earl miller

Slide1: 

Prefrontal cortex: categories, concepts and cognitive control Earl K. Miller Picower Center for Learning and Memory, RIKEN-MIT Neuroscience Research Center, and Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology www.millerlab.org

Slide2: 

Basic sensory and motor functions Executive (cognitive) control – The ability of the brain to wrest control of its processing from reflexive reactions to the environment in order to direct it toward unseen goals. Volition, goal-direction

Slide3: 

Sensory Motor Learning and memory (Hippocampus, basal ganglia, etc.) Memories, habits and skills Consolidation (long-term storage)

Slide4: 

Sensory Motor Executive Functions goal-related information Consolidation (long-term storage)

Slide5: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Selection (flexibility) Consolidation (long-term storage)

Slide6: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Consolidation (long-term storage) Selection (flexibility)

Slide7: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Consolidation (long-term storage) Selection (flexibility)

Slide8: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Selection (flexibility) Consolidation (long-term storage)

Slide9: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Consolidation (long-term storage) Selection (flexibility)

Slide10: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Consolidation (long-term storage) Selection (flexibility)

Slide11: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Selection (flexibility) Consolidation (long-term storage)

Slide12: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Consolidation (long-term storage) Selection (flexibility)

Slide13: 

Sensory Motor Executive Functions goal-related information Top-down Consolidation (long-term storage) Selection (flexibility)

Slide15: 

Train monkeys on tasks designed to isolate cognitive operations related to executive control. Record from groups of single neurons while monkeys perform those tasks. Our Methods:

Slide16: 

Sensory Motor Bottom-up Executive Functions goal-related information Top-down Consolidation (long-term storage) Selection (flexibility)

Slide17: 

Perceptual Categories David Freedman Maximillian Riesenhuber Tomaso Poggio Earl Miller www.millerlab.org

Slide18: 

Category boundary Prototypes 100% Cat 80% Cat Morphs 60% Cat Morphs 60% Dog Morphs 80% Dog Morphs Prototypes 100% Dog Perceptual Categorization: “Cats” Versus “Dogs”

Slide19: 

“Cats” “Dogs” Category boundary

Slide20: 

. . . . . Fixation Sample Delay Test (Nonmatch) (Match) 600 ms. 1000 ms. 500 ms. Delayed match to category task Test object is a “match” if it the same category (cat or dog) as the sample RELEASE (Category Match) HOLD (Category Non-match)

Slide21: 

A “Dog Neuron” in the Prefrontal Cortex

Slide22: 

To test the contribution of experience, we moved the category boundaries and retrained a monkey

Slide23: 

To test the contribution of experience, we moved the category boundaries and retrained a monkey Old, now-irrelevant, boundary New, now-relevant, boundary

Slide24: 

PFC neural activity shifted to reflect the new boundaries and no longer reflected the old boundaries Old, now-irrelevant, boundary New, now-relevant, boundary

Slide25: 

Freedman, D.J., Riesenhuber, M., Poggio, T. and Miller, E.K. (2001) Science, 291:312-316 Freedman, D.J., Riesenhuber, M., Poggio, T. and Miller, E.K. (2002) J. Neurophysiology, 88:914-928 Freedman, D.J., Riesenhuber, M., Poggio, T. and Miller, E.K, (2003) J. Neuroscience, 23:5235-5246 .

Slide26: 

Category Effects in the Prefrontal versus Inferior Temporal Cortex Activity to individual stimuli along the 9 morph lines that crossed the category boundary PFC Cats Dogs Cats Dogs

Slide27: 

Category Effects were Stronger in the PFC than ITC: Population Index of the difference in activity to stimuli from different, relative to same, category

Slide28: 

Quantity (numerosity) Andreas Nieder David Freedman Earl Miller www.millerlab.org

Slide29: 

Behavioral protocol: delayed-match-to-number task Preventing the monkey from memorizing visual patterns: Position and size of dots shuffled pseudo-randomly. Each numerosity tested with 100 different images per session. All images newly generated after a session. Sample and test images never identical. A. Nieder, D.J. Freedman, and E.K. Miller (2002) Science, 297:1708-1711. Numbers 1 – 5 were used Release Hold

Slide30: 

Standard stimulus Monkeys instantly generalized across the control stimulus sets.

