Frangipani
Top 20
Found a brand new University of Sydney PhD thesis, which is highly likely related to the GitHub “Akida Seizure” repository that one of the two thesis supervisors, Omid Kavehei, had set up under the GitHub name NeuroSyd just over two months ago (which seems to have disappeared since or has possibly been renamed?), spotted by @Fullmoonfever at the time.
Student Profile
www.sydney.edu.au
View attachment 88203
University of Sydney PhD candidate Andre Zainal, who has a B.Eng. (Biomedical) & B.MedSci background, embarked on his Research PhD in Biomedical Engineering just a week ago, “researching AI-driven neuromorphic hardware for early seizure detection”.
The title of his thesis is “Optimizing Training Algorithms for Neuromorphic Hardware: Enhancing In-Memory Computing with FPGA and Akida Neural Processors for Epileptic Seizure Detection”.
Akida doesn’t have “in-memory-computing”, though?
View attachment 88204 View attachment 88205 View attachment 88206
Andre Zainal’s first PhD supervisor Omid Kavehei has been conducting epilepsy research for years:
Staff Profile
www.sydney.edu.au
View attachment 88210
View attachment 88212
View attachment 88213
(…)
View attachment 88214
More on Andre Zainal’s two PhD supervisors:
Principal supervisor Omid Kavehei is not only Professor at University of Sydney’s School of Biomedical Engineering, but also Director and Founder of BrainConnect (https://brainconnect.com.au/), the stealth startup “developing novel solutions in long-term interfacing with the brain and body”, where co-supervisor Duy Nhan Truong works as Engineering Lead.
Prior to filling that position, Duy Nhan Truong also used to be at University of Sydney, initially as a PhD student in Electrical and Electronics Engineering (transferred from RMIT University in 2018, PhD thesis completed in 2020, titled “Epileptic Seizure Detection and Forecasting Ecosystems”, principal supervisor: Omid Kavehei; https://ses.library.usyd.edu.au/bitstream/handle/2123/21932/truong_nd_thesis.pdf?) and later as a postdoc.
View attachment 88215
View attachment 88216
Duy Nhan Truong - BrainConnect Pty Ltd | LinkedIn
I bring over 6 years of experience in electronics development for medical devices… · Experience: BrainConnect Pty Ltd · Education: University of Sydney · Location: Sydney · 481 connections on LinkedIn. View Duy Nhan Truong’s profile on LinkedIn, a professional community of 1 billion members.www.linkedin.com
View attachment 88217
View attachment 88218
View attachment 88219
View attachment 88220 View attachment 88221
New paper published today from the Sydney researchers around Omar Kavehei on epileptic seizure detection and prediction using Akida.
The paper’s first author is Luis Fernando Herbozo Contreras, who according to his LinkedIn profile graduated from University of Sydney with a PhD in Computer Science and Biomedical Engineering in August (but for some reason hasn’t yet updated his self-description, which still reads “PhD candidate”).
The title of his PhD thesis is “Bio-inspired Algorithms for Low-Power Seizure Detection: Towards Neuromorphic Neuromodulation AI”.
So here we’ve got yet another (former) PhD student supervised by Omar Kavehei who has evaluated Akida for epileptic seizure detection. See my above post about Andre Zainal, who started his PhD in July - the title of his thesis even contains Akida by name!
Staff Profile
While the Sydney researchers seemed overall happy with Akida’s performance, they also addressed certain limitations of BrainChip’s first generation of neuromorphic hardware and concluded by saying:
“7 Conclusion
This work demonstrates the effectiveness of a novel framework for seizure detection and prediction directly on the edge, enabling learning with lower computational com- plexity, reduced energy consumption, and in an unsupervised manner while processing streaming data. Looking ahead, if the long-term vision is neuromorphic neuromodulation, this framework establishes a strong foundation toward that goal. Future work will focus on designing architectures capable of better capturing temporal and spatial dependencies, as shown in recent studies [23], thereby achieving higher performance without increasing memory or computational requirements. Overall, these results pave the way for the next generation of AI-driven electroceuticals.”
(…)
(…)
Reference 23 is another publication by the same authors plus PhD candidate Isabelle Aguilar:
Maybe the option of now being able to access Akida 2.0 via the cloud will make them reconsider?
Anyway, yet another validation of Akida (and neuromorphic technology in general) for important future use cases in biomedical engineering.
)
