neural prosthetics iv: pushing preferred directions around beata jarosiewicz laboratory of andrew...

36
Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Upload: edward-miller

Post on 14-Jan-2016

226 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Neural Prosthetics IV:

Pushing preferred directions around

Beata JarosiewiczLaboratory of Andrew Schwartz

Page 2: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

The explanatory gap• Absolute gap:

Why is neural activity accompanied by any conscious experience at all?

• Comparative gap:Why does certain neural activity give rise to, say, visual rather than auditory experience? Or to the experience of red instead of green?

‘Red’

670 nm wavelength light

Neural activity resulting from 670 nm

light

Page 3: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

The explanatory gap• Absolute gap:

Why is neural activity accompanied by any conscious experience at all?

• Comparative gap:Why does certain neural activity give rise to, say, visual rather than auditory experience? Or to the experience of red instead of green?

‘Red’

670 nm wavelength light

Neural activity resulting from 670 nm

light

absolute gap

Page 4: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

The explanatory gap• Absolute gap:

Why is neural activity accompanied by any conscious experience at all?

• Comparative gap:Why does certain neural activity give rise to, say, visual rather than auditory experience? Or to the experience of red instead of green?

‘Red’ ‘Green’

670 nm wavelength light

Neural activity resulting from 670 nm

light

530 nm wavelength light

Neural activity resulting from 530 nm

light

comparative gap

absolute gap

Page 5: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Cortical “deference” vs. “dominance”

from “Neural Plasticity and Consciousness” (Hurley & Noë, Biology & Philosophy, 2003)

Page 6: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Cross-modal remapping: “vOICe”

Left and Right Image is scanned left to right, one snapshot per second. So left and right get encoded by time (and stereo).

Dark and Light Brightness is encoded by loudness: the brighter the visual pattern, the louder the sound.

Up and Down Elevation is encoded by pitch: the higher the position of the visual pattern, the higher the pitch.

Movie by Alex Storer, BU

Page 7: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Cross-modal remapping: “vOICe”2000:

“Well the other day I was again washing dishes. I had let the water out of the sink and turn to get a towel to dry my hands. Then when I turn back to rinse the sink I was stunned to see the sink in a “depth” like image. I stepped away from the sink and walked slowly up to it again to see if my mind was playing tricks on me. No, the feeling of seeing depth in the sink bowl was still there.”

2002:“Just sound?.... No, It is by far more, it is sight! There IS true light perception generated by the vOICe. When I am not wearing the vOICe the light I perceive from a small slit in my left eye is a grey fog. When wearing the vOICe the image is light with all the little greys and blacks. Yet a definite light image.”

- Pat Fletcher, long-term vOICe user

Page 8: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Cross-modal remapping: “vOICe”2000:

“Well the other day I was again washing dishes. I had let the water out of the sink and turn to get a towel to dry my hands. Then when I turn back to rinse the sink I was stunned to see the sink in a “depth” like image. I stepped away from the sink and walked slowly up to it again to see if my mind was playing tricks on me. No, the feeling of seeing depth in the sink bowl was still there.”

2002:“Just sound?.... No, It is by far more, it is sight! There IS true light perception generated by the vOICe. When I am not wearing the vOICe the light I perceive from a small slit in my left eye is a grey fog. When wearing the vOICe the image is light with all the little greys and blacks. Yet a definite light image.”

- Pat Fletcher, long-term vOICe user

Page 9: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

What caused the switch from cortical dominance to cortical deference?

Maybe what’s important is how the information is interpreted downstream.

Is there a way we can experimentally change how the activity of particular neurons is interpreted downstream, and test whether they can change to acquire their newly-assigned function?

Page 10: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

What caused the switch from cortical dominance to cortical deference?

Maybe what’s important is how the information is interpreted downstream.

Is there a way we can experimentally change how the activity of particular neurons is interpreted downstream, and test whether they can change to acquire their newly-assigned function?

Page 11: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

What caused the switch from cortical dominance to cortical deference?

Maybe what’s important is how the information is interpreted downstream.

Is there a way to experimentally manipulate the way neural activity is interpreted downstream, and test whether the neurons “oblige”?

Page 12: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

There is a way, with brain control!

• With the brain control paradigm, we know the precise relationship between the activity of each of the neurons under study and the “behavioral” output.

• This allows for a new kind of perturbation that can target selected subsets of neurons by altering the way those neurons contribute to the object’s movement, and to test whether neurons can change their activity specifically and selectively when the monkey learns to compensate for the perturbation.

Page 13: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

3D center-out task under “brain control”

zzyyxx vbvbvbb 0

Cosine tuning with (intended) movement

direction

cosvbbf 0

iii b(t)f(t)v ''

Population Vector Algorithm

Page 14: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

1) Adaptive: • Each cell’s tuning function (PD and modulation depth) was estimated by

iteratively regressing firing rate against target direction as the monkey performed the brain control task.

• Once the monkey’s performance stabilized, the tuning functions obtained from this iterative regression were frozen and used for decoding (“dPDs”).

2) Control 1 (C1): • Monkey performed brain control using the decoding

parameters obtained from the adaptive session.

3) Perturbation (P): • A subset (~25%) of the recorded units (the

“perturbed units”) were given reassigned dPDs by rotating their original dPDs 90 degrees about the x, y, or z axis (chosen randomly each day).

4) Control 2 (C2): • The perturbation was removed (i.e. the original

dPDs were reinstated).

Experimental setup: 4 sessions

Page 15: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Direction of perturbation

Deflection in early trajectory

Monkey learns to control cursor by end of Perturbation session

C1

Early P

Late P

Page 16: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Possible neural bases1) The perturbed units could have suppressed their contribution to

the PV by firing at baseline rate everywhere; i.e. by decreasing their modulation depths.