Slide31: 

Standard stimulus Equal area Sample Delay Average sample interval activity

Slide32: 

Standard stimulus Variable features Sample Delay Average delay interval activity

Slide33: 

Low density High density Sample Delay Average sample interval activity

Slide34: 

Characteristics of Numerosity Preservation of numerical order – numbers are not isolated categories. Numerical Distance Effect – discrimination between numbers improve with increasing distance between them (e.g., 3 and 4 are harder to discriminate than 3 and 7) PFC neurons show tuning curves for number.

Slide35: 

Characteristics of Numerosity Preservation of numerical order – numbers are not isolated categories. Numerical Distance Effect – discrimination between numbers improve with increasing distance between them. Numerical Magnitude Effect – discrimination between numbers of equal numerical distance is increasingly difficult as their size increases (e.g., 1 and 2 are easier to tell apart than 5 and 6).

Slide36: 

Numerical Magnitude Effect 1 2 3 4 5 0 5 1 . 0 1 . 5 2 . 0 2 . 5 3 . 0 Bandwidth of tuning curves Average population tuning curve for each number Neural tuning becomes increasing imprecise with increasing number. Therefore, smaller size numbers are easier to discriminate. Average width of population tuning curves Numerosity 1 2 3 4 5 0 2 5 5 0 7 5 1 0 0 N o r m a l i z e d r e s p o n s e ( % ) Numerosity

Slide37: 

Scaling of numerical representations Linear-coding hypothesis Non-linear compression hypothesis symmetric distributions on linear scale (centered on numbers) wider distributions in proportion to increasing quantities symmetric distributions on a logarithmically compressed scale standard deviations of distributions constant across quantities asymmetric on log scale asymmetric on linear scale

Slide38: 

Non-linear scaling of behavioral data Logarithmic scaling

Slide39: 

Logarithmic scaling Non-linear scaling of neural data

Slide40: 

Scaling of numerical representations Linear-coding hypothesis Non-linear compression hypothesis symmetric distributions on linear scale (centered on numbers) wider distributions in proportion to increasing quantities symmetric distributions on a logarithmically compressed scale standard deviations of distributions constant across quantities asymmetric on log scale asymmetric on linear scale

Slide41: 

Scaling of numerical representations Linear-coding hypothesis Non-linear compression hypothesis symmetric distributions on linear scale (centered on numbers) wider distributions in proportion to increasing quantities symmetric distributions on a logarithmically compressed scale standard deviations of distributions constant across quantities asymmetric on log scale asymmetric on linear scale

Slide42: 

Number-encoding neurons A. Nieder, D.J. Freedman, and E.K. Miller (2002) Science, 297:1708-1711. A. Nieder and E.K. Miller (in preparation) A. Nieder and E.K. Miller (in preparation)

Slide43: 

16

Slide44: 

Inferior Temporal Cortex

Slide45: 

Behavior-guiding Rules Jonathan Wallis Wael Asaad Kathleen Anderson Gregor Rainer Earl Miller www.millerlab.org

Slide46: 

What is a rule? Rules are conditional associations that describe the logic of a goal-directed task. Wallis et al (2001)

Slide47: 

Release Hold Match Rule (same) Wallis, J.D., Anderson, K.C., and Miller, E.K. (2001) Nature, 411:953-956

Slide48: 

Sample Nonmatch Rule (different) Test Wallis, J.D., Anderson, K.C., and Miller, E.K. (2001) Nature, 411:953-956 Release Hold Hold Release Test

Slide49: 

Sample Test Test Release Hold The rules were made abstract by training monkeys until they could perform the task with novel stimuli Match Rule (same) Nonmatch Rule (different) Hold Release

Slide50: 

Sample + Cue

Slide52: 

Wallis, J.D., Anderson, K.C., and Miller, E.K. (2001) Nature, 411:953-956

Slide53: 

Rule Representation in Other Cortical Areas PFC ITC PMC

Slide54: 

SAMPLE TEST ROC Value Number of neurons (All recorded neurons) Time from sample onset (ms) PFC Timecourse of Rule-Selectivity Across the PFC Population: Sliding ROC Analysis Note: ROC Values are sorted by each time bin independently Wallis, J.D. and Miller, E.K. (in press) J. Neurophysiology

Slide55: 

Rule Representation in Other Cortical Areas PFC ITC PMC

Slide56: 

PFC Abstract Rule-Encoding in Three Cortical Areas Wallis, J.D. and Miller, E.K. (in press) J. Neurophysiology

Slide57: 

PFC ITC Abstract Rule-Encoding in Three Cortical Areas Wallis and Miller, in preparation Wallis, J.D. and Miller, E.K. (in press) J. Neurophysiology

Slide58: 

Abstract Rule-Encoding in Three Cortical Areas PFC ITC PMC Wallis and Miller, in preparation Wallis, J.D. and Miller, E.K. (in press) J. Neurophysiology Wallis, J.D. and Miller, E.K. (in press) J. Neurophysiology

Slide59: 

Abstract Rule-Encoding was Stronger and Appeared Earlier in the PMC than PFC Latency for rule-selectivity (msec) Number of neurons Median = 410 Median = 310 PFC PMC Wallis and Miller, in press, J. Neurophysiol.

