Lecture 4: Cognitive Neuroscience Methods I

What this lecture covers

The lecture introduces David Marr's idea that understanding perception requires first asking what problem is being computed and why, before asking how neurons implement it. It works through this "computational theory" level in detail for color vision, covering the ill-posed nature of inferring an object's true color from the light reaching the eye. It then pivots to face perception as the case study for the rest of the methods unit, introducing behavioral (psychophysics) evidence and starting into functional MRI.

This is the fourth lecture in the course, following the neuroanatomy overview and dissection. After watching, you should be able to state Marr's three levels of analysis, explain why color and shape perception are ill-posed problems, and describe the logic of a basic fMRI experiment testing whether a brain region is face-selective.

Key ideas

• Marr's three levels: computational theory (what is computed and why), algorithm and representation (what code or process does it), and hardware implementation (how neurons carry it out); Marr argued the first level is a prerequisite for the other two.
• Ill-posed problems: many perceptual inferences, like recovering an object's reflectance from the light reaching the eye, don't have enough information to be solved uniquely, so the visual system must bring in assumptions.
• Color constancy: a demo shows that changing the color of light illuminating a scene changes the perceived color of an object even when the actual light reaching the eye from that object is held constant, showing the brain is solving for reflectance by estimating the illuminant.
• Prosopagnosia: a face recognition deficit, illustrated by Jacob Hodes, who cannot recognize even close friends and family despite normal intelligence, social skills, and object recognition.
• Super recognizers: people at the opposite end of the face-recognition spectrum, so accurate they sometimes hide the ability to avoid seeming unsettling.
• Dutch politician sorting task: a behavioral study (Jenkins et al.) showing people are poor at sorting photos of unfamiliar faces by identity, suggesting face recognition of unfamiliar people does not rely on an abstract, image-independent representation.
• fMRI and the BOLD signal: functional MRI measures blood oxygenation changes that follow increased neural activity, giving the best non-invasive spatial resolution available in humans but poor temporal resolution.
• Fusiform face area (FFA): a brain region that responds more to faces than to objects or hands in a basic fMRI contrast, used as the running example for how to reason about brain-imaging evidence.

Walkthrough

Marr's computational theory, applied to color (0:10)

Kanwisher frames the mind as a set of computations over representations and introduces Marr's argument that understanding a computation requires first understanding the problem it solves, illustrated with his comparison of studying bird flight through feathers alone versus understanding aerodynamics first. She sets up color vision as the running example: light reflecting off a surface (reflectance) combines with the color of the illuminating light to produce the luminance reaching the eye, and the visual system must recover reflectance from luminance alone.

Why color vision is ill-posed (9:21)

A live class discussion and demo (at the imaging center) explores what color is used for, such as finding ripe fruit and judging food safety, with a reference to a study showing macaques with three color photoreceptors outperform those with two at finding fruit. Kanwisher then formalizes the problem: luminance equals reflectance times illuminant, an equation with too many unknowns to solve directly, making color perception (like shape perception from a 2D image, and word-meaning inference in infants) an ill-posed problem that requires extra assumptions.

The car color demo and psychophysics (19:26)

Students identify the colors of several car images that all have the identical gray pixel value under differently colored lighting, demonstrating that the visual system estimates the illuminant to solve for reflectance. This illustrates Marr's algorithm and representation level: psychophysics (measuring what people perceive) reveals the assumptions the visual system uses, even without touching the brain directly.

Introducing face perception and prosopagnosia (25:32)

The lecture shifts to face perception as the case study for the rest of the methods unit. Kanwisher describes Jacob Hodes, a person with lifelong prosopagnosia who cannot recognize even close friends, and notes about 2% of the population has clinically significant face recognition difficulty, a trait uncorrelated with IQ or other perceptual skills, with "super recognizers" at the opposite extreme.

Behavioral evidence: the Dutch politicians study (42:38)

A class demo has students estimate how many distinct individuals appear in an array of unfamiliar face photos; most badly undercount, while people familiar with those individuals succeed instantly. This replicates a published study and suggests face recognition of unfamiliar people is not based on simple template matching or memorization, since neither could explain performance on people we already know well versus poorly.

Starting functional MRI (47:43)

Kanwisher explains the BOLD signal: increased neural activity draws more blood flow, and the resulting change in blood oxygenation is what fMRI detects, giving good spatial resolution but poor temporal resolution (changes take several seconds). She then describes a basic fMRI experiment: scanning subjects while they view faces versus objects reveals a region with a higher blood-oxygenation signal during faces, and the lecture opens the question of what else, besides genuine face-selectivity, could explain that result.

Before you watch

• Watch the neuroanatomy lecture first for background on cortex and subcortical structures referenced here.
• Familiarity with the previous lecture's discussion of visual motion as an example of a "what is computed and why" problem is helpful, since color vision is treated the same way.

Check your understanding

1. What does it mean for a perceptual problem to be "ill-posed," and why is inferring an object's reflectance from luminance an example?
2. In the car-color demo, why do the cars appear differently colored even though the physical light reaching the eye from each patch is identical?
3. What does the Dutch politicians sorting study suggest about how people recognize unfamiliar versus familiar faces?
4. What physiological process does the BOLD signal in fMRI actually measure, and why does that make its temporal resolution poor?
5. Besides genuine face-selectivity, what other explanations might account for a brain region responding more to faces than objects in a simple fMRI contrast?