Priorities - A: high, B: medium, C: low; Status - _: unfinished, X: finished, C: cancelled, P: pending, o: in progress, >: delegated. Covey quadrants - Q1 & Q3: urgent, Q1 & Q2: important
AXStudy chapter 13 (2005.12.12)
AXStudy chapter 12: stress and human error (2005.12.12)
AXStudy chapter 11: Attention, time-sharing and workload (2005.12.12)
AXStudy chapter 7: memory and training (2005.12.08)
AXStudy chapter 8 (2005.12.08)
AX@1200 Print and hand in my paper for mie1407f (2005.12.05)
AXWrite abstract (2005.12.04)
AXWrite conclusion (2005.12.03)
AXWrite about past recommendations (2005.12.03)
AXWrite everything else needed for paper (2005.12.02)
AXPoint out how system was interconnected with human operators (2005.12.02)
AXTie paper into decision-making (2005.12.02)
AXWrite about system constraints (2005.12.02)
Priorities - A: high, B: medium, C: low; Status - _: unfinished, X: finished, C: cancelled, P: pending, o: in progress, >: delegated. Covey quadrants - Q1 & Q3: urgent, Q1 & Q2: important
A_Write everything else needed for (2005.12.02)
AXDouble-check paper contents (2005.12.02)
AXFinish writing introduction to paper (2005.12.01)
AXReview structure of paper (2005.11.30)
AXHard to get a global picture, easy to miss important problems, had to confirm status by calling up other operators (2005.11.29)
AXInaccuracy of the reserve generation preview information led to insufficient action (2005.11.29)
AXSystem operator was unable to make recognition-primed decisions and lacked the knowledge or confidence to make decisions on the fly (2005.11.28)
ACCost of a false alarm made the system operator hesitate (2005.11.28)
AX@1400-1600 Describe task proximity problem for MIE paper (2005.11.25)
Lab reports
Priorities - A: high, B: medium, C: low; Status - _: unfinished, X: finished, C: cancelled, P: pending, o: in progress, >: delegated. Covey quadrants - Q1 & Q3: urgent, Q1 & Q2: important
AX@1330-1630 Q1 Write introduction and discussion for lab report (2005.11.24)
Priorities - A: high, B: medium, C: low; Status - _: unfinished, X: finished, C: cancelled, P: pending, o: in progress, >: delegated. Covey quadrants - Q1 & Q3: urgent, Q1 & Q2: important
A_Check out Engineering Communication Centre
A_@1500-1800 Q1 Go to engg psych lab in RS303 from 2005.11.09 (2005.11.09)
A_@0900-1100 Q2 Study for MIE1407F from 2006.01.26 (2006.01.26)
A_@1100-1200 Q1 Attend lecture in MP134 from 2006.01.26 (2006.01.26)
AC@1500-1700 Q1 Attend lecture in BA1230 from 2005.12.13 (2005.12.13)
AX@1100-1200 Q1 Attend lecture in MP134 from 2005.12.01 (2005.12.01)
AX@1500-1700 Q1 Attend lecture in BA1230 from 2005.11.29 (2005.11.29)
AX@0900-1100 Q1 Work on lab report (2005.11.24)
AX@1100-1200 Q1 Attend lecture in MP134 from 2005.11.24 (2005.11.24)
AX@1500-1800 Q1 Go to engg psych lab in RS303 from 2005.11.23 (2005.11.23)
AX@1500-1700 Q1 Attend lecture in BA1230 from 2005.11.22 (2005.11.22)
AXSketch system requirements for MIE paper (2005.11.21)
AX@1100-1200 Q1 Attend lecture in MP134 from 2005.11.17 (2005.11.17)
AX@1500-1800 Q1 Go to engg psych lab in RS303 from 2005.11.16 (2005.11.16)
AX@1500-1700 Q1 Attend lecture in BA1230 from 2005.11.15 (2005.11.15)
AX@1200-1230 Revise lab 4 graphs (2005.11.10)