Slide60: 

Abstract Rule-Encoding in Three Cortical Areas PFC ITC PMC Wallis and Miller, in preparation Wallis, J.D. and Miller, E.K. (in press) J. Neurophysiology Wallis, J.D. and Miller, E.K. (in press) J. Neurophysiology

Slide61: 

1. Goal-related information, including the categories and concepts needed for executive control, is represented in the PFC while irrelevant details are largely discarded. 3. This ability of the PFC and related areas to convey categories, concepts and rules may reflect their role in acquiring and representing the formal demands of tasks, the internal models of situations and courses of action that provide a foundation for complex, intelligent behavior. A Model of PFC function: Miller, E.K. (2000) The prefrontal cortex and cognitive control. Nature Reviews Neuroscience, 1:59-65 Miller, E.K. and Cohen, J.D. (2001) An integrative theory of prefrontal cortex function. Annual Review of Neuroscience, 24:167-202 For reprints etc: www.millerlab.org 2. Neural representations of categories and concepts are stronger and more explicit in the PFC than in cortical areas that provide the PFC with visual input (“cats and dogs”, numbers). Highly familiar rules may be more strongly encoded in the PMC than PFC. CONCLUSIONS:

Slide62: 

Active Inactive The PF cortex and cognitive control Phone rings Answer Don’t answer

Slide63: 

Active Inactive The PF cortex and cognitive control Phone rings Answer Don’t answer At home Guest

Slide64: 

Active Inactive The PF cortex and cognitive control Phone rings Answer Don’t answer At home Guest PF cortex

Slide65: 

Active Inactive The PF cortex and cognitive control Phone rings Answer Don’t answer At home Guest PF cortex

Slide66: 

Active Inactive The PF cortex and cognitive control Phone rings Answer Don’t answer At home Guest PF cortex

Slide67: 

Active Inactive The PF cortex and cognitive control Phone rings Answer Don’t answer At home Guest PF cortex

Slide68: 

Active Inactive The PF cortex and cognitive control Phone rings Answer Don’t answer At home Guest PF cortex

Slide69: 

Active Inactive The PF cortex and cognitive control Answer Don’t answer PF cortex Phone rings Guest At home

Slide70: 

Active Inactive The PF cortex and cognitive control Answer Don’t answer PF cortex At home Guest Phone rings

Slide71: 

Active Inactive The PF cortex and cognitive control Answer Don’t answer PF cortex At home Guest Phone rings

Slide72: 

Active Inactive The PF cortex and cognitive control Answer Don’t answer PF cortex Phone rings Guest At home

Slide73: 

Active Inactive The PF cortex and cognitive control Answer Don’t answer PF cortex At home Guest Phone rings

Slide74: 

Its integrative anatomy allows it to rapidly acquire a “map” that specifies which pattern of “tracks” (neural pathways) are needed to solve a given task. The prefrontal cortex may be like a switch operator in a system of railroad tracks:

Slide75: 

Its integrative anatomy allows it to rapidly acquire a “map” that specifies which pattern of “tracks” (neural pathways) are needed to solve a given task. The PF cortex actively maintains this pattern during task performance, allowing feedback signals to bias the flow of activity in other brain areas along task-appropriate pathways. The prefrontal cortex may be like a switch operator in a system of railroad tracks: GOAL-DIRECTION FLEXIBILITY

Slide76: 

Categories: David Freedman Max Riesenhuber (Poggio lab) Tomaso Poggio Numbers: Andreas Nieder David Freedman Rules: Jonathan Wallis Wael Asaad Kathy Anderson Gregor Rainer Other Miller Lab members: Tim Buschman Mark Histed Christopher Irving Cindy Kiddoo Kristin Maccully Michelle Machon Anitha Pasupathy Jefferson Roy Melissa Warden Miller Lab @ MIT (www.millerlab.org)