that was fun indeed
12.03.2026 14:54
👍 1
🔁 0
💬 0
📌 0
that was fun indeed
Lovely visit to Birkbeck to talk about avoidance, hosted by @silviaseghezzi.bsky.social. Good engagement from staff and students, really enjoyed visiting. Also got to talk with @matildevaghi.bsky.social, lots to think about in my way home.
Our back-translation of value-modulated attentional capture to rats and mice is now published in Journal of Experimental Psychology: Animal Learning and Cognition, and it's open access! Yay.
psycnet.apa.org/fulltext/202...
Reacquisition after standard extinction is sometimes faster and sometimes slower (doi.org/10.3758/BF03...) and to me the source of these discrepancies is not clear so its good that your model can shed light on this.
The way that all these models (R-W, SOP and comparator) handle differential and unpaired inhibition is by assuming that the context is excitatory and drives inhibition, so I think that the Dickinson & Burke can explain the retrospective extinction effects that you observed.
To mark the 80th anniversary of the Experimental Psychology Society, we invite proposals for a special anniversary symposium reflecting on the current state of experimental psychology and its future trajectory.
Please see below for information on how to submit a proposal!
eps.ac.uk/eps-80th-ann...
Thank you for engaging with all this.
Correct, the original SOP does not explain G3 but Dickinson & Burke (1996 doi.org/10.1080/7139...) proposed a modification (no new parameters) that explains these phenomena (so called "retrospective revaluation" - i.e., learning about absent cues)
If it is a "threshold" then I'd see less of a reason to assume that the memory has been "erased" (it is harder to cross the threshold), I guess it depends on how you define a memory
I think that given conditioned inhibition training, you expect to see animals that "never reacquire" 50% partial reinforcement (this obviously depends on what threshold you use), so I do not see a reason for discarding them. These animals learned inhibition and therefore show retarded acquisition
great that you could visit, very informative and insightful conversations - and obviously a very nice talk
Looks like an amazing opportunity to do a postdoc
New preprint and simulator of associative learning attentional models. Have fun! 👁️
arxiv.org/abs/2602.07519
cal-r.org/index.php?id...
#simulation #associative_learning #attention
As someone with an interest in 1) ways of enhancing extinction learning to attenuate recovery and 2) assessing the evidence for memory erasure, I read this preprint from the Namboodiri lab with attention. Here are a few thoughts on the preprint doi.org/10.64898/202...
A final thought on memory erasure. If a memory has been erased, the animal has no record of that prior experience. I would expect that reacquisition should occur at a similar rate as the original acquisition, not slower. The latter is consistent with conditioned inhibition.
The conclusion is that the treatments “truly erase” memory, and I doubt that this is the case. We have recently argued that there is no convincing evidence for memory erasure, instead the available evidence is better understood as new context-dependent learning doi.org/10.1037/rev0...
The paper is set to compare one model (ANCCR) with latent cause models, but the prediction is consistent with standard associative learning theory. It is argued that R-W does not account for spont recov, but Wagner’s SOP (conceptually, a real-time R-W) anticipates these findings
Compared to standard extinction learning, conditioned inhibition (either differential or unpaired) training resulted in retarded reacquisition and less spontaneous recovery. This extends the finding that conditioned inhibition training attenuates renewal
doi.org/10.1037/0097...
As someone with an interest in 1) ways of enhancing extinction learning to attenuate recovery and 2) assessing the evidence for memory erasure, I read this preprint from the Namboodiri lab with attention. Here are a few thoughts on the preprint doi.org/10.64898/202...
Heading back home from another @exppsychsoc.bsky.social brilliant meeting in London. Here's the information about the next one!
Excellent talk by @katepeters3.bsky.social, who visited us at @notts-psych.bsky.social today to speak about "Effects of environmental enrichment in reducing food seeking and the role of corticolimbic circuits"
This looks very appealing for someone looking to do a PhD in sunny Malaga (Spain) with an excellent mentor and team.
Yes, that phenomenon is also called the "gap-filling effect". However the contiguity that @gershbrain.bsky.social refers to is (to me) a CS duration effect, because in the figure above a delay procedure was used. The manipulation is not about temporal closeness between two events (CS and US)
It is all dependent on parameters and preparations, what's amazing to me about the figure is that the shape of the function is always the same.
It depends on who you ask, much of the evidence for contiguity comes experiments in which the trace interval is manipulated, to that the "delay" vs "trace" is the relevant comparison. But yes some of the wording in the Rescorla review takes on the view that you allude to
Mackintosh's figure (p 203) has a panel D in which contiguity (CS offset to US onset) does not follow the same inverted U pattern. As to why the inverted U, I take it that if the CS is too short it doesn't get processed and therefore less learning. Imagine a 3 ms tone and try to learn from it
I love this figure - it actually appeared first in Mackintosh's 1983 book and Rescorla adapted it - but I interpret it as showing a CS duration effect rather than proper contiguity (contiguity, that is the time between CS offset and US onset, is strong in all the examples by Rescorla)
Thought-provoking analysis by @markhaselgrove.bsky.social on a recent preprint looking at benchmarks for associative learning models 👇
In which 94 benchmark phenomena are identified.
16 are domain- and species-general and should be explained by any theory claiming generality.
30 are species- and/or domain-specific but highly robust across procedures
I looked into this two years ago and concluded that GPower cannot do this power calculation (2x2 W-S). Daniel Lakens has a shinny app that can do it (but there is a parameter [correlation between measures] which gives a lot of flexibility).