New methods to improve AI quality

Researchers at IBM Research Zurich in Rüschlikon in the canton of Zurich have made a discovery that could help to improve the quality of Artificial Intelligence (AI) together with an international research group. They have discovered that photonic computer processors are able to process information much more rapidly than conventional electronic chips. Moreover, they are able to process several information streams in parallel. Data is transported via light in photonic computer processors, while conventional computer chips rely on electronic data transmission.

As part of their experiments, the researchers presented a processor in which they had combined photonic structures with phase-change materials (PCMs) as energy-efficient storage elements, as outlined in a press release. In terms of a light source, the researchers used a chip-based frequency comb – a technology developed by the Swiss Federal Institute of Technology Lausanne (EPFL). The researchers used their processor in an experiment with neural networks designed to recognize hand-written numbers. The overall result was that more rapid and parallel data processing was possible.

Thanks to the discovery made by the researchers, a greater quantity of data could now be processed simultaneously in the area of Artificial Intelligence. The use of larger neural networks also allows more accurate forecasts, including some that would not have been possible previously, as well as more precise data analyses. This could, for example, aid medical diagnoses and benefit progress in the field of autonomous driving.

In addition to IBM Research Zurich, scientists from academic institutions such as WWU-Münster in Germany, Oxford University and the University of Exeter in the UK, as well as EPFL, were also involved in the research project.