AX@1100-1200 Q1 Attend lecture in MP134 from 2005.11.10 (2005.11.10)
AXRevise result descriptions (2005.11.09)
AX@1500-1700 Q1 Attend lecture in BA1230 from 2005.11.08 (2005.11.08)
AX@1100-1200 Q1 Attend lecture in MP134 from 2005.11.03 (2005.11.03)
AX@1500-1800 Q1 Go to engg psych lab in RS303 from 2005.11.02 (2005.11.02)
AX@1500-1700 Q1 Attend lecture in BA1230 from 2005.11.01 (2005.11.01)
AX@1100-1200 Q1 Attend lecture in MP134 from 2005.10.27 (2005.10.27)
AC@1500-1800 Q1 Go to engg psych lab in RS303 from 2005.10.26 (2005.10.26)
AX@1500-1700 Q1 Attend lecture in BA1230 from 2005.10.25 (2005.10.25)
AX@1100-1200 Q1 Attend lecture in MP134 from 2005.10.20 (2005.10.20)
AC@0900-1100 Q2 Study for MIE1407F from 2005.10.20 (2005.10.20)
AX@1500-1800 Q1 Go to engg psych lab in RS303 from 2005.10.19 (2005.10.19)
AC@1500-1700 Q1 Attend lecture in BA1230 from 2005.10.18 (2005.10.18)
AX@0800-1000 Write lab report, argh (2005.10.14)
AX@0730-1000 Q1 Toronto Western Hospital fieldtrip (2005.10.13)
AXWrite my part of lab report for MIE1407F (2005.10.13)
AC@0900-1100 Q2 Study for MIE1407F from 2005.10.13 (2005.10.13)
AC@1100-1200 Q1 Attend lecture in MP134 from 2005.10.13 (2005.10.13)
AX@1500-1800 Q1 Go to engg psych lab in RS303 from 2005.10.12 (2005.10.12)
AX@1500-1700 Q1 Attend lecture in BA1230 from 2005.10.11 (2005.10.11)
AX@1100-1200 Q1 Attend lecture in MP134 from 2005.10.06 (2005.10.06)
AX@1430-1440 E-mail Prof. Milgram asking if he can help me find another lab partner (2005.10.06)
AC@0900-1100 Q2 Study for MIE1407F from 2005.10.06 (2005.10.06)
AX@1500-1800 Q1 Go to engg psych lab in RS303 from 2005.10.05 (2005.10.05)
AX@1500-1700 Q1 Attend lecture in BA1230 from 2005.10.04 (2005.10.04)
AX@0900-1100 Study for MIE1407F from 2005.09.29 (2005.09.29)
AX@1100-1200 Attend lecture in MP134 from 2005.09.29 (2005.09.29)
AX@1500-1800 Go to engg psych lab in RS303 from 2005.09.28 (2005.09.28)
AX@1500-1700 Attend lecture in BA1230 from 2005.09.27 (2005.09.27)
AX@1200 Review lab report 1 : E-Mail from alex jim (2005.09.23)
AX@0700-0800 Explain experiment bias : E-Mail from alex jim (2005.09.22)
AX@0800-0900 Explain experiment weakness (2005.09.22)
AX@0900-1000 Write lab conclusion (2005.09.22)
AX@1100-1200 Attend lecture in MP134 from 2005.09.22 (2005.09.22)
AC@0900-1100 Study for MIE1407F from 2005.09.22 (2005.09.22)
AXSign up for hospital fieldtrip (2005.09.21)
AXPrepare resume for hospital fieldtrip {{Deadline: 2005.09.21}} (2005.09.21)
AC@1500-1800 Go to engg psych lab in RS303 from 2005.09.21 (2005.09.21)
AX@1500-1700 Attend lecture in BA1230 from 2005.09.20 (2005.09.20)
AX@2000-2300 Work on lab report (2005.09.18)
AX@0900-1100 Study for MIE1407F from 2005.09.15 (2005.09.15)
AX@1100-1200 Attend lecture in MP134 from 2005.09.15 (2005.09.15)
AX@1500-1800 Attend engg psych lab (2005.09.14)
AX@1500-1700 Go to MIE1407F in BA1230 from 2005.09.13 (2005.09.13)
AX@2200-2230 Read chapter 2, first pass (2005.09.12)
AX@1100 Attend class at MP134 (2005.09.08)
AXAdd CCNet/20059/mie448h1f/ to my daily bookmarks (2005.09.08)
AXGet textbook: Engineering Psychology and Human Performance (2005.09.08)
AXE-mail name and student number to Paul Milgram regarding ECFPC account (2005.09.08)

