Ariadna final presentation

Today I presented the final results of the Ariadna project at ESA-ESTEC, with the guys of the ESA/ACT and all the ESA people interested in this amazing cutting edge technology.

Overall, the project was a success! 🎉

Spiking Neural Networks are a viable alternative to traditional Artificial Neural Networks, and they can be used in space applications to perform complex tasks with a fraction of the energy consumption. SNN systems can be easily prototyped by means of traditional Deep Learning libraries, like PyTorch and TensorFlow, and the use of the metrics developed during the project allowed us to compare the computational demands of different models (both SNN and ANN) in a hardware-agnostic way, to possibly perform a trade-off and select the most promising architectures for actual implementation on neuromorphic devices.

The obtained results open up a new world of possibilities for future research. In fact, an effort is still needed, to look for architectures, regularization techniques, and initialization methods capable to exploit the peculiarities of latency-based SNN, and bring such technology to practical implementation in flight. 🚀

Detailed results will be published in the next months, so stay tuned! Meanwhile, in the main project page you can find a summary in the final report, and you can download the slides of the presentation.