I have extracted the following from the second linked article out of KP Labs News items. I have broken it up to make it easier to see the significance of the three different roles the consortium members are playing. As we know OHB Hellas has expressed some very favourable views regarding AKIDA technology since around 12 months ago when it first announced receiving an AKIDA neuromorphic processor:
The BOLERO project is propelled by the synergistic efforts of its consortium members. 1. As the project leader, KP Labs is primarily responsible for developing the Synthetic Data Generators (SDGs) and the continual learning models, ensuring their efficacy across multiple SatCom applications and hardware architectures.
2. OHB Hellas contributes by exploring novel machine learning methodologies suitable for streaming data, assessing continual learning applications in and beyond the space sector, and implementing two use cases in different hardware modalities.
3. Eutelsat OneWeb focuses on identifying strategic space-based applications for continual learning, evaluating their business impact, and analyzing the benefits of continual learning models, particularly in terms of performance and cost-efficiency.
Together, these entities combine their unique strengths to advance the BOLERO project, addressing the evolving demands of SatCom systems."
We are thrilled to announce that OHB Hellas has taken a leap forward in cutting-edge technology with the acquisition ofBrainChipAkida Neuromorphic Processor! Our engineering team has dived into the realm of neuromorphic computing, exploring its limitless potential to revolutionize Artificial Intelligence (AI). Neuromorphic processing holds immense potential for AI applications due to its ability to mimic the brain's architecture, offering high efficiency and adaptability in processing data. This is particularly important in environments with limited resources and high computational demands, as is the case for various space applications. Alexis ChatzistylianosandAntonis Karterishave conducted some initial experiments inferencing an image classification Spiking Neural Network (SNN), to evaluate the performance of the neuromorphic processor and benchmark it against traditional CPU and GPU architectures. The neural network was based on the AkidaNet architecture and the quantized model, pretrained on the ImageNet dataset, was obtained from the Akida model zoo and converted to an Akida compatible SNN. It consists of 1.4M parameters, 32 neural processors (NPs) and a quantization scheme of 8-bit weights in the input layer, 4-bit weights in other layers and 4-bit activations. As illustrated from the charts shared below, the Akida NPU brings astounding performance gains in power efficiency and inference time, while maintaining high accuracy standards. At OHB Hellas, we are committed to staying at the forefront of technological innovation, driving progress, and shaping the future of on-board high performance computing and AI-driven data processing. Stay tuned for more updates as we continue to chart new territories in the ever-evolving landscape of space technology!
A personalised tool to help users track selected stocks. Delivering real-time notifications on price updates, announcements, and performance stats on each to help make informed investment decisions.
(20min delay)
Neuromorphic processing holds immense potential for AI applications due to its ability to mimic the brain's architecture, offering high efficiency and adaptability in processing data. This is particularly important in environments with limited resources and high computational demands, as is the case for various space applications.
Alexis ChatzistylianosandAntonis Karterishave conducted some initial experiments inferencing an image classification Spiking Neural Network (SNN), to evaluate the performance of the neuromorphic processor and benchmark it against traditional CPU and GPU architectures. The neural network was based on the AkidaNet architecture and the quantized model, pretrained on the ImageNet dataset, was obtained from the Akida model zoo and converted to an Akida compatible SNN. It consists of 1.4M parameters, 32 neural processors (NPs) and a quantization scheme of 8-bit weights in the input layer, 4-bit weights in other layers and 4-bit activations. As illustrated from the charts shared below, the Akida NPU brings astounding performance gains in power efficiency and inference time, while maintaining high accuracy standards.
At OHB Hellas, we are committed to staying at the forefront of technological innovation, driving progress, and shaping the future of on-board high performance computing and AI-driven data processing.
Stay tuned for more updates as we continue to chart new territories in the ever-evolving landscape of space technology!