5. Looking forward to Monday: 16:13

Categories: 2005.12.02#2 -- Permalink
I am _so_ looking forward to being finished with my enggpsych paper. Grumble, grumble. Humbling and annoying to think I've only written ~ 4160 words so far. I have no idea how the people who wrote the A+ papers he posted managed to write that much. C'mon, 26 pages, single-spaced? Okay, well, the 26 page single-spaced one was a bit fluffy in terms of language... But still. <mumble>

I am relatively happy with it now. There are a couple of paragraphs here and there that I still might want to write, but overall I won't be too annoyed with myself if I handed this in. It's been a good way to review the textbook, too...

4. Justin Hollands

Categories: None -- Permalink
Human computer interaction group, defense r&D canada

Signal Detection Theory

Toronto lab focused on human. human behavior, physiology, human-systems integration, which includes HF but also related topics like personnel, training, manpower, like issues.

emphasis on take-home message:

obtaining separate measures of an observer's sensitiivyt and bias

Bias affects criteria.

TAKEHOME: SDT: Separate measures of sensitive and bias. Refined our measure of an observer's performance

3. isosensitiity

Categories: None -- Permalink

Announcements. Julie Stevensons. Engineering Career Office for both undegrad and grad. This career office is soley for the engineers. Sanford Fleming B670. Next week: One of North America's largest career fairs happening next week. Wednesday and Thursday. Register before you attend. www.career.school.ca - click on the career fair and then register. It will be in University College, 11 to 3. If you want company profiles, they're also available on the website. New grads: recruitments are mainly now: September to December. Summer: January to March.

Graduate school announcement. One of your options is to go to grad school next year (or in the future), and if you're contemplating that, then now is the time to apply for the important government scholarships: NSIRC (national, very high-level scholarship, A- average to apply, no visa students), OGS (very high-level, also pays well, and deadline for applying in September-October.) My experience has been that guys like you are wondering "What's he talking about? I just want to get out of here," so most of you want to get out there and work, but by the time summer wanders around, then they have second thoughts. If you get it, you can always turn it down. Two weeks Friday.

Another important announcement. I am flying off to China on Saturday. Justin Hollands will be doing the lecture on the 20th. Yesterday I sent out my first e-mail to the list, also. No lab next week. On Thursday, we have Dr. Jacob Lai, who is the chief of anesthesiology at the Toronto Western Hospital.

I have prepared the background for the field trip, a three-page document which I will put on the website this afternoon. It says where we're going, the objective, background preparation. Some of this is important. The first is the lecture on the 22nd. Very important: you can't just walk into the hospital. Security. They have to know who each and every one of you is. Everyone who's participating in this trip has to submit a resume. You're not looking for a resume, they just want to know who you are. Just take whatever resume you have... One page, very brief, so that they know who you are. It's written on this fact sheet. You can send me and Zhang Hai your resume in a PDF. Once we get a whole bunch, we will forward it to them. They'll have to process this, so we have to get them in early. Please do that right away. I'm curious to know if anyone's intending not to attend.

E-mail from Engineering Publicity office. There'll probably be an article in the engineering newsletter and the university newsletter. I hope this works. =)

Priorities - A: high, B: medium, C: low; Status - _: unfinished, X: finished, C: cancelled, P: pending, o: in progress, >: delegated. Covey quadrants - Q1 & Q3: urgent, Q1 & Q2: important
A_Prepare resume for hospital fieldtrip (2005.09.15)

Background. Most of you know nothing about hospital anesthesiology, which is okay. Read the background pamphlet and that'll give you some idea of what's going on.

Logistics. I'm not going to go over that now. Where the hospital is, which entrance you go in, where you meet. When you get there, you're going to have to change your clothes. There's a change room there and they'll give you scrubs. This is important: don't bring valuables. If you're coming from the university, leave your stuff in your locker, or something like that.

Scheduling. We're going to start doing this. We've mode it to Oct 33 onards. We're going to put a Calendar on the website and you can sign in. 7:30 in the morning, 9:30 in the morning, and 1 in the afternoon. You should be able to find some opening in your calendar for this. I would count on two hours. Ten-minute walk, hour or hour and a half there, another walk balk. 7:30 slot best because you can see everything from the beginning. If you're lucky, you'll get in at the beginning of the operation, and you can see the beginning, middle and end. If you arrive in the middle, I suggest staying until the beginning of the next operation. Feel free to stay longer. Grad students need to do a paper, so they may want to do a paper related to this. Take a look at the pamphlets before the lecture. (By the way, Dr. Lai has two kids in our department.)

