1In rational analysis, what is the core assumption about how people adapt to their environment and tasks?
People are treated as approximately optimal machines that adapt in the best possible way.
2How is learning modeled within the framework of rational analysis?
Learning is modeled as optimal in the sense of probability theory, typically via Bayesian update (Bayes' rule).
3What is the poverty-of-the-stimulus argument regarding the 'auxiliary fronting' rule in English questions?
Children correctly learn the structurally-highest verb rule without hearing direct evidence that the simpler closest-verb rule is wrong.
4In the generative grammar view, what enables children to correctly generalize syntactic rules without sufficient positive evidence?
An innate Universal Grammar that privileges hierarchical structure over linear order.
5In the Bayesian LOT model of syntax acquisition, why do learners converge on the structurally-based rule for question formation?
The structural rule is considered the simplest or most efficient program, giving it a higher prior probability.
6What operation on a set of strings is represented by the Kleene star, denoted as \$X^*\$?
The Kleene star generates the set of all possible finite concatenations of strings from the base set X, including the empty string.
7The formal language \$L = \{a^n b^n \mid n \in \mathbb{N}\}\$ is often used to model what type of grammatical construction in natural language?
Center-embedding, such as in relative clauses where the number of nouns must match the number of verbs.
8In Bayes' rule, what term represents the probability of the evidence given a specific hypothesis?
The likelihood, denoted as \$P(E|H)\$.
9What is the term for the probability of a hypothesis before considering new evidence, denoted as \$P(H)\$?
The prior probability.
10The posterior probability, \$P(H|E)\$, is proportional to the product of the _____ and the _____.
likelihood; prior probability
11In Bayesian learning, how can one 'reverse-engineer' cognitive structures like biases or primitives?
By working backward to find the priors and likelihoods under which observed human behavior is the optimal posterior result.
12What is the core idea of the Language of Thought (LOT) hypothesis proposed by Fodor (1975)?
Humans possess basic conceptual primitives and logical operations that are combined compositionally to form complex thoughts.
13In the Bayesian LOT framework, how is the prior probability of a hypothesis (a LOT program) determined?
It is a function of the program's complexity; more complex hypotheses requiring more primitives have lower prior probability.
14The complexity penalty on hypotheses in the LOT framework is analogous to what principle from information theory?
The Minimum Description Length (MDL) principle, which favors shorter descriptions or simpler models.
15In the Bayesian LOT model of number learning, what developmental shift is mirrored by the posterior probability shifting to favor a recursive counting hypothesis?
The shift mirrors children's progression from being one-, two-, or three-knowers to grasping the general counting principle (CP-knower).
16According to Alan Newell, why is 'playing 20 questions with nature' an inadequate scientific approach?
It only reveals behavior in isolated binary experiments, failing to show how underlying universal laws integrate.
17What alternative to the '20 questions' approach did Newell advocate for, which is the core of cognitive modeling?
Understanding by building, which involves creating process models and simulations to explain cognition.
18According to Kriegeskorte & Douglas, what is the role of Neuroscience in its relationship with Cognitive Science and AI?
To constrain models to be consistent with brain data and explain neuronal activation patterns and biological plausibility.
19What is the defining characteristic of a 'process model' in cognitive science?
It describes the sequence of information processing steps the mind goes through, typically at Marr's algorithmic level.
20In the model space proposed by Kriegeskorte & Douglas, what do the two main axes represent?
Biological fidelity (how well the brain is captured) and Cognitive fidelity (how well the mindehavior is captured).
21What level of analysis in Marr's framework addresses the question of 'why' a behavior occurs and what problem is being solved?
The computational theory level.
22What level of analysis in Marr's framework describes the specific information processing steps and representations used?
The algorithmic theory level.
23What is the central assumption of Computational Rationality theories?
Users act in accordance with what is best for them, given the limits imposed by their cognitive architecture and task environment.
24In the Computational Rationality framework, what are 'internal bounds'?
Constraints from within the agent's cognition, such as memory limits, perception, motor abilities, and history of experiences.
25What type of optimality explanation is used when behavior is optimal given the statistical regularities of the environment and the agent's history, without modeling internal cognitive constraints?
Type II - Ecological optimality.
26What type of optimality explanation is used when behavior is optimal given an agent's internal cognitive bounds and the immediate task environment, but not the broader ecological history?
Type III - Bounded optimality.
27The classical Psychological Refractory Period (PRP) effect, where the second of two rapid responses is delayed, is an example used to illustrate which type of optimality?
Type III - Bounded optimality, as it requires assumptions about internal processing architecture (e.g., a bottleneck).
