Lack of invariance problem for speech perception (Theodore)

Compare and contrast communication in humans compared to non-human species.

Be able to describe the primary methods and findings of Werker and Tees (1984).

Be able to define the lack of invariance problem and provide examples of this problem for speech, vision, and conceptual domains.

Be able to describe how voice-onset-time (VOT) is a physical acoustic property that listeners use to recognize /g/ vs. /k/ sounds, just like wavelength is a physical visual property that viewers use to recognize red vs. blue hues.

Be able to describe evidence indicating that (1) listeners process VOT variation categorically and (2) speech sound categories have a graded internal structure.

Describe and recognize an algorithm that could take VOT inputs and provide outputs that converge with categorical perception.

Describe and recognize an algorithm that could take VOT inputs and provide outputs that converge with graded internal structure.

Drawing on the Myers (2007) paper, be able to provide support for the following statement: "Neural basis of speech perception reflects a functional division of labor; front regions resolve category membership and temporal regions resolve category typicality."

Neuroscience 1 and 2 (Theodore)

You should be able to define and describe neuron, action potentials, neuronal computations, inhibitory connections, excitatory connections, and thresholding.

You should be able to determine activation of a post-synaptic neuron given its inputs and threshold.

What does it mean to say that a neuron can be considered as a binary computer?

You should be able to define and describe Hebbian learning. You should be able to explain how Hebbian learning can account for traditional "behaviorist" phenomena (e.g., Pavlovian response, eye blink to a tone).

Given a neural network, you should be able to calculate the output of post-synaptic neurons and determine what kind of logical operation the network is performing.

Given a neural network and a learning rate, you should be able to calculate the change in weights between neurons following Hebb's rule.

You should be able to state location and function of the following structures:

• V1 (primary visual cortex)
• A1 (primary auditory cortex)
• STG (superior temporal gyrus)
• IFG (inferior frontal gyrus)
• Broca's area
• Wernicke's area
• Hippocampus
• Primary somatosensory cortex
• Primary motor cortex
• Occipital cortex
• Frontal lobe
• Temporal cortex
• Parietal cortex

You should be comfortable describing differences between primary and association cortex.

You should be comfortable describing and defining tonotopy and retinotopy.

You should be able to define neural plasticity.

You should be able to discuss critical periods in language acquisition with respect to neural plasticity.

You should be familiar with evidence in support of experience-related plasticity.

Language and the brain (Drown)

Be able to describe lateralization of language function and how it relates to whether an individual is left- or right-handed.

Know the principal characteristics of Broca's aphasia and Wernicke's aphasia.

Describe the reading-specific and domain-general deficits associated with Developmental Dyslexia.

Be able to argue for and against modularity in the brain.

Connectionism 1 (Snyder)

You should understand the following components of a connectionist model: units, connections, weights, and threshold / bias.

Be able to calculate the activation of a unit when you are given information about the unit's inputs, bias, and activation function (e.g., Heaviside).

Be able to explain what a perceptron is.

For a two-place function specified in tabular form, you should be able to use linear separability to see if the function can be computed by a perceptron.

Connectionism 2 (Snyder)

Be able to explain what a hidden layer is, and how it relates to the discovery of the backpropagation algorithm.

Be able to explain the difference between supervised and unsupervised learning.

In the context of critiques of connectionism, be prepared to explain the relevance of the following characteristics of human cognition: systematicity, productivity, compositionality.

Be ready to explain how findings from the nervous system of the nematode raise questions about the explanatory power of connectionism.

According to Randy Gallistel, what are some key characteristics of human memory that are not captured in standard connectionist models? (Be prepared to discuss variable binding, as covered in lecture; and the engram, as covered in Gallistel 2020.)

Embodied cognition/classical view of concepts (Theodore)

You should be able to define concept, describe what is meant by the term conceptual knowledge, and provide an example of a concept.

You should be able to compare and contrast the classical view of concepts and embodied cognition with respect to how concepts are encoded and represented.

You should be able to describe and recognize evidence in support of embodied cognition, including evidence that mental states affect physical states, and that physical states affect mental states.

You should understand how motor theories of speech perception (e.g., analysis by synthesis) are an example of embodied cognition.

Be able to compare and contrast the classical view of concepts and embodied cognition according to Marr's computational and implementational levels of analysis.

Consciousness (Snyder)

How did Adrian Owen use fMRI to detect consciousness in patients who appeared to be in a vegetative state?

How does being conscious relate to having a mind? Contrast the views of Ren Descartes and David Armstrong.

Explain Armstrong's taxonomy of consciousness: Minimal, Perceptual, and Introspective Consciousness.

According to Armstrong, what is the relationship between Introspective Consciousness and Event Memory?

What does David Chalmers mean by "the hard problem" of consciousness?

Classical AI (Snyder)

Describe the Turing Test, and contrast Newell & Simon's definition of intelligence in terms of problem solving.

What is meant by an 'agent' in the context of AI?

What are some advantages of constructing an AI system to solve a specific, narrowly defined problem?

Explain what is meant by a state space (including nodes, actions, goal states, and solution paths). Be prepared to discuss the idea that intelligent decision-making can be viewed as a search problem.

Be able to apply the 2-ply Minimax algorithm, if you are provided with a state space and a heuristic for evaluating different states. Know how to optimize your search by means of alpha-beta pruning.

Modern AI (Snyder and Theodore)

Regarding large language models (LLMs, like Chat GPT), you should be able to:

• Define each of the following terms: large language model; deep learning; transformer; self-attention.
• Compare and contrast language acquisition in LLMs and humans
• Describe two approaches ('cognitive science'; 'dynamical systems') to investigating how LLMs work.
• Describe ethical considerations (as discussed in class) regarding LLMs.