Extra lecture will be held on Wednesday, the 28th. so that we don't fall behind. RS208.

Priorities - A: high, B: medium, C: low; Status - _: unfinished, X: finished, C: cancelled, P: pending, o: in progress, >: delegated. Covey quadrants - Q1 & Q3: urgent, Q1 & Q2: important
AX@1500 Attend extra lecture @RS208 (2005.09.28)

I'll be away on Oct 4, so we'll have another guest lecturer. He's fantastic. He was my thesis supervisor. He'll be talking about human factors, human error, medically related error. He did this for me last year when I was away at the conference, and one student wrote that it was the most educational experience in all his history, and that didn't make me look too good. <laugh>

Guest lectures: Full two hours.

Extra lecture: 2 hours. 3 - 5.


How can we make models of some kinds of behavior which are quantitative and which allow us to say how much information is being processed and how rapidly, so that we can design systems where the throughput we're demanding is not greater than the possible throughput?

We're modeling humans as a limited capacity noisy information channel.

I went through the stovetop stuff as a simple example of input and output. What I'm talking about now is giving you a sense of basic ideas in information theory

Information theory

Something happens where there are a number of possible outcomes. I'm talking and the next word that I say has a set of possible outcomes. The fact that there are a number of different possiblities means there's information. Each outcome has a probability. The amount of information in that source can be quantified using that formula. (DIAGRAM: Definition of "Information").

- number of possible events - distributional constraints - sequential constraints

What do we do with all of this?

Keeping in mind that we're looking at an information channel with information coming in, information being transmitted, and information coming out, there are two ways in which all of this could screw up. Noise. Output information would not be identical with the input information. Another way: lose information at the source.

DIAGRAM: Venn Diagram Model of Information Transmission.

We've got:

- stimulus information - response information

Red area shows information in both. The larger the area, the better everything is. More of what's in the input is getting through the output. Anything that's in the output that's not in the red area is in this area, and that represents the noise. Anything that's in the blue circle that's not in the red area represents information that's lost. We can look at the total amount of the information by looking at the union.

H(X/Y) - H of X given Y - equivocation H(Y/X) - H of Y given X - noise

I don't like when you use the word "memorizing", because it's almost nothing.

Only thing you'll probably memorize is the weighted sum of the log of probabilities.

You don't need to memorize anything because the formulae come directly from the diagram.

Matrix of stimulus and response. Once they get off the diagonal, then they're making errors.

DIAGRAM: Contingency table representation of experimental data

P(x_i) = n_i / N p(x_i, y_j) = n_ij / N

Interesting: Objective probability. Some of the time, you may have to think about the probabilities in your source.

If the dispersion is large, then the intersection is small.

IF YOU ARE CONSISTENT, IT'S POSSIBLE TO COMMUNICATE. Weakness of approach: Mathematically, we could have everyone saying Yes instead of No and No instead of Yes, and have everyone live normally. In human terms, we would have to adapt to that.

DIAGRAM: Fundaments of models of the human operator as a limited information channel

Uses a Mac.

2. Human information processing and information theory

Categories: None -- Permalink
(Wickens & Hollands, Ch. 2: 44-61; Ch. 9: 339-350; Ch. 10: 387-389)
Review: What is this course's purpose?

- Engineering psychology, applied to design

- Examples of high-tech work where there's a lot of information

processing involved, high-stake

Anesthesiology example

- Humans in a control loop. Anesthetist controls the system.

- Very complex system controlled there. Bringing live human being to

the threshold of death.

- More complicated now. Quality control. Quality of outcome. As few

side effects as possible.

Field trip

- Will be able to ask questions - Going to Western.

- Teaching hospital, so you have residents doing all the work. chief

surgeon, etc. watching over the residents. You can talk to them.

- Get some sense of the interaction between these guys. Training

expertise is also an issue that's part of this course. first-hand example of one expert training another expert.

Human information processing

Example of person driving a car

TAKEAWAY MESSAGE OF COURSE: understanding some basic principles in human-information processing and how those relate to the design of human-machine systems

Aspects of information processing (interactive, from class)

- sensory input: visual and audio - trying to sort out what's important and what's not. prioritize. attention - driver's doing some prediction - comprehension and planning (ex: black ice ahead, must think of detours) - memory processing going on - decision-making: different options of what you're going to do - motor control. manual control is the result of decisions. needs information processing as well.

Stimulus, Cognition, Response

- Simplest model - I make a noise, nobody's expecting it, you get startled, and jump. - Information comes in as external stimuli - Cognition to figure out what it is - Response - Really basic model. Tells us what we already know. Not much new insight.

What are the seven elements of a generic human-machine system?

- Human: Perceived information - Human: Information processing and decision making - Human: Motor responses for control activation - Human-machine interfacez - Machine: Control devices - Machine: System - Machine: displays

The system could be anything: hardware, software, peopleware.

The human's effectors control that system or influence that system through different mechanisms. Steering wheel, gas pedal, joystick, keyboard, whatever. You can't control the system unless you have some kind of feedback. Displays are not just visual; they can also be auditory or other stuff. For example, a flight simulator, a motion simulator, etc.

This is a generic model. It doesn't give us a lot of insight.

Figure 1.4: A model of human information processing strategies
One more representation. Figure 1.3 in your book.

We'll be seeing this model over and over. This is what drives the course.

Model: representation. There are toher ways of representing such a complex system. this is the one that's proposed by the authors of your textbook. The authors of your textbook are Chris Wickens and Justin Hollands. Chris Wickens just retired this year. Aviation lab, human factors department, Illinois. Huge HF factory down there, very very well known. He lives in Toronto, he works at defense R&D. Guest lecturer next Tuesday!! We really have to admt that this is Chris Wickens' model because he had it before Justin coauthored.

Added SCR framework there just to show that even if there are different models, they have something to do with each other. In general, when we look at the left side, we're looking at sensing. center: cognition. r: response.

psychological model, not an engineering kind of model. If any of you have taken control engineering, then there's a different kind of a focus: mathematical, formulae, equations. here, the focus is not on quantitative representations, but functional representations. human information processing functions. you can't make a plot of the different control groups. nevertheless, psychological model, recognizes most basic concept of control engineering: feedback.

wwen we talk about inputs, we're talking about sensing: receiving information. what do we do when we sense something? you are probably sensing your elbow on your desk right now, but it doesn't mean anything to you... if somebody were to be touching you from behind, you would first detect it, and then you decide whether or not it means something: identifying. coding and classifying. different kinds of input tasks.

cognition: thinking about. making decisions, manipulating, solving problems, planning ahead, things like that.

effect something, influence environment...

categories of errors you could be making. NOTE: Diagram. Output: Chaining: sequential chaining effects?

Model itself. STSS: Short-term sensory storage. BIG DEAL. One insight is that we've got many arrows going from going into processing. We are continually assaulted by many kinds of sensory input, but it's only the important ones that make their way fthrough our IP system. we have stuff coming in whether we like it or not. you could be receiving information without perceiving it. All sorts of stuff coming in whether you like it or not. Cognitive work. Boundaries are vague as you get toward the center. Perceive information. You identify it's there, you start attributing meaning to it.

Working memory, cognition: CPU. This is where all the work goes on. Arrows down to one.

Attentional resources. This model is telling us that hovering abve all of this SCR is attention. It's your attentional capabilities which determine what you're going to attend to (obvious) because you're limited. You can't do many many things at onece. tHat's the primary theme of the course. We are limited by the amount of information and the rate at which we can process information.

Arrow from attention rsources to our sensory processes as we select whilat we perceive. over here (STSS( we have no arrows. an interesting message from this is that there are information processes that are going on athat are pre-attentive. You take them in even if you're not consciously trying to. You can't process information unless you understand what it means. You have to have long-term memory. If you're reading a book in Armenian, but if the words don't mean anything to you, you can't process the book.

Long-term memory: storage. You can't do anything with your CPU unless you have RAM: working memory. Computation, translations, comparisons, etc.

And your attentional resources affect whether you're going to attend to something. Which of the many things you're going to attend to, whether you're going to attend to, how you put your priorities in your working memory, it's going to influence your long-term memory, deciding what to put in your long-term memory. long-deciding what to store for later. You're going to respond. Selectingi: I(A lot through eyes and head motion for communication). Have to execute that. So if I start stumbling over words, then having problems in haining and inserting and misordering in terms of response execution. Declining in performance in total information processd.

I'll be bringing this back during the course to show whow what we're talking about relates to the model.

Chatpure 1: Stuff about history not covered because most people have heard it before, but should read anyway.

Modeling the human as an information processing channel: information theoretic models.

We're going to look at more global models. The first part of the course is more quantiative than the first part of the course. We're going to look at actual quantitative modeling, all of which is concentrated in chapter 2. Chapter 2 talkes about information theory and signal detection theroy. comm engg should have an easier time with this. the rest haven't, so i spend an inordinate amount of time trying to explain it slowly.

information theory models. can we quantitatively model or characterize, typify human information processing. Yes, for some tasks. Let's look at what that means.

First: human as information processing channel / system. Quantifying. hat do we mean about quantifying informaiton.

Stoveburner example

Figure 10-3. Control-burner arrangements of a simulated stove used in experiments. Classic problem: not really a problem any more, but the kind of illustration that you'll find in an introductory text book or a historical book. The stovetop problem. The problem of designing a stovetop is that typically, you've got The problem then is, how can you look at the problem of : making a choice from this input set of burners, a two-dimensional array. communicate your deccision correctly through one dimensional array of dials. Picture shows there are different ways to do it. You would almost never expect of someone using the left-hand dials to turn on the right-hand burners. But you'd expect a lot of errors in II, ABCCD. Best one is I because you've added proximity compatibilty. You've added extra information so that the outer dial corresponds to something that's more outer, and so on.Trade-off: wider stove or smaller urners. Research: You can propose a design, you can talk about it, but how do you prove that your design is better than another design? You run a test.

Actual simulation or contrived test. Does not give us a lot of information about which design is better and in what respect is it better.

Information theory

4.301 information theory handout

The tools that we're using are from information theory (part of communicational engineering, part of electrical engineering). All of this comes from the late 1940s or mid 1940s. World War II. People were trying very hard to kill each other. One of the key technologies that developed during the war was radar. Arguably, that's what's won the sign on our side. Sending out signals and receiving signals, and interpreting from what you got back what was out there. Radar, telephone, all of these things being developed back then. Information theory: set of tools for analyzing, modeling, designing, developing communication channels.

shannon and weaver main names. Claud Shannon ideveloped information theory. The essense of this is that there's an information channel. Information travels through it. It has to go in. It has to be encoded, has to travel through it, decoded, and interpreted at the destination. Model applies to telephone. (example: trace through phone line). doesn't sound exactly like the person is there. not a perfect transmission. noise: anything that deviates us from perfection.

Human information processing models have done (very fashionable in 60s) take theory of information from 40sand 50s. stuff they use to model modems, radar, etc. corresponds to what humans are doing.Noise: We make errors, we don't understand everything that's given to us.

Using that approach:

Limited capacity noisy communication system

The basic principle of modeling a human operator: the human operator is a LIMITED CAPACITY NOISY COMMUNICATION SYSTEM.Limit to amount of information we can absorb. Limit to rate of absorbing information. We are noisy because there are imperfections in our ability to perceive visual, auditory information. We are a communication systems is the sense of having different kinds of input and we're trying to match the outputs to the inputs. We're trying to influence our environment in accordance to the information coming in. Which is what we do when we're faced with the stovetop burners. Input: which one. Output: action.

