Cian O'Donnell
LightNews LightNews
banner
cianodonnell.bsky.social
Cian O'Donnell
@cianodonnell.bsky.social
2.8K followers 1.8K following 900 posts
Computational neuroscientist.
Senior Lecturer at Ulster University in the Great City of Derry, Northern Ireland.
"not articulate enough"
https://odonnellgroup.github.io
Posts Replies Media Videos
cianodonnell.bsky.social
predictive coding, RPE, successor representation, and RL more generally. These have been the great success stories of theoretical neuroscience from the past 20 years.

Hard to see how these discoveries could have arisen via experiment and conceptual modelling alone
mgkumar138.bsky.social M Ganesh Kumar @mgkumar138.bsky.social · 4h
All theory is wrong until verified by data. Greatly indebted to @mhyaghoubi.bsky.social, @markbrandonlab.bsky.social, @douglasresearch.bsky.social for finding the hippocampus encoding reward prediction! Grateful to my advisor @cpehlevan.bsky.social, @kempnerinstitute.bsky.social.
#RL #hippocampus
nature.com Nature @nature.com · 4d
Nature research paper: Predictive coding of reward in the hippocampus

go.nature.com/49mB13V
January 19, 2026 at 9:46 AM
Reposted by Cian O'Donnell
mgkumar138.bsky.social
All theory is wrong until verified by data. Greatly indebted to @mhyaghoubi.bsky.social, @markbrandonlab.bsky.social, @douglasresearch.bsky.social for finding the hippocampus encoding reward prediction! Grateful to my advisor @cpehlevan.bsky.social, @kempnerinstitute.bsky.social.
#RL #hippocampus
January 19, 2026 at 9:32 AM
cianodonnell.bsky.social
the longer genAI has been around, the more and more 'professional' the work I am seeing students submit (code, reports, even presentations)

I'm sure they are learning some good practises from GenAI outputs, by osmosis if nothing else

BUT I'm not seeing basic understanding improving in parallel
profaliceroberts.bsky.social Alice Roberts @profaliceroberts.bsky.social · 4h
And that is: asking a student to do a literature review isn’t really about the output, the final product. It’s about learning the process - assessing the quality of the evidence as you read it, learning how your question may evolve as you engage with the literature, learning how to spot SALIENCE 1/2
January 19, 2026 at 9:11 AM
Reposted by Cian O'Donnell
crof.bsky.social
A long, absolutely must-read article from Cory Doctorow: www.theguardian.com/us-news/ng-i...
AI companies will fail. We can salvage something from the wreckage | Cory Doctorow
AI is asbestos in the walls of our tech society, stuffed there by monopolists run amok. A serious fight against it must strike at its roots
www.theguardian.com
January 18, 2026 at 9:01 PM
Reposted by Cian O'Donnell
tjryan.bsky.social
Starting 2026 with a thread a very recent manuscripts at Ryan Lab @tcddublin.bsky.social @tbsi-tcd.bsky.social

"Extinction and subsequent updating of innate fear responses to a visual looming stimulus rely on hippocampus-dependent mechanisms"
journals.plos.org/plosbiology/...

@PLOSBiology
Extinction and subsequent updating of innate fear responses to a visual looming stimulus rely on hippocampus-dependent mechanisms
How does the brain balance innate responses with adaptive learning when reacting to threats? This study shows that repeated exposure to looming stimuli reduces innate defensive responses via hippocamp...
journals.plos.org
January 2, 2026 at 4:34 PM
Reposted by Cian O'Donnell
hannahtarder-stoll.bsky.social
Can reward improve memory for what came before it? 🌟

In a registered report with @duncanlabuoft.bsky.social & @megschlichting.bsky.social, we reconcile mixed findings from past studies: reward retroactively boosts associative—but not item—memory, and only in reward-sensitive individuals!
OSF
osf.io
January 12, 2026 at 5:41 PM
Reposted by Cian O'Donnell
annaschapiro.bsky.social
Really thrilled that this paper led by @neurozz.bsky.social is now published in its final version in @elife.bsky.social!!

