Das war eine sehr unterhaltsame Recherche, bei der ich Sepp Hochreiter getroffen habe - ein Pionier des maschinellen Lernens, der mit seiner alten Idee (#lstm) jetzt OpenAi „vom Markt fegen“ will.
Ob dieser alte Algorithmus wirklich das Zeug dazu hat, große Sprachmodelle zu revolutionieren, kann ich schwer einschätzen. Was mir aber immer klarer wurde in letzter Zeit: Transformermodelle sind an ihrer Grenze. Von daher wird sich was bewegen müssen.
New paper ! We measured expected directional effects on thermal infrared satellite images from #TRISHNA, #LSTM or #SBG missions, using simultaneous acquisitions from #LANDSAT and an aerial imager #MASTER from @NASAJPL
Differences up to 4.5 degrees in #TRISHNA field of view have been observed, which can be corrected to less than 2°K using very simple models.
@elduvelle@LMPrida@biorxivpreprint@cogneurophys I’m stoked that it worked out so well! The assessment followed closely the methods in Navas-Olive CNN paper https://elifesciences.org/articles/77772 using F1 (balanced accuracy) to reflect both precision and recall (i.e. sensitivity). So both FN and FPs count against the score, equally. The human raters were around .7 and the monkey data started at ~.5 and reached ~.6 (same as mouse levels!) after retraining. A pleasant surprise, given visible differences in the SWR phenotype between rodent and primate clades!
I think Andrea will post more details soon, but meanwhile, some relevant keywords for interested folks (can you think of others we should use?)