Working with neuromorphic and analog computing. Curious about abstractions. Cares about #FOSS
Author of Neuromorphic Intermediate Representation in Nature Communications: https://www.nature.com/articles/s41467-024-52259-9
| [email protected] | |
| Website | jepedersen.dk |
| GitHub | github.com/jegp |
| https://www.linkedin.com/in/jens-egholm-pedersen-69543117/ |
How privileged is it that I was paid to study for 5 years? Thank you to my family, colleagues, supervisors, committee, opponent, and many many others who made this possible. I've been rewarded with personal and professional insights, skills, friendships, awards, and memories that will last me a lifetime.
❤️
https://kth.diva-portal.org/smash/record.jsf?pid=diva2%3A2001119&dswid=-1044
We built spiking neural networks that outperform ANNs on event-based vision by initialising them with spatio-temporal receptive fields. Our theory-guided design beats naive approaches by 42%!
New paper in Nature Communications 🧠⚡
Paper (open access): https://www.nature.com/articles/s41467-025-63493-0
Video summary: https://youtu.be/qtHkrx4tYfI
Neuromorphic computing mimics brain efficiency but lacks theoretical guidance. Here, authors develop a computational foundation for processing signals in space and time in spiking neural networks that can outperform standard neural networks in event-based vision.
Neuromorphic computing mimics brain efficiency but lacks theoretical guidance. Here, authors develop a computational foundation for processing signals in space and time in spiking neural networks that can outperform standard neural networks in event-based vision.
Why should you read our paper on covariant #neuromorphic networks, you ask? Good question!
We're connecting decades of work in computer vision with decades of work in spiking networks. It's cool because it explains signal processing in spiking networks really well.
The theory is also cool because it works in practice! Trained on an event-based vision task against regular ANNs of similar complexity, spiking networks wipe the floor with ANNs. Interesting, right?
Neuromorphic computing mimics brain efficiency but lacks theoretical guidance. Here, authors develop a computational foundation for processing signals in space and time in spiking neural networks that can outperform standard neural networks in event-based vision.
It's the first post out of three that describes our strategy, and we'll be launching a host of initiatives (check out the shiny new website!) to help accelerate open and accessible science in the neuromorphic domain!
Give it a read, join the discussion on Discord, and stay tuned for more! 🚀
2/2
Throughout my PhD I put in a lot of effort in building open-source software, because I fundamentally believe that software is *necessary* for good science.
As the newly elected chair of Open Neuromorphic, I've put elaborated the argument in our newly launched strategic vision for ONM: https://open-neuromorphic.org/blog/strategic-vision-open-neuromorphic/
1/2
This is all expertly put together by my colleague Juan Pablo Romero Bermudez, who demonstrated around 7ms reaction time with both SpiNNaker 1 and Norse. All using #opensource neuromorphics Open Neuromorphic 🥰
We're slowly making neuromorphic tech more and more accessible 💪
2/2
A new paper is out where we play air hockey with *millisecond* latency using #neuromorphic hardware. Milliseconds. That's fast! 🏃➡️
This is kind of the culmination of my work at KTH Royal Institute of Technology where I've built a spiking neuron simulator (Norse), a fast event-camera processor (AEStream), and cool spatio-temporal spiking receptive fields.
Read more here: https://iopscience.iop.org/article/10.1088/2634-4386/addc15
1/2
An Invitation to Neuroalgebraic Geometry
In this expository work, we promote the study of function spaces parameterized by machine learning models through the lens of algebraic geometry. To this end, we focus on algebraic models, such as neural networks with polynomial activations, whose associated function spaces are semi-algebraic varieties. We outline a dictionary between algebro-geometric invariants of these varieties, such as dimension, degree, and singularities, and fundamental aspects of machine learning, such as sample complexity, expressivity, training dynamics, and implicit bias. Along the way, we review the literature and discuss ideas beyond the algebraic domain. This work lays the foundations of a research direction bridging algebraic geometry and deep learning, that we refer to as neuroalgebraic geometry.
Bruno Gavranović