![]() They must know the set of possible rules, how often they switch, etc. In its simplest form, this type of latent state inference process presupposes that the animals have previously learned about the structure of the task. Lesioning prefrontal cortex impairs performance on tasks that require rule inference ( Milner, 1963 Dias et al., 1996) and neurons in prefrontal cortex track the currently inferred rule ( Mansouri et al., 2006). Such inference is associated with activity in the prefrontal cortex ( Durstewitz et al., 2010 Milner, 1963 Boettiger and D’Esposito, 2005 Nakahara et al., 2002 Genovesio et al., 2005 Boorman et al., 2009 Koechlin and Hyafil, 2007 Koechlin et al., 2003 Sakai and Passingham, 2003 Badre et al., 2010 Miller and Cohen, 2001 Antzoulatos and Miller, 2011 Reinert et al., 2021 Mansouri et al., 2020), suggesting that its neural mechanisms are distinct from incremental learning. ![]() These more task-specialized dynamics are often modeled by a distinct computational mechanism, such as Bayesian latent state inference, where animals accumulate evidence about which of several ‘latent’ (i.e., not directly observable) rules currently applies ( Sarafyazd and Jazayeri, 2019 Collins and Koechlin, 2012 Behrens et al., 2007 Gershman et al., 2014 Bartolo and Averbeck, 2020 Qi et al., 2022 Stoianov et al., 2016). ![]() However, in other circumstances, learning can also be quicker and more specialized: for instance, if two different stimulus–response rules are repeatedly reinforced in alternation, animals can come to switch between them more rapidly ( Asaad et al., 1998 Rougier et al., 2005 Harlow, 1949). This type of learning works by incremental adjustment that can allow animals to gradually learn an arbitrary task – such as a new stimulus–response discrimination rule. A long tradition of research in areas like Pavlovian and instrumental conditioning has focused on elucidating general-purpose trial-and-error learning mechanisms – especially error-driven learning rules associated with dopamine and the basal ganglia ( Daw and O’Doherty, 2014 Daw and Shohamy, 2008 Daw and Tobler, 2014 Dolan and Dayan, 2013 Doya, 2007 O’Doherty et al., 2004 O’Reilly and Frank, 2006 Rescorla, 1988 Schultz et al., 1997 Yin and Knowlton, 2006 Day et al., 2007 Bayer and Glimcher, 2005 Lau and Glimcher, 2008 Samejima et al., 2005 Padoa-Schioppa and Assad, 2006). Intelligence requires learning from the environment, allowing one to modify their behavior in light of experience. Our results shed light on the computational interactions between rule switching and rule learning, and make testable neural predictions for these interactions. By modeling behavior, we found the animals learned the axis of response using fast inference ( rule switching) while continuously re-estimating the stimulus–response associations within an axis ( rule learning). Each rule was compositional, requiring the animal to discriminate one of two features of a stimulus and then respond with an associated eye movement along one of two different response axes. To this end, we studied how monkeys switched between three rules. Here, we show these two processes, rule switching and rule learning, rely on distinct but intertwined computations, namely fast inference and slower incremental learning. Often we must do both at the same time, switching between known rules while also constantly re-estimating them. To adapt to a changing world, we must be able to switch between rules already learned and, at other times, learn rules anew.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |