Lecture 11 + Passingham Ch.2 (Classifying Objects) NEW

What is the role of the lateral occipital cortex (LOC)?

The LOC responds preferentially to coherent objects over scrambled ones and supports viewpoint and size invariance in object recognition.

What is the difference between recognizing an object and making sense of it?

Recognizing identifies physical features; making sense involves semantic meaning, salience, use, and classification within the environment.

Why is classification a form of “making sense”?

Classification situates an object within learned conceptual categories (e.g., animate vs. inanimate), linking perception to knowledge.

What is the function of the inferotemporal (IT) cortex?

The IT cortex is critical for object categorization and representing semantic meaning.

How does the ventral visual stream categorize objects?

At multiple levels: superordinate (animate/inanimate), basic (faces/houses), and subordinate (specific exemplars like dog breeds).

How does category size relate to cortical area in encoding studies?

Larger, more abstract categories occupy larger cortical regions; finer distinctions are represented in smaller, distributed areas.

Why can’t V1 distinguish between animal categories?

V1 encodes low-level visual features; category distinctions emerge in higher areas like IT.

How are subordinate categories represented in the brain?

Through distributed voxel patterns across higher visual areas rather than single specialized landmarks.

What did Connolly & Haxby (2012) demonstrate?

That ventral stream activation patterns mirror biological class structure and behavioral similarity judgments.

What continuum organizes object representations in LOC?

A medial-to-lateral continuum from inanimate (medial) to highly animate (lateral).

What is multi-voxel pattern analysis (MVPA)?

A method that analyzes activation patterns across voxels rather than overall activation magnitude.

What is Representational Similarity Analysis (RSA)?

A decoding approach that compares voxel activation patterns to determine how similarly different stimuli are represented.

What is “representational space”?

A conceptual space where stimuli are organized according to similarity in neural activation patterns.

What was the key finding of RSA in ventral stream areas?

Neural similarity patterns matched participants’ behavioral similarity ratings.

What does this imply about category representation?

The ventral stream reflects learned semantic hierarchies that mirror biological taxonomy.

Are fine-grained categories innate?

Evidence suggests they are learned through experience during childhood and adolescence.

What is blindsight?

A condition where patients with V1 damage can respond to visual stimuli without conscious awareness.

What does blindsight demonstrate about V1?

V1 is necessary for visual awareness but not for all visual processing.

What is backward masking?

A method where a stimulus is quickly followed by another, preventing conscious awareness of the first stimulus.

What brain regions are active when stimuli are consciously reported?

Ventral visual areas plus parietal and dorsal prefrontal cortex (often synchronized).

What is Ned Block’s critique?

Some activations may reflect reporting awareness (“access consciousness”) rather than phenomenal awareness itself.

What did eyes-open vs. eyes-closed studies suggest?

Ventral visual stream activation underlies phenomenal awareness without requiring parietal/prefrontal activation.

What brain areas are part of the salience network?

Anterior insula and anterior cingulate cortex.

What happens to awareness under anesthesia (e.g., propofol)?

Reduced activation in insula, ACC, and thalamus; sensory cortices become unresponsive.

Why is the thalamus important for awareness?

It relays sensory information to cortex and may gate conscious access.

What is predictive coding?

A model where the brain generates predictions about sensory input and updates them based on prediction error.

Who first proposed perception as unconscious inference?

Hermann von Helmholtz

What is a generative model?

A high-level internal model that predicts expected sensory input.

What are prediction errors?

Mismatches between predicted input and actual sensory input.

How does perception occur in predictive coding?

Through minimizing prediction error; the “winning hypothesis” becomes perception.

How do predictive coding models differ from feature detection models?

Feature detection is feedforward and stimulus-driven; predictive coding includes top-down expectations and error signals.

What are the two unit types at each level in predictive coding?

Representation (prediction) units and error (mismatch) units.

Why is perception sometimes called “controlled hallucination”?

Because the brain’s expectations shape what we consciously perceive.

What was the big question of Egner et al.?

Whether FFA activity reflects expectation + surprise (predictive coding) or just face features (feature detection).

What does the feature detection model predict for FFA?

Greater activation only when face features are present.

What does predictive coding predict for FFA?

Activation reflects an additive function of expectation and surprise.

What are the FFA and PPA selective for?

FFA: faces; PPA: places/houses.

If predictive coding is correct, what should happen when a face is expected but not shown?

FFA should still show activation due to expectation signals.

How does hierarchical organization apply to both object recognition and predictive coding?

Lower areas process simple features; higher areas integrate them into abstract categories and predictions.

How does classification relate to semantic memory?

Categories are learned associations formed through hierarchical integration across development.

What overall structure does the ventral stream reflect?

A hierarchical and continuous semantic organization from simple features to abstract biological classes.

What is the key shift from classic models to predictive coding?

From purely bottom-up feature detection to dynamic interaction between top-down expectations and bottom-up error correction.