2) The perturbed units could have shifted their actual PDs toward their reassigned dPDs.

3) The monkey could have “re-aimed” the cursor to offset the perturbation caused by the reassignment, disregarding the relative contributions of the perturbed vs. unperturbed units to the error. This would appear as a shift in all (perturbed and unperturbed) measured PDs in the direction of the perturbation.

Page 17: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Possible neural bases1) The perturbed units could have suppressed their contribution to

the PV by firing at baseline rate everywhere; i.e. by decreasing their modulation depths.

2) The perturbed units could have shifted their actual PDs toward their reassigned dPDs.

3) The monkey could have “re-aimed” the cursor to offset the perturbation caused by the reassignment, disregarding the relative contributions of the perturbed vs. unperturbed units to the error. This would appear as a shift in all (perturbed and unperturbed) measured PDs in the direction of the perturbation.

Page 18: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Possible neural bases1) The perturbed units could have suppressed their contribution to

the PV by firing at baseline rate everywhere; i.e. by decreasing their modulation depths.

2) The perturbed units could have shifted their actual PDs toward their reassigned dPDs.

3) The monkey could have “re-aimed” the cursor to offset the perturbation caused by the reassignment, disregarding the relative contributions of the perturbed vs. unperturbed units to the error. This would appear as a shift in all (perturbed and unperturbed) measured PDs in the direction of the perturbation.

Target

Deflection caused by

perturbation

Re-aiming direction

Page 19: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Testing the possible neural bases

• To obtain each unit’s “measured” tuning function in each session, the same linear regression was done on all firing rates vs. target directions in that session (but post-hoc):

• measured PD = direction for which its firing rate was the highest (direction at the peak of the cosine fit)

• measured modulation depth = the magnitude of the fitted cosine (the difference between the peak firing rate and the baseline firing rate)

zzyyxx dbdbdbbf 0

Page 20: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Modulation depths did not change

Page 21: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Possible neural bases1) The perturbed units could have suppressed their contribution to

the PV by firing at baseline rate everywhere; i.e. by decreasing their modulation depths.

2) The perturbed units could have shifted their actual PDs toward their reassigned dPDs.

3) The monkey could have “re-aimed” the cursor to offset the perturbation caused by the reassignment, disregarding the relative contributions of the perturbed vs. unperturbed units to the error. This would appear as a shift in all (perturbed and unperturbed) measured PDs in the direction of the perturbation.

Target

Deflection caused by

perturbation

Re-aiming direction

Page 22: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Testing for shift in measured PDs

O

azimuth

MC1

0

0

MR

R

elev

atio

n

M

M

Page 23: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

C1 -> P

P -> C2

Page 24: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

“Dose-response”

Page 25: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Possible neural bases1) The perturbed units could have suppressed their contribution to

the PV by firing at baseline rate everywhere; i.e. by decreasing their modulation depths.

2) The perturbed units could have shifted their actual PDs toward their reassigned dPDs.

3) The monkey could have “re-aimed” the cursor to offset the perturbation caused by the reassignment, disregarding the relative contributions of the perturbed vs. unperturbed units to the error. This would appear as a shift in all (perturbed and unperturbed) measured PDs in the direction of the perturbation.

Target

Deflection caused by

perturbation

Re-aiming direction

Page 26: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Conclusions• This study demonstrates selective functional plasticity, even when

only a global feedback signal is available.

• How can the brain do that? Perhaps noise allows it to explore synaptic weight spaces that optimize cursor control?

• We still don’t know where or how the changes took place that underlie the PD shifts in motor cortex.

• Further studies using the brain control paradigm promise a deeper understanding of plasticity as a neural substrate of learning.

• In combination with the dominance/deference studies, this result might have interesting implications for “neural coding”: perhaps it isn’t as important what “activates” a neuron (i.e. its “tuning function”) – as how that neuron’s activity is used downstream.

Page 27: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Thank you!

• Rob Kass• Meel Velliste• George Fraser• Chance Spalding • Ingrid Albrecht• Nathaniel Daw• Valerie Ventura

• NIH-NINDS-NO1-NS-2-2346

Page 28: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz
Page 29: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Introduction

• Many studies have shown that neurons can change their “coding” properties in association with learning.

• In motor cortex, neurons can change their tuning functions when monkeys adapt to perturbations that interfere with the execution (e.g. Bizzi and collaborators) or visual feedback (e.g. Wise, Paz, Vaadia) of their movements.

• With the brain control paradigm, we know the precise relationship between the activity of each of the neurons under study and the “behavioral” output.

• The brain control paradigm also allows for a new kind of perturbation that can target selected subsets of neurons by altering the way those neurons contributes to the object’s movement; this makes it possible to test whether neurons can change their activity specifically and selectively when the monkey learns to compensate for the perturbation.

Page 30: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Cursor deflection resulting from perturbation

Expected Actual

Rotation axes: red = X

green = Y blue = Z

Page 31: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz
Page 32: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz
Page 33: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz
Page 34: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Expected cursor perturbations

rotation axis:

X = redY = greenZ = blue

Page 35: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

cos)(tvbbf(t) 0

ii

0m

0i bbb

b(t)f(t)v

ii

i '

The “Population Vector” algorithm:

= firing rate at time t

= baseline firing rate

= cell’s preferred direction

= (intended) movement velocity at time t

= angle between the movement direction and the cell’s preferred direction

b

)(tv

0b

f(t)

ib '

imb

= normalized preferred direction of cell i

= maximum firing rate of cell i

Estimating preferred directions:

zzyyxx vbvbvbb 0

Page 36: Neural Prosthetics IV: Pushing preferred directions around Beata Jarosiewicz Laboratory of Andrew Schwartz

Re-aiming occurs later in trajectory