People of ACM - Carla Chiasserini

July 22, 2025

How is machine learning improving computer networking?

The way computer networks benefit from machine learning is manyfold.

Firstly, the configuration of networks, services, and applications often involves setting a large number of parameters as well as dealing with time-varying quantities and behaviors of the system components. Additionally, the effects of all these factors may be intertwined. Machine learning can help address the high complexity of advanced network systems, especially in dynamic environments where mobile users operate.

Secondly, it is beneficial to forecast the behavior of the network system such as user mobility or traffic generation patterns in order to proactively adapt the system settings to changed operational conditions or context.

Finally, new areas of research also look at large language models as a way to interpret human requirements and translate them into commands or instructions for the configuration of network systems. These new areas of research also support semantic communication approaches aimed at reducing the amount of information transferred through a communication network.

One of your most cited papers is “Energy Efficient Battery Management” (co-written with Ramesh R. Rao). The paper introduces algorithmic approaches, which influence electrochemical processes to make batteries more energy efficient. Will you discuss why this was a novel approach at the time? This paper was published in 2001, what is an example of a current challenge with regard to battery efficiency?

Thanks to Rao’s inspiring input, we started looking into the chemical processes that affect the battery lifetime, and we found that this could be prolonged by optimizing the energy demand, the usage, of a mobile device. This was the result of a truly interdisciplinary work and it still represents an example of how interdisciplinary research can lead to breakthrough results.

How to extend the lifetime of a battery and to increase battery efficiency are still open challenges which have become even more relevant as we have realized the importance of reducing our energy footprint.

An interesting medical application of your O-RAN project is to enable remote surgery, which requires “highly reliable and low-latency support from the network infrastructure.” Will you discuss the larger goals of the O-RAN project?

The O-RAN project in collaboration with Ahmed Badawy is targeting applications that require high reliability and low latency in data transfer. Besides medical applications, examples include automated manufacturing and connected autonomous vehicles. In all these cases, it is critical that the state of the system and system performance are reliably and timely collected. On the other hand, commands and instructions generated according to the collected state information are enacted within a very short time span. As O-RAN makes use of machine learning-driven intelligence at the radio interface, it can enable a network system to meet these goals. However, in implementing a distributed network design it also suffers from conflicts that may arise between different decision makers at the radio interface. How to make O-RAN effective is in fact a fascinating research area.

In a description of your Connected Autonomous Cars research field, you note that an important challenge is to “remove technological barriers that currently prevent the full exploitation of connected car data.” Will you explain what you mean by this?

Connected cars are equipped with a plethora of sensors such as cameras, radars, and lidars. The multi-modal data collected from these devices is essential for perceiving the vehicle surroundings and then making the car move autonomously, efficiently, and safely. However, the accuracy and reliability of vehicle perception dramatically increases when connected cars share their data, or share abstractions derived from such data. Information sharing may be challenged by unreliable data transfer due to the wireless channel conditions, the communication technology used for the data transfer, or the inaccuracies in the gathered data introduced by real-world sensors. In addition, such data may entail user privacy leakage. To take full advantage of connected cars data is thus essential to tackle these challenges.

It seems like everyone today has a mobile phone and that they are using their phones throughout the day. How are mobile networks keeping up with this demand?

This is indeed one of the main reasons why wireless networks have evolved so much in the last decades, and they keep evolving to increase their capacity. This evolution has concerned all system components and all protocol layers, from large antenna arrays to sophisticated scheduling of radio resources and mechanisms for information compression. However, mobile networks have evolved not only to support the ever-growing traffic demand but also to increase their capability to support new services and applications that, as mentioned, require ultra-high reliability and low latency.