks, these models are particularly valuable for forecasting future traffic loads on 5G base stations.

According to Claudio Fiandrino, Senior Researcher at IMDEA Networks and one of the paper’s authors, “It is an unprecedented work for several reasons. Firstly, it addresses a critical need: the lack of clarity in traditional explainability techniques. Unlike those techniques that provide relevance scores on input data regardless of the data’s own structure, AIChronoLens relates explanations to the presence or absence of maximums and minimums and other temporal properties of the input data. Thus, AIChronoLens creates a relationship between the model’s operation and the physical system.”

Secondly, it opens up various avenues of research. Fiandrino highlights that “the analysis of AIChronoLens output allows for a deeper understanding of the model’s behavior and detection of the hidden cause of errors, such as missing data in the training set or the intrinsic complexity of certain data. It can also compare different models applied to the same dataset beyond simple prediction accuracy metrics, making it a valuable verification tool for current and future AI models.”

Regarding practical applications, the researcher notes: “We are at the dawn of the era of trust in AI, and AIChronoLens is a small step further towards realizing this vision. It can be used by the research community during the development of new AI models to verify their internal logic and compare their performance with known models in a much more sophisticated way than is currently possible. For example, it allows for understanding the confidence of a prediction given the knowledge that the AI model has acquired after training.”

This recognition at the IEEE INFOCOM 2024 conference highlights the importance of AIChronoLens in advancing explainability and confidence in AI models, promising significant benefits across various scientific and technological domains.