28A model that explains behavior as being optimal relative to both internal agent constraints and broader environmental history falls under which category of optimality?
Type IV - Ecologically bounded optimality.
29What is the key difference between a decoding model and an encoding model in neuroscience?
A decoding model maps brain activity to a stimulus, while an encoding model specifies how stimulus features are transformed into neural responses.
30In the visual encoding model by Kay et al. (2008), what mathematical functions were used to approximate the receptive fields of orientation-selective neurons in V1?
Gabor-like filters.
31What is the term for a systematic organization on the cortical surface where neighboring neurons respond to neighboring locations in the sensory world (e.g., visual field or body surface)?
A topographic map.
32The discovery by Harvey et al. (2013) of gradual changes in preferred numerosity across the cortical surface provides evidence for what kind of cognitive structure?
A topographic map for the abstract quantity of numerosity.
33Studies comparing responses to dot patterns versus Arabic numerals in numerosity-selective brain regions suggest these maps encode _____ quantity, not abstract symbolic number.
perceptual
34What network of brain regions is known to respond to a wide variety of cognitively demanding tasks, including quantitative cognition, working memory, and attention?
The multiple demand network (MDN).
35Deep Convolutional Neural Networks (DCNNs) are inspired by what organizational principle of the biological visual cortex?
A hierarchical organization where successive layers process the outputs of preceding layers to build increasingly complex feature representations.
36The process of modifying an input image to maximize the activity of a specific unit or layer in a DCNN is used to achieve what?
To visualize the features that the unit or layer has learned to prefer.
37What is Mitchell's (1998) formal definition of machine learning?
A program learns from experience E on task T with performance P, if its performance on T, measured by P, improves with E.
38What is the goal of supervised learning?
To learn a function that maps new inputs to correct outputs, based on training data consisting of labeled input-output pairs.
39What is the primary goal of unsupervised learning?
To find structure, such as clusters or lower-dimensional representations, in input data that has no pre-existing labels.
40What kind of feedback signal does a reinforcement learning agent use to learn?
Delayed and coarse feedback signals, such as a reward or punishment (e.g., winning or losing a game), rather than step-by-step labels.
41What is the key advantage of Multi-Voxel Pattern Analysis (MVPA) over traditional univariate fMRI analysis?
MVPA can detect subtle, distributed patterns of activity across voxels, even when the average activation in a region does not change between conditions.
42The finding that fMRI can decode stimulus orientation from V1 using large voxels was initially puzzling because each voxel averages over many fine-scale _____, which should cancel out.
orientation columns
43What is the core data structure used in Representational Similarity Analysis (RSA) to characterize the relationships between neural patterns evoked by different stimuli?
A Representational Dissimilarity Matrix (RDM).
44How is a Representational Dissimilarity Matrix (RDM) constructed?
By computing a measure of dissimilarity (e.g., 1 - correlation) for every pairwise combination of stimulus-evoked neural response patterns.
45What is a major advantage of using RSA and RDMs for comparing data from different sources (e.g., human fMRI, monkey single-unit, computational models)?
It provides a common framework for comparing representational geometry, abstracting away from the specific units of measurement of each source.
46What statistical technique can be used to visualize the high-dimensional structure of an RDM in a low-dimensional (e.g., 2D) space?
Multidimensional Scaling (MDS).
47What is the primary trade-off between EEG and fMRI as neuroimaging methods?
EEG has very high temporal resolution but coarser spatial resolution, while fMRI has high spatial resolution but poor temporal resolution.
48What is a major limitation of classical Event-Related Potential (ERP) analysis that makes it difficult to study internal cognitive stages?
Latency jitter across trials smears the averaged signal, obscuring the timing of processes not strictly locked to an external event.
49In the HSMM-MVPA method developed by Anderson et al. (2016), what brief feature in the EEG signal is hypothesized to mark the onset of a new cognitive processing stage?
A 'bump,' modeled as a 50 ms half-sine wave.
50What is the 'linking assumption' that connects the ACT-R cognitive architecture to the HSMM-MVPA analysis of EEG data?
Each firing of a production rule in the ACT-R model corresponds to the onset of a bump in the EEG signal.
51In the associative recognition task, the 'fan effect' refers to the finding that response times are slower and errors are more frequent when a word is associated with _____ items in memory.
multiple (a high-fan word)
52How does constraining a bottom-up HSMM model with the stage structure predicted by a top-down ACT-R model affect its performance?
It improves the cross-validated fit to the data, suggesting the theoretical structure from ACT-R is meaningful.
53What is 'bump-aligned' averaging, and how does it improve upon classical ERPs?
