Note that Rudy Pei didn’t unambiguously reply to Philip Dodge with “Yes, this IP belongs to BrainChip” either, so things might actually not be as straightforward as simply inferring the IP must belong to BrainChip because Rudy Pei and Olivier Coenen were both BrainChip employees at the time of the PLEIADES paper’s first publication. It might just turn out to be a little more complicated than that.
One thing seems pretty obvious to me: Ever since Rudy Pei left for NVIDIA, he has been trying to distance himself from BrainChip. He still expresses his appreciation for the work of some of his former colleagues, whose LinkedIn posts he continues to like, but he evidently doesn’t want his own work done over the past few years automatically be associated with his former employer, with which he seems to have had some sort of fall-out.
Have a look at the following picture from the ICLR Conference in Singapore (24 - 28 April 2025), which he posted on LinkedIn at the time. While the conference paper published on 9 April 2025 (
https://arxiv.org/pdf/2501.13230) still identifies him as someone who worked for BrainChip during this research…
“Published as a conference paper at ICLR 2025
LET SSMS BE CONVNETS: STATE-SPACE MODELING WITH OPTIMAL TENSOR CONTRACTIONS
Yan Ru Pei
Brainchip Inc.
Laguna Hills, CA 92653, USA
yanrpei@gmail.com”
… the photo of him posing in front of the conference poster in Singapore two to three weeks later is IMO testament to Rudy Pei’s emotions “post-divorce” so to say: He made an effort to put stickers over what I assume would have revealed where he had been working at the time of his research:
Amazing time at #ICLR2025 today. Absorbed an incredible amount of AI knowledge in a span of 8 hours (standing and walking). Met a lot of new friends, particularly in the SSM space, doing amazing stuff that I would not have even imagined. This sort of diversity in the field of AI is what draws me...
www.linkedin.com
View attachment 92544
View attachment 92545
Neither did he mention his previous employer during this video presentation of
“Centaurus: Let SSMs be ConvNets (ICLR 2025 spotlight)”:
I have never believed Rudy Pei was poached by NVIDIA. Instead, I think he was no longer happy at BrainChip (the why is of course speculation) and actively reached out to look for a new job - also cf the info “LinkedIn helped me to get this job”, and found it with the company he had always dreamed working for.
And it is highly likely no coincidence that he started his new job with NVIDIA just after the Chinese New Year holidays (which this calendar year fell on 29 and 30 January) - after all, the Lunar New Year is considered an auspicious time for new beginnings.
I'm passionate about machine learning research and engineering, with a particular focus… · Experience: NVIDIA · Education: University of California San Diego · Location: San Diego · 500+ connections on LinkedIn. View Rudy Pei’s profile on LinkedIn, a professional community of 1 billion members.
www.linkedin.com
View attachment 92549
The mysterious disappearance of various “TENNs” repositories from BrainChip’s GitHub account sometime in February or during the first days of March also suggests to me that this IP issue may be more complicated than it seems at first glance:
On 2 February, I had posted about my discovery of TENNs-Eyes on BrainChip’s GitHub:
https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-449579
A month later, I noticed all the TENNs model repositories previously found there had disappeared:
https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-452508
“Now here comes the weird thing: When I just revisited BrainChip’s GitHub page (https://github.com/Brainchip-Inc), I noticed that not only has TENNs-Eye disappeared, but so have aTENNuate, TENNs-PLEIADES and Centaurus. Of the 14 repositories (see my 2 February screenshot), four have simply vanished into thin air, it seems… 
[…] Is there possibly any connection with Rudy Pei’s departure, given he was instrumental in leading the R&D for the TENNs model family? (Although I’d expect any rights would actually have been assigned to BrainChip, his former employer?)
A read-only archived TENNs-Eye repository can still be found on his GitHub page. It says the repository has been moved to Brainchip-Inc/TENNs-Eye (presumably in September 2024, when the repository got archived), but that is now an empty link (Error 404).”
It’s actually quite telling to compare the two different versions of the PLEIADES paper - the one dated 31 May 2024 (when both authors were still BrainChip employees, but even then Rudy Pei preferred to use his private email address rather than his work email address) and the recently updated one, dated 24 October 2025 (after Olivier Coenen had been sacked for reasons unknown to us):
In the earlier version of the paper, the title says “TENNs-PLEIADES” and the abstract’s first sentence also refers to the TENNs architecture. As does another sentence in the Introduction, which further mentions the broader class of networks called TENNs were developed by BrainChip Inc.
This reference to TENNs & BrainChip was completely dropped in the updated October 2025 version. The only remaining reference to the researchers’ former employer can be found in a small footnote saying “Work done while at BrainChip Inc”.
View attachment 92540
View attachment 92541
View attachment 92538
(…)
View attachment 92539
The other striking omission in the October 2025 version is of Course who aT BrainChip nO longer gets named under “Acknowledgement(s)” - see for yourself…
brainchip.com
www.autotechagil.de
