Tracking the Emergence of Conceptual Knowledge during Human Decision Making. Kumaran, Summerfield, Hassabis, Maguire. Neuron 2009.

  1. Shows that hippocampus and ventromedial PFC “underpins the emergence of conceptual knowledge and its effect on choice behavior.  Critically, the hippocampus alone supported the efficient transfer of knowledge to a perceptually novel setting.  These findings provide compelling evidence that the hippocampus supports conceptual learning…”
  2. Mentions in related work RL
  3. The paradigm used here is based on a task known to be dependent on the hippocampus <well in this case, if you find activity in the hippocampus you shouldn’t be surprised, but on the other hand you should not take it to mean that these things don’t happen outside the hippocampus>
  4. They had to play a meterologist <!> and figure out how each of 8 weather patterns (deterministically) determined rain or shine
  5. <I don’t know what this means, but seems important:> “Since all eight patterns were constructed from the same four fractals, successful performance required participants to use associative information consisting of shape-location and shape-shape conjunctions rather than elemental information (e.g., single shape) as studied previously…”
  6. Although subjects could just learn a mapping from the stimuli to responses, there were other cues that could be leveraged to make predictions
  7. Stimulus had  either patterns 1+3 or 2+4.  The location of #1 gives the correct response, and the position of #3 or #4 when coupled with #2 also gives the correct response.
    1. This allows for correct responses in some cases given partial information, and was therefore tested as a way of measuring conceptual understanding
  8. Listing of other ways categorical knowledge was tested previously, this task is different in that it test “a higher-order conceptual structure.”
  9. These tasks, as well as learning artificial grammars often are not related to the hippocampus <bingo – see comment in #3> and involve straitum and lateral PFC
  10. Didn’t find differences in fMRI between the two classes of stimulus
  11. For models, compared against Q-learning and “the moving average” <I don’t know what the latter is, but QL doesn’t make much sense to compare against, because (I don’t think) there is a sequential aspect to the problem.>
    1. “For our data set, the state-space  model generated closer fits to the observed data compared to these two other approaches…” <wait – whats the state-space model?>
    2. Anyway, that model “…allowed us to create participant-specific trial-by-trial parametric regressors… that we used to regress against the learning trial fMRI data.”
  12. Rewards were given for correct answers, and therefore areas of the brain commonly found to activate in RL tasks showed up <wait I thought before they said only the hippocampus>
  13. <I think they can make both claims because they use the delta of two different tests to determine what is active for different functions>
  14. “our findings are consistent with the notion that  areas such as the parahippocampal cortex play a greater role in memory for individual associative pairings.”
  15. “Thus far, our findings provide behavioral evidence that conceptual knowledge is acquired gradually during learning, plays a significant role in guiding participants’ choices, and is underpinned by neural activity in hippocampus, posterior cingulate cortex (PCC), and vmPFC.”
  16. They move on to test more specifically that the hippocampus and vmPFC (“two interconnected brain regions”) interact during the emergence of conceptual knowledge during decision making.  A different analysis shows this “… greater conceptual knowledge was associated with stronger coupling [between the two].  This finding provides evidence that the hippocampus and vmPFC act as a circuit during the acquisition of conceptual knowledge during decision making.”
  17. In “probe trials” there was no reinforcement, and cues were incomplete (as the prediction could still be made with partial information if the encoding scheme was processed).  Some trials allowed correct determination of the answer, and some did not provide enough information (so accuracy would be at chance).
    1. When the stimulus provided enough information to accurately predict the outcome, there was significantly more activity in left hippocampus and vmPFC than in cases where the correct determination was not possible
  18. “Taken together, our results show that neural activity in the hippocampus, and vmPFC tracks the emergence of knowledge during the Initial session and its deployment during conceptual decision making.”
  19. Then, if the rules stayed the same, but the shapes used differ, people were able to transfer the rules from one set of images to another
    1. In this task, activity in the hippocampus correlated with performance in this task, but not so in the vmPFC
    2. Hippocampal activity was not correlated to reaction time
  20. In this session, the hippocampus, vmPFC, and PCC correlated with a correct response
  21. “…support a model in which the hippocampus and vMPFC interact during conceptual decision making but play dissociable roles… [the hippocampus]  lmay house abstract conceptual representations of the task structure, which endure across time  (e.g., across experimental sessions).  In contrast, the vMPFC may act primarily to guide choices online, by integrating abstract information received from the hippocampus with stimulus-bound value information.”
  22. “…the hippocampus supports conceptual learning through its unique associative memory capabilities, which also explain its critical role in other relational memory domains…”
  23. “…this neural circuit  may support a common core function during goal-directed cognition, regardless of whether it is oriented to the past, present, or future, whereby the vMPFC mediates the online integration and evaluation of associative information conveyed by the hippocampus.”

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: