clean discussion

Author: yukomunakata

ch8/bg/README.md

@@ -72,13 +72,13 @@ One interesting side-effect of PD meds is that a subset of patients develop a ga
 
 Although this very simple model can account for the key qualitative features of dopamine-based learning in the BG to promote adaptive action selection, and also for these fascinating patterns of effects in PD patients, there are a number of more complex issues that must be solved to produce a more realistic model of the full complexities of the decision making process that underlies complex behavior.
 
-To start, decision making is not just a simple Go/NoGo decision on a single action. The more elaborate models of the BG circuitry (available at [Michael Frank's website](http://ski.clps.brown.edu/BG_Projects/Emergent_7.0+/) allow for it to select among multiple actions (where action selection involves both Go and NoGo activity for multiple actions in parallel), and explore i) the differential roles of dopamine on learning vs. choice (i.e., risky decision making), ii) the function of the subthalamic nucleus and the 'hyperdirect' pathway for preventing impulsive actions in response to decision conflict, (iii) the role of cholinergic interneurons for optimizing learning as a function of uncertainty, and (iv) hierarchical interactions among multiple cortico-BG circuits for more advanced learning and abstraction of hierarchical task rules during decision making that supports generalization to new situations.
+To start, decision making is not just a simple Go/NoGo decision on a single action. The more elaborate models of the BG circuitry (available at [Michael Frank's website])(http://ski.clps.brown.edu/BG_Projects/Emergent_7.0+/) allow for it to select among multiple actions (where action selection involves both Go and NoGo activity for multiple actions in parallel), and explore i) the differential roles of dopamine on learning vs. choice (i.e., risky decision making), ii) the function of the subthalamic nucleus and the 'hyperdirect' pathway for preventing impulsive actions in response to decision conflict, (iii) the role of cholinergic interneurons for optimizing learning as a function of uncertainty, and (iv) hierarchical interactions among multiple cortico-BG circuits for more advanced learning and abstraction of hierarchical task rules during decision making that supports generalization to new situations.
 
 Beyond these functions, in the real world, the dopaminergic outcomes associated with actions almost never come immediately after the action is taken -- often multiple sequences of actions are required, with outcomes arriving some minutes, hours or even later! In our more complete PBWM (prefrontal-cortex basal-ganglia working memory) model covered in the Executive Function Chapter, we show how these same BG dynamics and learning mechanisms can support maintenance and updating of activation-based "working memory" representations in PFC, to bridge longer gaps of time. 
 
-Furthermore, properties of brain systems that drive phasic dopamine firing, covered in the RL and PVLV models, help to transfer phasic dopamine signals from firing at the time of later outcomes, to earlier stimuli that reliably predict these later outcomes -- this is helpful for driving action learning to achieve *sub-goals* or *milestones* along the way toward a larger desired outcome. Furthermore, we'll see that a synaptic tagging-based *trace* learning mechanism is very effective in bridging these temporal gaps, and solves a number of different problems that other mechanisms cannot.
+Furthermore, properties of brain systems that drive phasic dopamine firing, covered in the RL and PVLV models, help to transfer phasic dopamine signals from firing at the time of later outcomes, to earlier stimuli that reliably predict these later outcomes -- this is helpful for driving action learning to achieve *sub-goals* or *milestones* along the way toward a larger desired outcome. We'll also see that a synaptic tagging-based *trace* learning mechanism is very effective in bridging these temporal gaps, and solves a number of different problems that other mechanisms cannot.
 
-Another critical element missing from this model is the ability to explicitly represent the nature of the outcomes of different actions, and to reason about these outcomes in relation to factors such as effort, difficulty and uncertainty -- these capabilities require the functions of the orbitofrontal cortex (OFC), anterior cingulate cortex (ACC), and other ventral / medial PFC brain areas, all working in conjunction with these basic BG and dopaminergic systems. Developing such models is at theforefront of current research.
+Another critical element missing from this model is the ability to explicitly represent the nature of the outcomes of different actions, and to reason about these outcomes in relation to factors such as effort, difficulty and uncertainty -- these capabilities require the functions of the orbitofrontal cortex (OFC), anterior cingulate cortex (ACC), and other ventral / medial PFC brain areas, all working in conjunction with these basic BG and dopaminergic systems. Developing such models is at the forefront of current research.
 
 # References