Types of information tasks
If we think in terms of the communication channel. There are different kinds of tasks. Information reduction, transmission, elaboration.
  • Reduction: Internet searching, summarizing. for example, what you're doing what now.
  • Trasmission: If you're copying down everything I'm showing you, that's a transmission task. If you're transcribing written text into typed text...
  • Elaboration: Example: 3000 word essay on my shirt. Efforts to show the same kind of information theory modeling for just the physical information processing in our body. The message here is the size. At the sense organs, you get 1 gigabit/s. Nerve junctions: 3 megabits. Conscious awareness: 16 bits. Lasting impression: 0.7 bits / second. _ Check out these numbers to see if they make sense What's important: less than one bit per second. goldfish 7 seconds, every time it swims around the bowl, it's seeing things for the first time
Quantifying information transmission
DIAGRAM: Figure 9.5: Possible arrangements of stove burner controls. How do we quantify performance on this kind of a task, on this particular task? Why/how can this be modelled as an information transmission task? For example: | A | B | C | D
A 4 1 . .
B . 5 . .
C . . 5 .
D . . . 5
Measures how good or bad information transmission is. Look at what people could be doing and what they do do. It's possible to quantify their performance. If you think about that stovetop example. If we look at the number of burners. If you have a hotplate--single burner--you would never make any mistakes, right? My question is,
Factors that affect amount of information being presented to system
What are the factors that influence the AMOUNT of information being presented to a system?
  • The number of elements
  • Spatial constraints: layout, color and formatting
  • Distributional constraints: stovetop problem with multiple pots. As soon as you put constraints... Entropy. If everything can achieve the state of maximum entropy, maximum dispersion, then there's a lot of information out there. As soon as you put constraints... DIAGRAM: Definition of information Coin flip: Event. There is an outcome.
  • Information content
    Information content.
  • number of possibleevents
  • probabilities of events (i.e. distributional constraints): distributional constraints
  • sequential constraints of events, etc: redundancy
  • Amount of information

    more outcomes: more information. Amount of information in the task. What do I mean when I talk about transmitting it? Amount of information : average amount of information. If I tell you I'm thinking about a person, there are 6 billion, but if I'm thinking of politicians, then you can drill down. Picture: Average minimum number of questions that you can ask to get an outcome. =) The stuff I teach! I've got a number in my head between 1 and 8. Approach: linear search? If you do that, 7 guesses. Intelligent approach: binary search. Average minimum number of guesses will be three. (Is that average or maximum number of guesses?) H = the sum of (probability of each outcome multiplied by the log-base-2 of 1 / that probability) \bar{H} = +\sigma{i=1}{N}{p\sub{i}log\sub{2}\frac{1}{p\sub{i}}} If we have equally likely events, then it's pretty straightforward. lg N. Every time we double the number of outcomes, we add one bit of information. If we look further than that... Let's look at the influence of distributional constraints.

    Distributional constraints
    INFORMATION CONTENT: INFLUENCE OF DISTRIBUTIONAL CONSTRAINTS. Example: Surprisal function. Function(1/p_i), H_i = log 1/p_i. Low probability: surprised, high information, and vice versa. What do you think the information content of those two outcomes is? We're looking at AVERAGE nformation. Information content: weighted average by probabilities. Maximum entropy: evenly distributed. Less information otherwise. This is an example of a distributional constraint. Conditional probabilities. Try it with series consisting of A (P(A) = 0.9) and B (p(B) = 0.1).
    Redundancy: Redundancy = ((H_max - H_actual) / H_max) x 100 Example: warning systems. You want high redundancy. You need very highly redundant signalling systems. But if I want to finish this course on time, I need to lower my redunancy and increase the rate of information. I don't agree with the definition because you're subtracting logs and that's already really a division. Anyway, the general message doesn't change.
    Sequential constraints
    Here's another example. P(A) = P(B) = 0.5. but P(A/A) = P(B/B0 = 0.9. Probability of A after choosing an A is 0.9. X Read about sequential constraints Ahh. It's 0.1 because that's the probability for new information? I need to read about this more. If what I'm saying now affects what I'm going to say in the next... sentence... There's a sequential constraint that makes that word almost insignificant. Sequential constraints. Remember there were three kinds of constraints? Spatial constraints. Sequential constraints. Let's think of the weather in Arizona. If we sample weather reports randomly over a year, then if you have 50 rainy days and 300 sunny days, then the probability of rain is 1/7th and the probability of sun is 6/7th. We can figure out the average amount of information like before.. Let's say that rainy days tend to group together. Is that going to decrease or increase the amount of information. It's going to decrease because there's less to be learned. Here's an example of sequential constraints.
    Effect of redundancy
    EXAMPLE: scrmabled text This is an example of sequential constraints. The context reduces the number of possibilities these jumbled words couldbe. DIAGRAM: APPROXIMATIONS TO ENGLISH: Letter approximations Word approximations _ Try letter and word approximations Example: Martians looking at English. For example: 27 letters (26 + space). Equiprobable: 4.75 bits per symbol. Use the actual frequencies: 4.03 bits per symbol. second order markov approximation Redundancy: 68% Actual: H = 1.5 bits / letter