This is a memory-focused (as opposed to RL-focused) account of the detailed characteristics of forward and backward awake and sleep replay!

elifesciences.org/articles/99931
A unifying account of replay as context-driven memory reactivation
A context-driven memory model simulates a wide range of characteristics of waking and sleeping hippocampal replay, providing a new account of how and why replay occurs.
elifesciences.org
January 15, 2026 at 1:57 PM
cianodonnell.bsky.social
my favourite thing about the rise of the xG metric is how much it has boosted statistical thinking among causal football fans

even the 'x' part being short for 'expected'... surely invented by some statistician who once wrote

E[G] = ...
cianodonnell.bsky.social Cian O'Donnell @cianodonnell.bsky.social · 4d
such a great #dataviz showing that football chance creation (xG) is an excellent predictor of team points

if you squint a bit you can also see that good teams overperform xG because they have elite goalscorers, and vice versa

also enjoying the Villa outlier 💜💙

from www.bbc.co.uk/sport/footba...
scatter plot of 200 premier league teams points per game vs xG difference per game from last 10 premier league seasons. Clear correlation, almost hugging the identity line but with some outliers like Sunderland and Aston Villa highlighted [Data source Opta, image from BBC website]
January 15, 2026 at 1:11 PM
cianodonnell.bsky.social
such a great #dataviz showing that football chance creation (xG) is an excellent predictor of team points

if you squint a bit you can also see that good teams overperform xG because they have elite goalscorers, and vice versa

also enjoying the Villa outlier 💜💙

from www.bbc.co.uk/sport/footba...
scatter plot of 200 premier league teams points per game vs xG difference per game from last 10 premier league seasons. Clear correlation, almost hugging the identity line but with some outliers like Sunderland and Aston Villa highlighted [Data source Opta, image from BBC website]
January 15, 2026 at 1:07 PM
Reposted by Cian O'Donnell
jorgefmejias.bsky.social
We recently did just that to validate a method to measure functional inter-area interactions, and it worked like a charm (academic.oup.com/nc/article/2...). It also puts the model results appear early in the story (in our case, in fig 3 of a 11-fig paper).
Functional connectivity drifts during sleep as a marker of fluctuations in the level of consciousness
Abstract. During the wake–sleep cycle, consciousness waxes and wanes, and this is thought to be reflected in varying levels of integration between brain ar
academic.oup.com
January 15, 2026 at 10:11 AM
Reposted by Cian O'Donnell
athenaakrami.bsky.social
We're already in the phase that each of us (Iranians) by now know someone killed/injured in our circle of friends/family. And this is despite the continued internet blackout, when millions haven't still managed to hear from their family/friends since Thursday, January 8th. This tells a lot ... 😥😥
January 14, 2026 at 9:18 PM
Reposted by Cian O'Donnell
kristorpjensen.bsky.social
Personally I think this is actually very valuable. 'Conceptual models' can easily ignore important details and may not work in practice. Coding it up (i) makes the details explicit, (ii) ensures they are compatible, and (iii) clarifies to everyone else _exactly_ what the model is.
January 14, 2026 at 5:05 PM
Reposted by Cian O'Donnell
drjoehanson.bsky.social
I am calling for a complete and total boycott of the Mercator projection in all news stories about Greenland until every member of the American public has seen this
January 13, 2026 at 5:32 PM
Reposted by Cian O'Donnell
cianodonnell.bsky.social
I like the postdiction idea, but tbh feel a lot of 'Figure 7' models in otherwise experimental papers don't meet that bar.

Instead they are often used as confirmation that the main conceptual model put forward by the paper can be reproduced computationally, without necessarily adding extra insights
January 14, 2026 at 4:40 PM
Reposted by Cian O'Donnell
fleurzeldenrust.bsky.social
Was reading this: press.princeton.edu/books/hardco... Relevant for here: if there are always multiple theories that could fit with an experimental outcome and vice versa, a lot of back and forth is needed. In that sense it is silly to insist on either one order.
Why Trust Science?
Why the social character of scientific knowledge makes it trustworthy
press.princeton.edu
January 14, 2026 at 8:11 AM
Reposted by Cian O'Donnell
csdashm.com
I think that’s a key point: how large is the space of possibilities and how much of it is covered by the modeling approach? That said, I also think of Barbara Webb saying “model early and often” to get true iteration in ideas and experiments and analysis. No one size fits all here.
January 12, 2026 at 6:02 PM
Reposted by Cian O'Donnell
hritz.bsky.social
This is the way. Starting very early, eg with parameter/model recovery, can even inform task design. Wish I was more consistent with this.

