psych 421

deefer


  • a year ago
  • 30
files (15)

Psych421chapter5678.docx

Chapter 5 - Artificially Modeling the mind (200 words)

Hello; for this week, please, read chapter 5 and answer the following questions:

1. How do artificial neural network models work alongside neuroscientific techniques to help us study the brain? What do they add to our understanding?

2. Why do you think training is such a key part of neural network modeling? How does it shape the way these models work?

3. How similar do you think artificial neural networks are to the real brain? Are there aspects where they feel biologically realistic or where they fall short?

4. What are the main features of artificial neural networks? How would you compare them to the physical symbol system model of information processing?

Chapter 6 - The Dynamical Mind (200 words)

During this week, please, read Chapter 6 and answer the following questions:

1. What’s the main idea behind the dynamical systems hypothesis? Do you find it convincing, and why or why not?

2. How does the example of the Watt governor help Van Gelder make his point? Can you think of other examples that might support his argument?

3. Dynamical models seem great for studying how skills like walking develop. Do you think they could also be useful for exploring other cognitive processes? Why or why not?

4. The A-not-B error is a fascinating phenomenon. Do you think there might be other plausible cognitive explanations for why it happens?

Resources for Chapter 6, others are attached

https://www.youtube.com/watch?v=LSHZ_b05W7o

http://www.scholarpedia.org/article/Chinese_room_argument

Chapter 7 - Bayesian Systems in the mind (200 words)

For this week you will read Chapter 7. After that, answer the following questions:

1. Do you think the Bayesian idea that beliefs can have degrees of certainty makes sense? Why or why not?

2. Can you explain how Bayes’s Rule works in simple terms? Why do you think it’s so widely used?

3. What’s the connection between Bayesian thinking and Gestalt approaches to perception? How do they help us understand how we make sense of what we see?

4. What’s the challenge posed by the St. Petersburg game? Do you think distinguishing between monetary value and utility helps solve it?

Chapter 8 - Modules to Build a Mind (200 words)

For this week, please, read Chapter 8 and answer the following questions.

1. What are the main differences between the three agent architectures described in the text? How would you explain these differences to someone new to the topic?

2. What are the key features of modular processing, according to Fodor? Why do you think these characteristics are important?

3. How does the massive modularity thesis differ from Fodor’s ideas about modularity? Which approach do you find more convincing, and why?

4. Why do Cosmides and Tooby argue that a module for detecting cheaters evolved? How convincing do you find their reasoning?

Chapter5.4_Informationprocessinginneuralnetworks-revisedfor3rdedition.ppt

Chapter 5.4:
Information processing in neural networks

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Level of analysis

  • Neural networks are models of information-processing at the algorithmic level

• alternatives to symbolic algorithms

• not implementations of symbolic algorithms

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Modeling information-processing

• Neural networks are algorithmic in a limited sense

• algorithms for updating activation levels

• learning rules are algorithmic

• But not algorithmic in the same way as PSS

• algorithms are not task-specific

• algorithms do not operate over representations

• Algorithms change weights and thresholds at the level of individual units

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Representations

• Representations in a neural network need not be located in distinct physical locations

• The network’s “knowledge” lies in its pattern of weights and thresholds

• The power of distributed (as opposed to localist) networks comes from the fact that the network doesn’t need a separate unit to code every feature to which it is sensitive

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Problem

• In what sense do neural networks actually contain representations?

Superpositional storage

• Once a network has been trained, all its knowledge is encoded in a single set of weights

• Each instance of information-processing involves an input vector and the weight vector

• This makes it difficult to think about the network’s knowledge as composed of discrete items (e.g. particular beliefs)

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

No clear distinction between information storage and information processing

• Parallel rather than serial processing as activation spreads through a network

• Knowledge distributed across a network (rather than stored in discrete symbol structures)

• Processing does not reply on explicit rules (other than those governing how activation flows through the network)

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Biological plausibility: compare to PSS

  • Many cognitive abilities are difficult to model in a rule-based way

Context effects in perception (explosion of rules)

Pattern completion

Language (particularly languages such as English with many irregular forms)

  • Connectionist networks are very successful on pattern recognition tasks (e.g. mine/rock)

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Chapter8.1_Architecturesforintelligentagents-revisedfor3rdedition.ppt

Chapter 8.1:
Architectures for intelligent agents

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Overview

• review general issue of mental architecture

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Agent architectures

  • Agent architectures illustrate the different components that make up an agent and how those components are organized
  • What is an agent?

Definition: an agent is a system that perceives its environment through sensory systems of some type and acts upon that environment through effector systems.

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Note that mental architectures were introduced in the slides to Chapter 5.

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Agent architectures

  • How do cognitive scientists decide which sorts of agents are worth modeling?
  • Three main types of agents

Simple reflex

Goal-based

Learning

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Simple reflex agent

  • Production rules (e.g., IF… THEN rules)
  • Not a cognitive system

No information processing

Simply acting on information

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Simple reflex agent

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Goal-based agent

  • Does not simply act on information
  • Works out the consequences of different possible actions and then evaluates those consequences in light of its goals
  • However, no capacity for learning

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Goal-based agent

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Learning agent

  • Can detect errors

  • Can experiment with different ways of achieving its goals, in light of past failures

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Learning agent

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

*

Chapter5.1_Neurallyinspiredmodelsofinformationprocessing-revisedfor3rdedition.ppt

Chapter 5.1:
Neurally inspired models of information processing

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Overview

  • Introduce basic principles of connectionist networks
  • Introduce different types of activation function

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Features of connectionist networks

• Exploit parallel processing

• Can be used to model multiple satisfaction of soft constraints

• Do not feature explicit (content-specific) rules

• Exhibit graceful degradation

• Intended as models of information-processing at the algorithmic level

• Capable of learning

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Note that this was introduced in chapter 3

*

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Neurons and network units

Fig. 2

An artificial neuron:

Ii Input i

Wi The weight attached to input i

T The threshold of the neuron

X The total input to the neuron

S The output signal

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Activation functions

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Networks and layers

• Basic distinction

Single-unit networks [a.k.a. single layer networks]

Multilayer networks

• Different learning rules

• Only multilayer networks have hidden units

Cognitive Science  José Luis Bermúdez / Cambridge University Press 2020

Chapter5.3_Multilayernetworks-revisedfor3rdedition.ppt
This file is too large to display.View in new window
Chapter8.4_Hybridarchitectures-revisedfor3rdedition.ppt
This file is too large to display.View in new window
Chapter8.2_Fodoronthemodularityofmind-revisedfor3rdedition.pptx
This file is too large to display.View in new window
Chapter7.3_Neuroeconomics_Bayesinthebrain-createdfor3rdedition.ppt
This file is too large to display.View in new window
Chapter7.2_PerceptionasaBayesianproblem-createdfor3rdedition.ppt
This file is too large to display.View in new window
Chapter6.1_Cognitivescienceanddynamicalsystems-revisedfor3rdedition.ppt
This file is too large to display.View in new window
Chapter6.2_Applyingdynamicalsystems-revisedfor3rdedition.ppt
This file is too large to display.View in new window
Chapter5.2_Single-layernetworksandBooleanfunctions-revisedfor3rdedition.ppt
This file is too large to display.View in new window
Chapter8.3_Themassivemodularityhypothesis-revisedfor3rdedition.ppt
This file is too large to display.View in new window
pssh.pdf
This file is too large to display.View in new window
Chapter7.1_Bayesianism_Aprimer-createdfor3rdedition.ppt
This file is too large to display.View in new window
other Questions(10)