Modeling the Brains Operating System Dana H. Ballard Computer - - PowerPoint PPT Presentation

Modeling the Brain’s Operating System

Dana H. Ballard Computer Science Dept. University of Austin Texas, NY, USA International Symposium “Vision by Brains and Machines” November 13th-17th Montevideo, Uruguay

Embodied Cognition

Maurice Merleau-Ponty 1906- 1961

World Body Brain

Timescales

10 -2 10 -1 10 1 10 2 10 0 Round-trip through Cortical Memory Shortest Recognition time Modal fixation time Attention Switching Time Sentence generation Speed Chess minimum search Activity time 10 3 Memory encoding

sec Continuous Discrete

QuickTime and a MPEG-4 Video decompressor are needed to see this picture.

Marr Brooks

=

Behaviors compete for body’s motor resources Behaviors obtain sensory information Behaviors are scheduled from a pool

Three Levels of a Human “Operating System”

Behavior

1 2 3

Task: Make a PBJ sandwich

Computational Abstraction Hierarchy

Component: Remove Jelly Jar Lid Routine: Locate Lid

Multi-tasking

As revealed by gaze sharing in human data

10 20 30 40 50 60 172 10

3

177 10

3

182 10

3

187 10

3

Mid-Block 40m

time(sec) Mid-block sign Car Intersection sign Eye Stop sign

0 5 10 15

Shinoda and Hayhoe, Vision Research 2001

Roelfsema et al PNAS 2003

Visual Routines

Introducing “Walter” Pickup cans Stay on sidewalk Avoid obstacles QuickTime and aYUV420 codec decompressorare needed to see this picture.

Control of visuo-motor routines

“active” “inactive”

+ only ~4 can run simultaneously + 100-200ms update per behavior

Walter’s Visual Routines

image Can locations Sidewalk location 1-d obstacle locs

You are here state action

Reinforcement Learning Primer : Before Learning

γ γ γ γ γ γ γ γ γ

policy value

Reinforcement Learning Primer : After Learning

Microbehavior for Litter Cleanup

• 2b. Value of Policy

Q θ,d

d θ

• 2a. Policy
• 1. Visual

Routine Heading from Walter’s perspective

Learned Microbehaviors

Litter Sidewalk Obstacles

Microbehaviors and the body’s resources

“active” “inactive”

• Walking direction uses weighted average of Q values.
• Gaze direction must use a single best Q value.

The best Q given a sample state The expected Q given the state uncertainty

Which Microbehavior should get the gaze vector?

• bs

can side

• bs

can side

Performance Comparison

Walter crosses the street Pickup cans Stay on sidewalk Avoid obstacles QuickTime and aYUV420 codec decompressorare needed to see this picture.

QuickTime and a decompressor are needed to see this picture. QuickTime and a decompressor are needed to see this picture.

Running Behaviors: Eye Movement Trace

Three trials

Eyetracker in V8 helmet

QuickTime and aTIFF (Uncompressed) decompressorare needed to see this picture.

A curved path in real space Produces the perception

• f a

straight path in visual space Human Ss walk Walter’s route in Virtual Reality. Their 6 dof head position and 2 dof gaze positions are continuously tracked. Three subjects were used. The resultant video and eye track signal are scored frame-by-frame.

Methods

QuickTime and aYUV420 codec decompressorare needed to see this picture.

A human walks Walter’s route

Obstacle Litter Sidewalk Corner Crosswalk Otherside

Human data: individual fixations

20 40 60 80 100 120

• bs

side litter Other 10 20 30 40 50 60 70

• bs

side litter Other

Walter(3 trials) Human subjects(3)

Walter and the humans have similar task prioritie

Human data: Two samples with different contexts

Obstacle avoidance Litter Sidewalk Crosswalk Otherside

Near

• bstacles

Approaching crosswalk

Walter

On Crosswalk Approaching crosswalk Walter and the human Ss all exhibit context sensitivities. Human gaze locations are interpreted based on gaze location. The actual internal state is unknown. Waiting for light Walter Human Ss

5 10 15 20 25

corner side

• bs
• therside

0.5 1 1.5 2 2.5 3 3.5 4 4.5

corner light side

5 10 15 20 25 30 35 40 45 50

crosswalk

• therside
• ther

10 20 30 40 50 60 70 80

corner light side

5 10 15 20 25

crosswalk

• herside
• ther

Scheduling Context

2 4 6 8 10 12 14

corner side

• bs
• therside

Rewards can be changed quickly

Litter Sidewalk Obstacles

10 20 30 40 50 60

• bs

side litter

Walter Humans

Changing the reward schedule

10 20 30 40 50 60

• bs

side litter 10 20 30 40 50 60

• bs

side litter 10 20 30 40 50 60

• bs

side litter

“ignore the litter” “ignore the obstacles”

2 4 6 8 10 12 Match No Match

Saliency Map vs Gaze

courtesy of program provided by

• Dr. Laurent Itti

at the iLab, USC Match No match

Credit Assignment - MIT model

Credit Assignment - Our Model

The laboratory at Rochester

Computer Science Cognitive Science

Dana Ballard Mary Hayhoe Brian Sullivan Jelena Jovancevic Constantin Rothkopf

Alumni

Chen Yu Pili Aivar Nathan Sprague Jochen Triesch Al Robinson Neil Mennie Weilie Yi Jason Droll Xue Gu Jonathan Shaw

QuickTime and aTIFF (Uncompressed) decompressorare needed to see this picture.

QuickTime and aTIFF (Uncompressed) decompressorare needed to see this picture.