If you need something to point to, great tutorial paper by Anne Collins & Bob Wilson:
elifesciences.org/articles/49547
Ten simple rules for the computational modeling of behavioral data
Computational modeling of cognitive and neuroscience data is an insightful and powerful tool, but has many potential pitfalls that can be avoided by following simple guidelines.
elifesciences.org
January 12, 2026 at 1:10 PM
Reposted by Cian O'Donnell
lukesjulson.bsky.social
On some level, this is just another version of the hypothesis-driven vs. discovery-driven science question. We rarely model first bc we don’t collect data with fine-grained hypotheses in mind. Our hypotheses are things like “mPFC activity will be different in groups A and B.” We collect the data,
January 12, 2026 at 1:45 PM
cianodonnell.bsky.social
when doing neuroscience projects I often advocate for computational modelling, followed by data analysis to test model's predictions.

however a few times now I have had pushback from collaborators/reivewers suggesting it would be better to do the data analysis first, then the modelling.

thoughts?
January 12, 2026 at 12:47 PM
Reposted by Cian O'Donnell
profrachelgan.bsky.social
Following coverage over the weekend of Sir Paul Nurse's comments that suggested that the only reason that a Fellow should be expelled from @royalsociety.org is scientific misconduct, I have written to him to explain the risks such an attitude poses of increasing sexual harassment in STEM.
Dear Sir Paul, Re: Royal Society Code of Conduct I am sure that many scientists have written to you about the specific question of Elon Musk’s Fellowship and whether, under the Royal Society’s Code of Conduct, his retaining that Fellowship is appropriate. I will not rehash these issues. Instead, as a female scientist with extensive experience of activities aiming to increase equality, diversity and inclusion in the engineering and physical sciences sector, I am writing to you (in a personal capacity) to ask you to reconsider the statements you have recently made in this context to the UK press about the Royal Society’s Code of Conduct and how it is applied. A 2018 report from the joint National Academies of the United States of America, concluded that “sexual harassment is common in academic science, engineering, and medicine” and that “greater than 50 percent of women faculty and staff and 20–50 percent of women students encounter or experience sexually harassing conduct in academia”. This report described codes of conduct that make clear that sexual harassment is unethical and will not be tolerated as a “powerful incentive for change”. The authors also noted that sexual harassment can have significant and damaging effects on the integrity of research. In my own praxis, I have found that clear and consistently-implemented codes of conduct that address these issues make female scientists and engineers safer, and allow them to focus more effectively on their research. For codes of conduct to have such a positive effect, it is vital that sanctions for actions which transgress the code are meaningful and substantial. I was hence aghast to realise that in an interview with the Financial Times published on 9/1/26, you appear to have suggested that the Royal Society “should only expel fellows if their science proved “faulty or fraudulent or highly defective””. Moreover, in a further interview with the Guardian on 11/1/26 you suggested that the code “may need to be looked at again”, with the implication that your aim would be to remove the option of sanctions on Fellows for reasons not strictly related to faults or defects in their research. I suggest that changing the Royal Society’s code of conduct so that the likelihood of serious sanctions for sexual harassment is reduced, would directly endanger women who interact with the Royal Society at events or otherwise, and would provide a licence to harass to the already powerful people on whom the Society bestows fellowship. The implications of your words - that under your leadership the only infringements of the code which are likely to receive the sanction of the Fellowship being removed are those related to research misconduct - already risk empowering harassers. You stated, in the Financial Times interview, that “there’s many bad people around, but they have made scientific advances”. Given this awareness of the possibility of bad actors in our scientific community, it is wholly irresponsible to suggest that the Royal Society would not act to sanction these people if they harass more vulnerable scientists. I am hence writing to request that you retract any suggestion that the Society’s Code of Conduct should be changed so that the only reason a Fellow might be sanctioned by the removal of their Fellowship is “faulty or fraudulent or highly defective” research. This action is necessary to safeguard female scientists, a requirement placed on the Society by safeguarding legislation and UK statutory guidance. Yours sincerely, Professor Rachel A. Oliver.
January 12, 2026 at 8:59 AM
Reposted by Cian O'Donnell
esqueer.net
I know it's hard to shock these days but Elon Musk going full Rhodesia defender and calling for the ethnic cleansing of the US should still shock you.

These are neo nazi views that were only previously shared on explicitly neo nazi sites.
A screenshot of a post on X (formerly Twitter) by Elon Musk. Musk replies "Yes" to a post by "The American Tribune." The American Tribune post reads: "Further, what they've managed to accomplish in South Africa is what they want to inflict on all of us. The anarchotyranny, expropriation, and race communism that destroyed Rhodesia and South Afric..." The text cuts off at the "Show more" link. A screenshot of a post by "The American Tribune" quoting a post by "Humble Flow." The American Tribune text reads: "Further, what they've managed to accomplish in South Africa is what they want to inflict on all of us. The anarchotyranny, expropriation, and race communism that destroyed Rhodesia and South Africa are the same things they are bringing to America and the rest of the Occident to turn us into the Global Favela." The quoted post by Humble Flow includes a vintage photograph of three white soldiers in short shorts and military fatigues holding rifles in a brush environment (evoking the Rhodesian Bush War). The caption reads: "What they once did to Rhodesia, they now attempt to the whole world. Resist." A screenshot of a thread on X involving Elon Musk. At the top, a reposted tweet by "Hugh Anthony" reads: "I will never stop. REMIGRATION TILL I DIE!" Below this, Elon Musk replies "Yes" to a post by user "RAMZPAUL." RAMZPAUL’s post quotes the Hugh Anthony tweet and states: "Demographic replacement is genocide." A screenshot of a post by Elon Musk quoting a post by "Will Tanner." Musk writes: "Unfortunately, this is where America is headed." Will Tanner's post reads: "High civilization existed for a brief flicker of a moment in Rhodesia, a few decades where the primordial darkness of the Stone Age was vanquished. But then it returned, in the name of equality, and Zimbabwe is once again a ruin..." Below the text is a six-panel comparison image titled "THEN" and "NOW." ​Rhodesia: "Then" shows a clean, orderly city street; "Now" shows people behind a locked gate. ​South Africa: "Then" shows a bustling city skyline with traffic; "Now" shows a crowd holding a sign that reads "kill WHITE." ​America: "Then" shows a historic city street scene; "Now" shows a white square with the text "COMING SOON." A watermark at the bottom right reads "NOWHITEGUILT.ORG."
January 11, 2026 at 2:43 PM
Reposted by Cian O'Donnell
nicolecrust.bsky.social
And this model by Daniel Bennett/@yaelniv.bsky.social links mood & optimal RL under changing conditions:

pubmed.ncbi.nlm.nih.gov/34516150/
A model of mood as integrated advantage - PubMed
Mood is an integrative and diffuse affective state that is thought to exert a pervasive effect on cognition and behavior. At the same time, mood itself is thought to fluctuate slowly as a product of feedback from interactions with the environment. Here we present a new computational theory of the va …
pubmed.ncbi.nlm.nih.gov
January 11, 2026 at 12:10 PM
Reposted by Cian O'Donnell
nicolecrust.bsky.social
There's much talk these days about what's required for "real" intelligence, such as world models and the like. One thing we know about the human evolution of it: it required/requires emotion. That may have been better appreciated in 1938 than today.

archive.org/details/in.e...
Subtract the emotional element from religion, literature, and art and there would be left the merest dross, an utterly meaningless mass of symbols and responses. Emotionless, man could not produce poetry, music, the graphic arts, or any of the splendid products of a refined culture. The generous sweeps of the imagination that have brought forth romantic movements in the world^s history, the noble aspirations of the intellect that have led whole peoples to seek a higher life, the tremendous activities of organized groups which have spread the best tokens of their civilizations to other lands, the splendid achievements of individual pioneers, adventurers, craftsmen, artisans, composers, inventors, scientists, and spiritual leaders, would never have come to pass in an emotionless world.
January 11, 2026 at 7:51 AM
cianodonnell.bsky.social
Such a great idea
aforli89.bsky.social Angelo Forli @aforli89.bsky.social · 8d
For decades, imaging has forced the same trade-off:
Set a frame rate → sacrifice resolution or field of view to go fast.

We break that rule

Using an EVENT-BASED CAMERA, we record neural activity without frames. Signals are captured only where and when they occur:

www.biorxiv.org/content/10.6...
Ultrafast Frame-Free Imaging of Neural Activity with Event Cameras
Frame-based fluorescence imaging has long defined how neural activity is optically measured. This approach requires acquiring all pixels within an image, regardless of whether they carry meaningful ne...
www.biorxiv.org
January 10, 2026 at 2:27 PM
Reposted by Cian O'Donnell
aforli89.bsky.social
For decades, imaging has forced the same trade-off:
Set a frame rate → sacrifice resolution or field of view to go fast.

We break that rule

Using an EVENT-BASED CAMERA, we record neural activity without frames. Signals are captured only where and when they occur:

www.biorxiv.org/content/10.6...
Ultrafast Frame-Free Imaging of Neural Activity with Event Cameras
Frame-based fluorescence imaging has long defined how neural activity is optically measured. This approach requires acquiring all pixels within an image, regardless of whether they carry meaningful ne...
www.biorxiv.org
January 10, 2026 at 1:41 PM