1. Interaction of Sensory and Value Information in Decision-Making
Institute for Theoretical Physics and Mathematics
Tehran
January 16, 2006

2. Alan Rorie

3. low level sensory analyzers
SENSORY INPUT
low level sensory analyzers
REWARD
HISTORY
representation
of stimulus/
action value
DECISION MECHANISMS
Psychologists and economists, however, have long understood that to produce adaptive behavior decision mechanisms must analyze sensory information in the context of the value or utility of available stimuli and actions.
The brain for its part must represent the value of competing stimuli and actions and update this representation
based upon the outcome of ongoing behavior.
Recently we have become interested in studying the neural basis of this representation.
However, we first needed a reliable means of quantifying the relative value that an animal ascribes to competing actions.
To address this problem we have employed a behavioral phenomenon called matching behavior that I will now briefly describe to you
motor output structures
ADAPTIVE BEHAVIOR

4. low level sensory analyzers
SENSORY INPUT
low level sensory analyzers
representation
of stimulus/
action value
DECISION MECHANISMS
Psychologists and economists, however, have long understood that to produce adaptive behavior decision mechanisms must analyze sensory information in the context of the value or utility of available stimuli and actions.
The brain for its part must represent the value of competing stimuli and actions and update this representation
based upon the outcome of ongoing behavior.
Recently we have become interested in studying the neural basis of this representation.
However, we first needed a reliable means of quantifying the relative value that an animal ascribes to competing actions.
To address this problem we have employed a behavioral phenomenon called matching behavior that I will now briefly describe to you
motor output structures
REWARD
HISTORY
ADAPTIVE BEHAVIOR

8. Motion discrimination task with multiple reward conditions.
Monkey must discriminate the direction of the motion.
Differs from the matching task because target values are fixed
Variable coherences span psychophysical threshold, creating a range of difficulties
Imagien this situation …
Creates conflict between sensory and reward information
Only correct choices are rewarded

10. “Absolute” and “relative” reward magnitude
Differ in absolute
reward magnitude
Differ in relative
reward magnitude

11. Effect of absolute and relative reward magnitude on behavior
Absolute magnitude
No effect on choice
Relative magnitude
Biases choices T1 T2
n=51 T1 T2 T1 T2
T1 = 12
T2 = -14
AVE
T1=16
T2= -13
RELATIVE MAGNATUDE
ADDITIVE
Lack of interaction menas that he is always combinin g the motions and reward information
Beta ONE T1 T2 T1 T2

12. We know from behavior: We ask:
Absolute magnitude does not influence choices
Relative magnitude influences choices
Motion coherence influences choices
We ask:
Whether, and how, absolute magnitude, relative magnitude, and motion coherence are represented in LIP as the decision unfolds in time?

13. Area LIP in the Macaque Brain
Sensory-based decisions (Shadlen & Newsome, ‘96, ‘01)
Value-based decisions (Sugrue, Corrado & Newsome, ‘04)

14. LIP neurons are spatially selective
LIP RESPONSE FIELD
GO!

15. LIP neurons are spatially selective
LIP RESPONSE FIELD
GO!

17. Imagien this situation …

18. Representation of Absolute Reward Magnitude in LIP

19. Representation of Absolute Reward Magnitude in LIP

20. Representation of Absolute Reward Magnitude in LIP
Chose In

21. Representation of Absolute Reward Magnitude in LIP
Chose In
Chose Out

22. Representation of Relative Reward Magnitude in LIP

23. Representation of Relative Reward Magnitude in LIP

24. Representation of Relative Reward Magnitude in LIP
Chose In

25. Representation of Relative Reward Magnitude in LIP
Chose In
Chose Out

26. Representation of Relative Reward Magnitude in LIP
Chose In
Chose Out

27. Summary of population activity
Absolute
Relative
Choice
Relative
Absolute
Absolute
Choice
How can we quantify these dynamics?

28. Modeling Dynamics

29. Modeling Dynamics

30. Modeling Dynamics

31. Modeling Dynamics

32. Modeling Dynamics

33. Modeling Dynamics

34. Modeling Dynamics

35. Modeling Dynamics

36. Conclusions: behavior
Absolute reward magnitude does not affect choice
Relative reward magnitude biases choice
Motion coherence biases choice
The biasing effects of relative magnitude and coherence are additive: reward information does not change psychophysical sensitivity to motion coherence (or vice versa).

37. Conclusions: physiology
The representation of sensory and reward information is dynamic; the profile changes dramatically during the course of a trial.
The critical decision variables—relative reward magnitude and motion coherence—are present in LIP at the precise time when the decision is being formed.
Absolute reward magnitude is represented in LIP even though it does not influence choice behavior.
Most single LIP neurons show effects of multiple variables; the representation is multiplexed.

38. Future Directions: Origins of sensory and reward signals
How are sensory and reward signals cast into a common additive currency for guiding decisions?
Why is the profile of signals in LIP changing so dramatically throughout the trial? What does this imply for the computational strategy embodied in cortical circuitry?

39. Indeed there are now no logical (and I believe no
insurmountable technical) barriers to the direct study
of the entire chain of neural events that lead from the
initial central representation of sensory stimuli…to
the detection and discrimination processes themselves,
and to the formation of general commands for
behavioral responses and detailed instructions
for their motor execution.
V . B . Mountcastle, Handbook of Physiology, 1985

40. The Optimal Bias 87% -> 88% OPT= 5.9%
Gets 2%fewer rewards then if he had done nothing at all

43. 55%
78%
47%