Resource-rational analysis is a new cognitive modeling paradigm that can be appliedto all aspects of human cognition (Lieder & Griffiths, 2020), including decision-making,reasoning, memory, perception, judgment, planning, learning, and problem-solving
I think this is a bit crazy
and handling uncertainty according to thelaws of probability are out of reach for people
Ths is also obviously wrong
The brain’s finite computational power limits people’s ability to achieve these nor-mative standards.
I think this is totally wrong.
early stopping—using a fixed and small numberof gradient descent steps in the weights Wi —in this meta-learning algorithm is the same asin section 12.3.2, in that it corresponds to a Gaussian prior with variance proportional to tover the weights Wi
I think this is too much
For example, the dropout algorithm(Srivastava, Hinton, Krizhevsky, Sutskever, & Salakhutdinov, 2014), in which a subset ofweights are not included in each weight update, has been connected to the idea of hav-ing a distribution over parameter θ , and in some cases, this distribution can be shown toalign with the Bayesian posterior
I think this is a bit too much
discriminative approach only models the distribution of the class labels.
I am not sure this is correct.
a hierarchical Bayesian model is a probabilistic model defined over anabstraction hierarchy
Not necessarily.
common cause structure,
fork
. Figure 4.1d is a commoneffect structure, with X2 being caused by both X1 and X3
collier
Figure 4.1All possible sections of a path between two variables in a directed graph. (a)–(b) Causal chains. Dependency flowsalong chains of edges, unless x2 is observed. (c) Common cause. Dependency flows between children of a commonparent unless x2 is observed. (d) Common effect. Dependency does not flow between parents of a common childunless x2 is observed
This is misleading
What constraints on learning are necessary for children to acquire language?• What kinds of category structures are easy or hard to learn?• Do people prefer simpler hypotheses?• What expectations do people have about the strength of causal relationships?• What kinds of structures will people tend to identify in random data?
I don't think it's trivial to formulate these using "prior".
Conventional
"Conventional" here I believe are those approaches before large foundation models.
” Why is one particular way of going beyond the data given here so much morecompelling than others?
A hidden concept here is Occam's razor. In many cases, we somehow come up with parsimonious theories of the generative rule behind the data. This will (I think) be discussed in the following chapters.
How do our minds get so much from so little?
This claim needs context. Consider Figure 1.1: it illustrates what machine learning typically calls few‑shot learning—inferring a task from a small number of labeled examples and then generalizing to new instances of the same task. The crucial question is whether such generalization truly happens without relying on previous experience.
hello
研究計画調書の内容、及び「研究費の応募・受入等の状況」欄の表示内容に不備（文字や図表等の欠落、文字化
heell