It time-warps EEG trials to align them based on the onset of internal stages (bumps), resulting in sharper, more interpretable neural components.
54According to the lecture on reasoning in communication, what type of reasoning allows us to understand messages beyond their plain, conventional meaning?
Reasoning about what has been said, why it was said, and what has not been said.
55In the 'reference game' experiment, if a speaker sees a robot with a blue cap and a green monster with a scarf, but the listener only sees those two plus a green monster with a blue cap, what message should the speaker send to identify the green monster with a scarf?
The speaker should send 'scarf' because it is unambiguously true and more informative than 'green monster', which could refer to two different creatures.
56In Gricean reasoning, the inference that saying "I read most of the papers" implies "I did not read all of the papers" is known as a _____.
scalar implicature
57What is the central idea behind Grice's Maxim of Quantity that leads to scalar implicatures?
A cooperative speaker will be as informative as is required and will not use a weaker statement if a stronger, true statement is available.
58What is the name for a computer program trained on human-provided labels to classify emails as spam or not spam?
A supervised learning classifier.
59What is the primary function of cross-validation in machine learning?
To estimate how well a model will generalize to an independent, unseen dataset.
60When a machine learning model is trained on N-1 examples and tested on the remaining one, repeating for all N examples, what is this procedure called?
Leave-one-out (LOO) cross-validation.
61What technique trains a classifier in one context (e.g., awake state) and tests its performance in another (e.g., sleep state) to probe for stable neural representations?
Cross-classification.
62What is the term for a cognitive model based on probability theory, which is widely used in psychology and linguistics?
A probabilistic cognitive model.
63What is the main critique of classical generative grammar's explanation for syntax learning, from a Bayesian perspective?
It posits a complex, innate 'organ' (Universal Grammar) rather than explaining learning through probabilistic inference and priors favoring simplicity.
64In the Language of Thought (LOT) framework for string generation, what is the role of the `flip(p)` primitive?
It is a stochastic primitive that returns TRUE with probability p, enabling the creation of probabilistic grammars.
65A LOT program for an `a^n` language that uses a recursive call, an `if` statement, and a `plus` operation would have a prior probability of \$p^3\$ if each of these _____ primitives has a prior of \$p\$.
string-generating
66The Bayesian LOT learner struggles with languages like '1,2,3,4' (limited embedding) because its simplest hypotheses naturally express _____ recursion.
unbounded (or full)
67Newell's ambition to 'make psychology more like physics' was based on the idea that the mind, existing in the physical universe, should be describable by _____.
universal laws
68A model of a driver texting that explores different strategies for interleaving attention between the road and the phone is operating at which of Newell's bands of cognition?
The cognitive band (time scale of seconds).
69According to Anderson & Schooler (1991), the fact that human memory seems to decay is ecologically optimal because it reflects what statistical property of the environment?
Information encountered more recently is more likely to be needed again in the near future.
70In the Computational Rationality framework, the set of behaviors possible given both internal cognitive constraints and external environmental rules is the _____ of the two sets.
intersection
71In the Kay et al. (2008) study, the model could not freely reconstruct images but instead performed image ____ by selecting the best match from a candidate set.
identification
72The fact that color appeared in reconstructions from the Nishimoto et al. (2011) movie study, despite color not being in the V1 model, is due to an _____.
implicit prior (average colors associated with certain contrast patterns in the YouTube candidate set)
73The encoding model used to find numerosity maps fits a Gaussian tuning function to each voxel's time series, defined by what two parameters?
Preferred numerosity (the mean/peak) and tuning width (the standard deviation).
74A key difference between classical DCNNs and the human visual system is that DCNNs are typically feedforward, while the brain has extensive horizontal and _____ connections.
feedback (or recurrent)
75When decoding dream content, Horikawa et al. (2013) first trained a decoder on voxel patterns from _____ subjects viewing labeled images.
awake
76In RSA, when testing a hypothesis about animacy, the model RDM would assign low dissimilarity to a pair of stimuli if they share the same _____.
animacy state (i.e., both are animate or both are inanimate)
77Why is a tutor system that can infer the process behind a student's error considered superior to one that only checks the final answer?
It can provide tailored feedback and instruction targeted at the specific misconception or procedural mistake.
78EEG signals measured at the scalp are primarily generated by the summed _____ of large populations of aligned pyramidal neurons in the cortex.
postsynaptic potentials
79In the ACT-R model of the associative recognition task, the 'retrieve-to-reject' strategy is used to correctly identify _____ foils.
repaired
80In the Anderson et al. (2016) paper, the sustained parietal old/new effect in the EEG was not modeled as a bump but as occurring during a _____ preceding the response.
flat