This preview shows page 1. Sign up to view the full content.
Unformatted text preview: Neural Representation: Examples from the motor system Motor System Motor System Motor System Georgopoulos et al. 19861990 Central light and 8 lights in a circle around the center Allows for visually guided reaching movement of the arm Measure direction and speed of movement Georgopoulos et al. Train on task 12 months Surgery to install recording chamberover motor cortex Recording over about 20 days, recording from hundreds of neurons Georgopoulos et al. 19861990 Single neuron 6 recording sessions T=cue to reach M=onset of reaching Neurons are encoding the plan to reach Record from each neuron, for each reaching direction Tuning curve for reaching direction: Directional sensitivity Broad tuning Single neuron's Response to multiple reaching directions What is the preferred direction? Georgopoulos et. al. Results: In the population of neurons recorded from, found preferred directions arrayed around the full 360 degrees of the circle Question: How is movement direction for a particular movement act represented? Local vs. Distributed encoding? Local representation For each movement direction only those neurons tuned to that direction are active Georgopoulos hypothesis: Population coding Direction coded by firing pattern across entire population of directionally sensitive cells Cells respond to a greater or lesser extent above or below baseline depending on how close the tobe produced movement direction corresponds to its preferred direction Pattern of firing across the entire population specifies movement direction, with all cells contributing to the representation Distributed representation Combining the responses of a population: The Vector hypothesis Each cell's activity can be thought of as a vector having: Direction (cell's preferred direction) Magnitude (strength of response) Movement direction is coded by the population vector the sum of the vectors represented by all of the cells. Combining vectors (analogous to forces pushing on a ball) Combining vectors Responses of neuron population recorded while planning movements to different directions and coded as vectors Each cluster corresponds to the responses produced in planning a movement to a particular direction With each cluster: Each line = one neuron the direction of the line= the preferred direction of a neuron The length of the line= the strength of the response population vector indicated by the dashed line But Does the Direction Tuning Generalize Across Start Locations? Movement from different start locations Wu & Hatsopoulos (2006) Only 4% of 680 M1 cells showed highly invariant directional tuning across start locations for an external Cartesian reference frame. 8% showed highly invariant tuning in two other reference frames (shouldercentered and jointangle). Takehome message: Most M1 cells do not actually have invariant direction tuning (in any reference frame tested so far). The Problem of Serial Order When you execute a series of movements, how is the sequence encoded? Two possibilities: Associative chaining Parallel, graded activation Associative Chaining Writing B I R D B I R D Direct connections from B to I, I to R, R to D Parallel, Graded Activation Writing B I R D Direct connections output B B I R D Explains transposition errors output I B I R D Averbeck et al. (2002) Monkeys trained to copy figures with joystick. 30 trials for each figure. Activity of neurons in prefrontal cortex recorded. Segment Pattern the average activity pattern over the time period in which the Segment is being drawn. (For correctly executed trials only.) Classification then done for 25 ms bins across each entire trial. Each time bin's activation pattern is compared to the shape's Segment Patterns, and classified as the Segment it is most similar to. Within a time bin across trials, a Segment's Representation Strength is the percentage of trials classified as that Segment. How does a Segment's Representation Strength vary over time? st 1 seg Expected Patterns nd 2 seg rd 3 seg th 4 seg Chaining strength Parallel time Experimental Results st 1 seg nd 2 seg rd 3 seg th 4 seg th 5 seg Supports parallel encoding of sequences Shows Primacy and Recency effects stronger activation for first and last elements of a sequence. Time = 0 is the onset of target figure. Same figure copied in blocks of trials, so monkey knew shape during the waiting period (Time < 0). Errors Primacy and Recency Effects Classification over time bins equal to entire Segments. Shows that errors occurred because upcoming segments activated too strongly. BrainMachine Interface: Using population recordings for neural prosthetics Carmena et al. (2003) Brainmachine interface for restoring motor behaviors in severely paralyzed (spinal cord injury) patients, amputations Logic: record motor planning info from the brain and use it to move a prosthetic What neural signals should be recorded? Carmena et al. create a brainmachine interface that uses the electrical activity of frontoparietal neuronal ensembles to extract motor parameters BrainMachine Interface: Using population recordings for neural prosthetics Carmena et al. (2003) Three tasks: Move the polemove the cursorreach the target Squeeze the poleenlarge the cursor `grab" the target Move and squeeze = reach and grab Three stages: 1. Pole control of cursor 2. Brain control of cursor 3. Brain control of robotic arm Pole control of the cursor: Learning results (blue lines) Stage 2: Brain control of the cursor 1recorded from M1 (3356 cells), S1 (2139), PP (6364) SMA (1619) while monkey performed the task 2algorithms used to determine neural correlates of: hand position, velocity and gripping force 3converted these to parameters that controlled cursor movements brain control "during brain control mode, animals initially produced arm movements, but they soon realized that these were not necessary and ceased to produce them.....we removed the pole after the monkey ceased to produce arm movements in a session" Brain control of the cursor: Learning results (red lines) Stage 3: Brain control of robotic arm Parameters extracted from brain signals used to drive a robotic arm Robotic arm reached and grasped> these parameters moved the cursor Monkeys learned to perform the task by "directing" the robotic arm Brain control of the robotic arm: Results Brain control occurs without arm muscle activity Absence of EMG activity in several arm muscles: ...
View Full Document
- Spring '08