#Neuromorphic #computing just got more accessible! Our work on a Neuromorphic Intermediate Representation (NIR) is out in @SpringerNature Communications. We demonstrate interoperability with 11 platforms. And more to come!
https://www.nature.com/articles/s41467-024-52259-9
NIR is a data format that is understood by 7 software libraries and 4 hardware platforms. Before NIR, models from one framework could not transfer to another. Now, we can develop software and hardware independently.
A 🧵 1/3
Neuromorphic intermediate representation: A unified instruction set for interoperable brain-inspired computing - Nature Communications
Neuromorphic software and hardware solutions vary widely, challenging interoperability and reproducibility. Here, authors establish a representation for neuromorphic computations in continuous time and demonstrate support across 11 platforms.
Controversial opinion: not every website needs a selectable dark mode. I appreciate your efforts, prepaid phone card company, but it’s truly not necessary.
Another Elsevier paper with obvious AI-written text.
“In summary, the management of bilateral iatrogenic I'm very sorry, but I don't have access to real-time information or patient-specific data, as I am an AI language model.”
https://www.sciencedirect.com/science/article/pii/S1930043324001298
Photo by Marcello Gennari on Unsplash
esperienze expo milano – Download this photo by Marcello Gennari on Unsplash
Our approach for building and deploying SNN applications to mixed-signal processors is highly extensible, and can be adapted to additional processor families! Of course all our code is available open-source, and we encourage developers to experiment:
https://github.com/synsense/rockpool
GitHub - synsense/rockpool: A machine learning library for spiking neural networks. Supports training with both torch and jax pipelines, and deployment to neuromorphic hardware.
A machine learning library for spiking neural networks. Supports training with both torch and jax pipelines, and deployment to neuromorphic hardware. - synsense/rockpool
We made use of JAX-based JIT compilation at training time, which enabled us to train models on GPU, TPU or CPU. On several platforms this accelerated training by more than 3000 times, reducing training time from days to minutes.
Using differentiable computing, coupled with a hardware-aware neuron model and a model of parameter variation during training, we build applications which are robust against mismatch. For deployment we have an optimisation-based solution for parameter quantisation and mapping.
Uğurcan Cakal has built the most advanced and easiest to use toolchain for building and deploying SNN applications to mixed-signal NM processors.
https://doi.org/10.1088/2634-4386/ad2ec3Coupled with uncontrolled variation in parameters and neuron dynamics caused by fabrication non-idealities, deploying robust applications to these processors is challenging.
Building and deploying applications to mixed-signal neuromorphic processors has historically been a difficult task. These processors often have a high degree of complexity with a low degree of parameterisation, meaning that parameters are highly constrained across a network.
Jens Egholm
Carl T. Bergstrom
