exam 2 Flashcards
what is one use for both language and music?
communication
are language and music processed separately or together in the brain?
separately
what types of music are most easily recognized across cultures?
dance, lullaby (and healing and love, but less obvious)
music + language (theory that these developed from a common precursor)
musilanguage
speech vs. music (4 each)
speech: timbre, pitch, rapid processing (20-40ms), left hemisphere
music: timbre, pitch, slower but precise changes, right hemisphere
how do we read a spectrogram?
x = time, y= frequency, z = energy/power/intensity (color)
what is the speech-to-song illusion?
when a phrase was repeated enough, it started to sound like a tune
mapping drequencies (apex, basilar membrane, base)
apex: low frequencies (least stiff)
basilar membrane: vibrates up and down
base: high frequencies (stiffest)
maps by frequency in the auditory system
tonotopic maps (similar to retinotopic maps in the visual system)
types of brain areas (music and language) (5)
temporal modulation (doesn’t like sounds that stay the same for a long time), likes pitch, speech-selective component (reacts to structure, not meaning), music-selective component (responds to music sounds), language-selective component
types of aphasia (2)
Broca’s aphasia: trouble producing speech (comprehension intact)
Wernicke’s aphasia: trouble comprehending speech (production intact, word salad)
amusia
fine-fitch discrimination
does musical meaning fit into syntax and semantics?
no; music is less strict than language
what brain areas respond to groove (music)?
motor and reward
are concepts fixed?
no
neural effect of expertise
higher activity in right FFA (and in high-level visual cortex) for expertise-related category
what does the greebles study show us?
FFA can become selective for categories that are completely novel and not relevant for naturalistic perception (but will not activate more than it does with faces)
neural coupling/inter-subject correlation (ISC)
how aligned our brains are to each other while performing a (highly complex) task; brain alignment increases with mutual understanding; strength of neural alignment with experts predicts exam scores
generative vs. discriminative inference
generative: use information from prior stages to determine what is happening in the world; can estimate a boundary from prior knowledge
discriminative: what you think the world should look like; determines whether features are dissimilar enough to be categorized differently; doesn’t use prior knowledge to help with categorization, but can learn from prior experience
inference
the act of computing unknown/latent/hidden variables, given observed variable
[inference by] sampling
a way of approximating a result using prior beliefs without having to use probability distributions (active neurons = 1, inactive neurons = 0)
Marr’s level of analysis
computational –> representation/algorithm –> implementation; interdisciplinary connections across levels needed to understand what’s going on in the brain
principal component analysis (PCA)
reduce dimensionality + reduce feature collinearity + increase useful variance + reduce noise; ignore dimensions/parts of our data that do not maximize variance (want 2-dimensional data)
encoding model
estimating the semantic tuning of each voxel in the brain (how the brain will respond to each stimulus)
key take-away of semantic category lecture
category information is widely distributed across many areas of the cortex; stimulus-selective regions are “selectivity peaks” in the distribution
modern neural networks
larger versions of earlier networks (more layers ~ deep)
AlexNet
the first neural network that could distinguish between thousands of items
deep neural networks (DNNs)
good prediction for human behavior; greatly improved the ability to predict performance, judgments, and error patterns compared to e.g., HMAX; capture an internal structure/similarity pattern (RSA) that matches better to our cognition
representational similarity analysis (RSA)
quantifying the joint similarity structure of a set of items; the more similar results are, the more likely that those artificial models are doing the same thing; internal structure/similarity pattern captured by DNNs that matches better to our cognition
a theory of how categorization works
categories as multidimensional Twizzlers; if two things have features in common, different regions might respond almost the same and you can no longer tell them apart in said region
deepfakes
mimic appearance, voice, mannerisms, etc.; can recreate content that maintains features enough that you can fool the sensory system of a cognitive animal (e.g., humans)
word embedding models
capture semantic relationships between concepts; words that co-occur are often encoded similarly in hidden layer
