More effective algorithms to reduce AI energy consumption

Developing more effective algorithms to reduce the currently extremely high energy consumption of artificial intelligence systems is one of the challenges that statistical physics is preparing to tackle at its world conference, which opens in Florence on July 13, with over 1,500 scientists from around the world. Hosting the conference are Roberto Liv, president of the Italian Society of Statistical Physics, and Stefano Ruffo, head of the conference's steering committee, called Statphys29 and scheduled to run until July 18. Nobel laureate Giorgio Parisi will also open the proceedings. "Statistical physics was born to study the complex behavior of matter , but today it has a much broader impact ," observes Livi. One of the topics this discipline is addressing, and one that represents a real challenge, concerns neural networks . The challenge is to ensure that this field of research communicates with that of neuroscience to obtain a more complex and realistic representation of how networks work. "These are fields that are intersecting after having followed different paths ," Livi told ANSA, and "our scientific community wants to ensure that these different fields communicate with each other and find common goals , in cutting-edge research that combines applied aspects and neural models." The potential applications are many, but one of the main ones is developing artificial intelligence algorithms that are more effective than current ones. "This result could have a huge impact , considering the high energy consumption of artificial intelligence. The great challenge ," Livi noted, "is to reduce consumption , and for this reason, algorithms that are less trivial than the current ones are needed, which are not easy to achieve." It is therefore necessary to move beyond the old model of neural networks , as Parisi has repeatedly emphasized. "Statistical physics is the basis of extraordinary discoveries: from the physics of materials to the study of complexity even in living systems," said the Nobel Prize winner, "and this conference is an opportunity for us all to come together and take stock of the situation." Parisi himself has repeatedly observed that, compared to the early models of the 1980s and 1990s, current research on deep neural networks has the major problem of not being supported by a robust theory.

Non-equilibrium and active matter: the new research objectives

From the behavior of the climate to that of a flowing river , systems that change over time and are therefore defined as 'non-equilibrium ' are among the new objectives of research in statistical physics , as are studies on active matter —that is, systems composed of cells, bacteria, or robots that consume energy to interact with each other. These emerging fields of research are definitely at the forefront of the Statphys29 World Statistical Physics Conference, which opens today in Florence. "For the first time, these topics predominate . For example, papers on non-equilibrium systems occupy almost double the space compared to other topics, thanks in part to the important results obtained in past years," Stefano Ruffo, head of the conference's steering committee, told ANSA. "These are formal and mathematical results, which could lead to numerous applications in materials physics. "There are also many papers on active matter," Ruffo continued, "with interesting applications in biology, for example, relating to the microbiota." Statistical physicists are " atypical physicists ," observes Roberto Livi, president of the Italian Society of Statistical Physics. "We are interested in methodologies related to statistical mechanics, which has a vast number of applications. Behind all this effort, there are bets on new ideas that will need to be verified. No serious scientist can promise results, only new paths to explore." With 1,500 scientists from 50 countries , "the world conference includes Russians and Ukrainians , Iranians and Israelis : "the spirit of science prevails , which," noted Ruffo, "speaks a universal language and sees the unity of all peoples, beyond conflicts." With the same spirit, the community of statistical physicists looks to Africa , where one of the satellite sessions of the world conference was organized at the University of Rwanda . "It's the first time it's happened," said Ruffo, "and it was a success, with the participation of around a hundred students from various African countries. A qualitative effort has been made to enable African researchers to enter this field of study at the highest level.