    1. MIE1407F: Engineering Psychology and Human Performance: 13:59

    Categories: 2005.09.08:2 -- Permalink
    Paul Milgram talked about the course mechanics and gave us a whirlwind tour of incredibly complex systems that are out there. I've heard this course is a lot of work, but it'll teach me a lot. I'm going to need to work on statistics and experiment design, as I don't have much background in that.
    random stuff

    absolute judgment


    multidimensional absolute judgment

    correlated dimensions

    your only task as a designer is to decide where to plunk this criterion. Andonce you plunk the criterion down, everything flows out of it. The performance is perfectly predictable. The setting of the criterion is done through computing called the optimal beta. and that is done according to this formula. I don't think he derived the formula, but it's not really important. It trades off semuthng thot,s rufluctdng feir sibkutdvu volius. the setting of this criterion has nothenig to do with the strength of the signal. noise, signal of a certain strength. i'm fre to set a criterion. indepe the setting of this criterion reflects two aspects of my subjective mental world one is my estimate of the relative probabilities of the signal being there vs the signal not being there. if i think that the signal is very likely, then p(s) is large,...

    Application of SDT in Engineering Psychology

    How do you study for this course? The kinds of questions I ask, I try not to make this into a course in memorizing. In other words, you don't need to memorize all the important points. I stay up late at night trying to think of good questions that test your ability to understand the basic principles. It's a lot harder to mark. That's so I can get away from the idea of "list 4 points here", etc.

    First part of the course more quantitative than the second part. Listen to what I say in the lectures because I emphasize what's important. Read the book. Th edifference is left up to your judgment. You should be reading the book, you should be getting an idea of the main points, you should be reading the examples there that I don't cover. You don't have to remember all of the deep examples that they give; they give a lot of references. That's a level of detail which is memory.

    The approach to determining the optimal criterion differs. If you were a designer, you'd start with the objective parameters of the problem, and you would come ppwwith a design, and your design would predict performance.

    The slope of the RT:HT line tells you how your subject performs. You can sy something about their basic reaction time and the rate at which they make decisions.

    Analyst: estimates. Ahhhh... model handout shows it going the other way around. Observed frequencies -> descriptive model -> estimates of subjective parameters. Estimates what the operator's beta was. Contrast: normative model tells you what the optimal beta is.

    Example of the operator inspecting cloth. Insights: 1 - 2> adjusting beta, 2 - 3: adjusting sensitivity.

    1. P_hat(H) = 0.6

    P_hat(FA) = 0.05

    B_opt = (/ 0.80 0.20) P(A) = (P(H) + (1 - P(FA))) / 2

    - 1. Draw a signal-noise diagram.

    - 2. Label the evidence variable. Represents whatever it is that is being

    observed, no matter what. Draw the noise and signal curves. Label the curves P(X_N) and P(X_S)

    - 3. Select an X_c so that the area under the signal curve, to the right, is

    roughly P(H), and the area under the noise curve, to the right of X_c, is roughly P(FA).

    - 4. Estimate d' by calculating the distances between the two means in

    terms of standard deviation. You can calculate this based on the area underneath the curve and a table that matches area under the curve to zed. d' = 1.645 + 0.255 = 1.90

    - 5. Use ordinates of the standard curve table to find Y position

    given area at a certain zed. Answer: B_hat = a / b = 0.387 /

    1. 103 = 3.76

    - 6. Calculate B_opt.

    - 7. Thnk about the data you don't have. Document your assumption that

    everything is neutral.

    Powerful things: model whatever people are doing when they take in information without trying to figure out how they make their observations.

    How do we determine d'? The units of d' are units of standard deviations. We assume that the variance of both of the distributions are the same in this simple model because the signal is discrete. What values of standard deviations correspond to these areas?


    Estimate of actual bata = P(X_c at S) / P(X_c at N) = a / b.

    Optimal beta = P(0.80) / P(0.20)


    2. d' = 1.96, b = 1.81 3. d' = 2.32, b = 3.09