From July 10–12, 2025, the Czech Institute of Informatics, Robotics and Cybernetics at CTU in Prague hosted the inaugural edition of the international conference Prague Synapse: Workshop on Neural Networks and Reasoning. The three-day event brought together researchers from top global universities and industrial labs to discuss the future of neural networks and their ability to solve problems through processes that resemble human thinking.
The conference united two distinct worlds—symbolic logical reasoning and neural network-based approaches. This exchange between different schools of thought opened up new perspectives on the challenge of building general artificial intelligence.
“The most fascinating part was watching how traditional logic-based reasoning methods clashed and connected with those aiming to achieve the same goals using neural networks. This interdisciplinarity was one of the workshop’s biggest strengths,” said conference organizer Jan Hůla from CIIRC CTU.
Ideas that challenge the status quo
Among the provocative topics discussed were critical perspectives on current training methods for neural networks. “Today’s models go through costly training processes and then stop learning altogether. That’s a sharp contrast to biological neural networks. However, some talks explored ways to enable continuous learning in neural networks even after deployment,” Hůla added.
Speakers included representatives from Harvard University, Stanford, Oxford, Cambridge, ETH Zurich, EPFL, as well as researchers from leading companies like Google DeepMind, OpenAI, and Sakana AI. Notable participants included Róbert Csordás (Stanford), Federico Barbero (Oxford), Mirek Olšák (Cambridge/DeepMind), and Sebastian Risi (Sakana AI).
The conference hosted over 50 experts—according to organizers, an ideal number for in-depth discussion and meaningful networking. Given the overwhelmingly positive feedback, a follow-up edition is already in the works.
