Simon Kern
@skjerns.de
Sleep & Memory researcher @ CIMH Mannheim with @gordonfeld.bsky.social . Interested in replay and applied machine learning in the context of episodic and declarative memory.
MEG and Python enthusiast.
MEG and Python enthusiast.
Not disagreeing. I find the results impressive that models first learn by heart until model weights are exhausted, then abstract and compress knowledge by necessity to reuse weights as input entropy exceeds exclusive node capa. IMO that in itself shares a lots w/ "intelligence" however you define it
November 11, 2025 at 1:02 PM
Not disagreeing. I find the results impressive that models first learn by heart until model weights are exhausted, then abstract and compress knowledge by necessity to reuse weights as input entropy exceeds exclusive node capa. IMO that in itself shares a lots w/ "intelligence" however you define it
Shoutout to my fellow TrainEEr assistants @iamra.bsky.social @neurogima.bsky.social @willenjoy.bsky.social @remymasson.bsky.social maybe next PracticalMEEG it's us organizing?! 😄
November 2, 2025 at 9:20 AM
Shoutout to my fellow TrainEEr assistants @iamra.bsky.social @neurogima.bsky.social @willenjoy.bsky.social @remymasson.bsky.social maybe next PracticalMEEG it's us organizing?! 😄
I'll be promoting this at #practicalMEEG2025 next week as well, so if you have any questions feel free to hit me up there :)
October 24, 2025 at 1:40 PM
I'll be promoting this at #practicalMEEG2025 next week as well, so if you have any questions feel free to hit me up there :)
Sorry had to repost - previous link was slightly wrong!
October 24, 2025 at 7:20 AM
Sorry had to repost - previous link was slightly wrong!
We still believe that there is an overlap of activity that can be extracted. However, we think current approaches don't capture it well! Therefore, we started this
#kaggle challenge. Have fun and good luck! :)
#kaggle challenge. Have fun and good luck! :)
October 24, 2025 at 6:55 AM
We still believe that there is an overlap of activity that can be extracted. However, we think current approaches don't capture it well! Therefore, we started this
#kaggle challenge. Have fun and good luck! :)
#kaggle challenge. Have fun and good luck! :)
In our recent study, we struggled to cross-decode using the standard approach used in our line of research (train a time-by-time decoder on the peak of visual decoding accuracy). We know that the decoders work in the context of visual reactivation (cf Kern 2024), but seem to fail in auditory cueing.
Reactivation strength during cued recall is modulated by graph distance within cognitive maps
Items retrieved from cognitive maps are reactivated simultaneously or sequentially depending on the performance, where in the first case the reactivation strength reflects the distance within the map.
elifesciences.org
October 24, 2025 at 6:55 AM
In our recent study, we struggled to cross-decode using the standard approach used in our line of research (train a time-by-time decoder on the peak of visual decoding accuracy). We know that the decoders work in the context of visual reactivation (cf Kern 2024), but seem to fail in auditory cueing.
It has been shown that there is overlap between evoked visual activity and evoked auditory activity, but especially with visual mental imagery (cf the amazing work of
@nadinedijkstra.bsky.social)! Memory replay studies rely on this overlap to detect reactivation during resting state
@nadinedijkstra.bsky.social)! Memory replay studies rely on this overlap to detect reactivation during resting state
October 24, 2025 at 6:55 AM
It has been shown that there is overlap between evoked visual activity and evoked auditory activity, but especially with visual mental imagery (cf the amazing work of
@nadinedijkstra.bsky.social)! Memory replay studies rely on this overlap to detect reactivation during resting state
@nadinedijkstra.bsky.social)! Memory replay studies rely on this overlap to detect reactivation during resting state
We still believe that there is an overlap of activity that can be extracted. However, we think it can be improved! Therefore, we started this @kaggle.com challenge. Have fun and good luck! :)
October 23, 2025 at 12:32 PM
We still believe that there is an overlap of activity that can be extracted. However, we think it can be improved! Therefore, we started this @kaggle.com challenge. Have fun and good luck! :)
It is assumed that there is overlap between evoked visual activity and evoked auditory activity, but especially with visual mental imagery (cf work of @nadinedijkstra.bsky.social )! Nevertheless, in our data, it turned out to be quite difficult to establish a pattern common to the three modalities.
October 23, 2025 at 12:32 PM
It is assumed that there is overlap between evoked visual activity and evoked auditory activity, but especially with visual mental imagery (cf work of @nadinedijkstra.bsky.social )! Nevertheless, in our data, it turned out to be quite difficult to establish a pattern common to the three modalities.
In our own data, we struggled to find replay (cf Kern et al 2025) using such an approach. While we know that this is also due to the method's sensitivity, another bottleneck might be the classifiers. Therefore, we recorded a condition in which participants heard words associated to the stimuli
Challenges in Replay Detection by TDLM in Post-Encoding Resting State
elifesciences.org
October 23, 2025 at 12:32 PM
In our own data, we struggled to find replay (cf Kern et al 2025) using such an approach. While we know that this is also due to the method's sensitivity, another bottleneck might be the classifiers. Therefore, we recorded a condition in which participants heard words associated to the stimuli
Background: In the field of memory research, many studies perform a functional localizer by showing participants stimuli and extracting activity patterns. Then these patterns are compared to a condition of interest (e.g. resting state) - however, how much overlap is there between the representation?
October 23, 2025 at 12:32 PM
Background: In the field of memory research, many studies perform a functional localizer by showing participants stimuli and extracting activity patterns. Then these patterns are compared to a condition of interest (e.g. resting state) - however, how much overlap is there between the representation?