This movie about the @HumanBrainProj is well done but thousands of PhDs (like me) were funded with this ~1B grant and never worked on Markram's Human Brain Simulation.
Here is my perspective on what truly happened inside this gigantic European research program 🧵 1/12
Henry Markam was the first director and public figure of HBP. He pitched the entire HBP around what he had done for a while (see TED 2009) but the detailed HBP proposal was divided in many work packages mostly disconnected from Markram's idea. ted.com/talks/henry_ma…
Although many associate the HBP to Henry Makram, he left the project early on after the controversy around his Human Brain simulation. His most prominent publication as a HBP leader appeared in 2015 and was mainly funded before the HBP started. sciencedirect.com/science/articl…
So what is the HBP truly about?
Human brain imaging,
Mouse brain physiology,
Neuroscience theory,
Infrastructure for neuro-informatics,
Robotics,
Medical research,
Experimental neuromorphic computers,
And more.
Certainly the money was not wasted! Researchers have worked hard and many publications came out.
A harder question is to know whether the outcome is more valuable than with other European grants like ERC which is small sized.
From my HBP funded PhD, roughly 1 every 4 papers was a HBP collaboration. So I would say it did not change research dynamics compared to regular grants.
I think the main difference was that group leaders had to play with interal HBP politics instead of writing grant proposals.
Sadly, HBP did not become like NASA or CERN since we we have not seen a product like the James Webb telescope of the Brain.
HBP knows it, so they launched EBRAINS @HBPHighPerfComp but imo it's a collection of papers/softwares/datasets and not a product. ebrains.eu
What's missing for a big HBP product?
A precise goal: @AllenInstitute or @IntlBrainLab who focused on gathering clean and open sourced neuro data have a much clearer than HBP. I use their data.
Mixing neuro datasets with neuromorphic computing on a single platform is absurd!
The HBP success is the pile of publications, this is what most HBP researchers were fighting for anyways. Trying to build a product-chimera from this cosmos of research result is a political move.
But for successful product it would have been needed to start with the vision.
Sadly, this brings me back to Markam, although most HBP members who signed up for the money did not want to simulate the Human Brain it is partly how people will remember our work inside the HBP.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
There are interesting prospects for engineering applications, but let's not forget that spiking neurons are precise models of biological neurons.
In a paper accepted at #NeurIPS2021 we use back-prop in spiking RNNs to fit cortical data 1/8
Given that the biological network is
(1) strongly recurrent and (2) some neurons are not recorded,
this is a profound statistical problem for which the best existing formalizations are still based on GLMs with the max. likelihood (MLE, @jpillowtime). 2/8
A limitation of MLE training is to be conditioned on recorded data only. So when one simulates the fitted network, it explodes as soon as it's different from the data.
This is why back-prop in spiking RNNs is useful: one can now train the model using simulated spikes! 3/8