Cian O'Donnell
@cianodonnell.bsky.social
Computational neuroscientist.
Senior Lecturer at Ulster University in the Great City of Derry, Northern Ireland.
"not articulate enough"
https://odonnellgroup.github.io
Senior Lecturer at Ulster University in the Great City of Derry, Northern Ireland.
"not articulate enough"
https://odonnellgroup.github.io
multiplicative downscaling encourages a competition between synapses (assuming eg sum of weights is kept constant). Weight decay just encourages each weight to be small without direct competition
February 13, 2026 at 4:19 PM
multiplicative downscaling encourages a competition between synapses (assuming eg sum of weights is kept constant). Weight decay just encourages each weight to be small without direct competition
I always thought multiplicative downscaling was more like a co-ordinated, active process. Unlike 'weight decay' which implies a passive return to baseline
was also kind of asking if you meant the ML L2 regularisation sense
was also kind of asking if you meant the ML L2 regularisation sense
February 12, 2026 at 11:58 PM
I always thought multiplicative downscaling was more like a co-ordinated, active process. Unlike 'weight decay' which implies a passive return to baseline
was also kind of asking if you meant the ML L2 regularisation sense
was also kind of asking if you meant the ML L2 regularisation sense
as far as I understand passive weight decay is not established... for reasons you say
it might be possible that (some aspect of) synaptic weights can last decades
it might be possible that (some aspect of) synaptic weights can last decades
February 12, 2026 at 11:55 PM
as far as I understand passive weight decay is not established... for reasons you say
it might be possible that (some aspect of) synaptic weights can last decades
it might be possible that (some aspect of) synaptic weights can last decades
Can I ask what is this neurobiological weight decay you are talking about?
February 12, 2026 at 10:19 PM
Can I ask what is this neurobiological weight decay you are talking about?
I feel like peripheral sensory and motor systems are the bits you would want least drift in... too tethered to the external physical world
February 12, 2026 at 6:39 PM
I feel like peripheral sensory and motor systems are the bits you would want least drift in... too tethered to the external physical world
Can I ask, why does some good-hearted billionaire not just bankroll a high-quality news outfit with no political oversight attached? Could literally change planet earth for the better
February 12, 2026 at 6:34 PM
Can I ask, why does some good-hearted billionaire not just bankroll a high-quality news outfit with no political oversight attached? Could literally change planet earth for the better
Reposted by Cian O'Donnell
The lesson for all the students out there is that science is a community project. Most of us make individually small contributions to this project. Success is measured at the collective level. Many of our professional (and personal) dysfunctions could be fixed by more fully embracing this view.
February 7, 2026 at 11:20 AM
The lesson for all the students out there is that science is a community project. Most of us make individually small contributions to this project. Success is measured at the collective level. Many of our professional (and personal) dysfunctions could be fixed by more fully embracing this view.
It's not only a community process, but a social process. A contribution includes persuading and convincing your peers of why your findings matter, and a collegiate and respectful tone is a key part of that
February 10, 2026 at 8:33 PM
It's not only a community process, but a social process. A contribution includes persuading and convincing your peers of why your findings matter, and a collegiate and respectful tone is a key part of that
a deadline is a deadline!
February 10, 2026 at 11:19 AM
a deadline is a deadline!
any idea why they hate Unis?
February 10, 2026 at 10:39 AM
any idea why they hate Unis?
Reposted by Cian O'Donnell
'One potential option to streamline the REF process could lie in the adoption of artificial intelligence tools.'
Mai oui! The answer is of course AI. Because we have no evidence to base its use on in this context, and when we do test it, will use datasets for a small number of STEM disciplines. 2/2
Mai oui! The answer is of course AI. Because we have no evidence to base its use on in this context, and when we do test it, will use datasets for a small number of STEM disciplines. 2/2
February 10, 2026 at 10:00 AM
'One potential option to streamline the REF process could lie in the adoption of artificial intelligence tools.'
Mai oui! The answer is of course AI. Because we have no evidence to base its use on in this context, and when we do test it, will use datasets for a small number of STEM disciplines. 2/2
Mai oui! The answer is of course AI. Because we have no evidence to base its use on in this context, and when we do test it, will use datasets for a small number of STEM disciplines. 2/2
sounds cool but trying to wrap my head around this... is it another way of saying that the timescale of synaptic plasticity induction is slower than the ~100 ms timescale of neural population dynamics?
February 8, 2026 at 8:20 PM
sounds cool but trying to wrap my head around this... is it another way of saying that the timescale of synaptic plasticity induction is slower than the ~100 ms timescale of neural population dynamics?