The most memorable conference talks I’ve ever heard were given by John Krumm. John is a researcher at Microsoft Research, and he works on measuring the location of people and utilizing this information in a way beneficial to people. There are multiple things that make his talks great, but one thing that certainly makes them memorable is that John uses humor very effectively. E.g. in one talk John made fun of sponsorship deals by including fake sponsor messages on slides.

In July 2007, John and three of his Microsoft Research colleagues, Patrick Baudisch, Mary Czerwinski and Richard Szelinski organized a session at Microsoft Research on how to give a great presentation for a job interview or a conference presentation. The video of this session is available here. Here are the pictures of the four speakers:

Here are some of the ideas from the session that I really liked. All of the presenters (with the exception of Patrick Baudisch) underscored the importance of practice. I couldn’t agree more. All of them also agreed that visuals (pictures and videos) are an excellent way to create a good presentation.

John Krumm also pointed out that the presentation has to bring people up to speed on what you’re doing, without assuming that they already know the intricate details of your area. Most people don’t and they will be lost without your help. Patrick Baudisch recommends taking pictures of your work from beginning to end (e.g. take pictures of messy prototypes). This will provide you with a large collection of photos to use in an exciting presentation.

Rick made an excellent point about which part of our work has the largest impact - it is not the code we write, it is the presentation we make, since that’s what will touch the largest number of people (say at a conference). So, it makes sense to put a lot of effort into the presentation. Mary gave interesting suggestions about how to deal with rude questions. Above all, she recommends staying humble. I can remember at least one occasion where I really could have used this advice.

Take a look at the video (it’s about 75 minutes long). It’s well worth watching, probably multiple times. You can see that I’ve learned something from it: I included the photos of the speakers above. Visuals matter!

Andrew Kun

Today, Jan 29, 2008, I was a participant in a psychology experiment which is trying to understand working memory. The experiment consisted of me sitting in front of a computer screen with head phones while different images with sounds appeared. Then one of the images and sounds would appear again and my jobs determine if anything was different. I experiment consisted of about 10 different set ups. Each setup had different combinations of images and sounds ( i.e. color balls and a tones, color balls and animal noises, Grey balls and tones, pictures of familiar things with tones, etc.)

Below are some excerpts from the debeifing.

“Past work has shown that visual modality can quickly ‘bind’ or associate different features with 3-4 objects, giving us the ability to detect changes int those objects after a short delay( Vogel, Woodman, and Luck 2001).”

“Data from this experiment will indicated whether associating features from different modalities (such as color and sound) takes more effort than associating features within one modality (such as color and position). Also, following on earlier research by Ceci and collegeues (1994), we will asses whether more recognizable real world objects such as bugs behave differently in working memory than abstract ones such as domes.”

~Jon Oppelaar

While skiing with my friends (Zeljko and Andras), I’ve made a statement: “Skiing is an investment for me! Lessons that improve my skiing can be transfered to other areas, for example, programming…”. Surprisingly, my friends were paying close attention to my words and immidietly requested a support for my statements (let’s not point finger at Zeljko). Luckily for me we were already off the chair-lift, and hence there was not time to continue this conversations. After thinking a little bit I found a good illustration (at least I think, it’s good) that supports my statement (after all, I only need a single example that proves my point). So here it is.

We were practicing formation skiing (that’s how we called it), meaning that three of us tried to move as a single entity while skiing down the slope. What I found out is that the major problem with this exersise for us was that attention that we payed to the formation-mates was the attention we did not pay to our own skis… which lead to close encounters with the ground (it did for me). I have found that staying close together made it easier to pay attention to my skis. I explain it by the fact that I did not have to move my head as much, hence the time that otherwise would be spent turning my head around and locating the mates was directed to my skis. This lesson can be abstracted as follows: the fewer or faster the attention shifts that are required to observe parts of an entity, the easier it is it maintain this entity.

One illustration of this lesson in programming is “Declare variables as localy as possible”. In fact, Herb Sutter and Andrei Alexandrescu mention this rule in their book C++ Coding Standards: 101 Rules, Guidelines, and Best Practices (rule 18).

Now the question is “Am I really stretching it?”

Alexander Shyrokov

Last week I started teaching a Speech Signal Processing course. This promises to be a fun course for me, since it mixes DSP and human speech production, both topics I’m very interested in. I keep track of links relevant to this course here. The latest link I’ve added to this list is a link to SPAN, the Speech Production and Articulation Knowledge Group at USC. They have fascinating movies showing human speech production as it happens. The movies were made using USC’s high-speed MRI technique. There are several movies on the SPAN home page, and for more click on Videos.

A couple of notes on viewing the SPAN videos. First, to see a video of a particular sound being produced, select the sound and then select the speaker (American or Indian English speaker). Second, on one computer I had problems viewing the videos from Firefox, however they worked fine from IE. So if you’re having problems with viewing try switching browsers (or let me know if you have a better explanation as to why the Quicktime plugin works OK on the home page but not the videos page).

Andrew Kun

An excellent AP article by Todd Lewan discusses how RFID technology can be used to spam the living lights out of you, as well as how to track you and in general violate your privacy.

While I believe RFID technology is extremely useful, it must be regulated in order to avoid abuses. We cannot simply take the word of corporations and/or government that they’ll take good care of our personal data. In fact, the article shows that corporations are not concerned with our privacy, and primarily worry about it when confronted with consumer backlash and the prospect of regulation.

Without regulation things will escalate out of control, and we’ll all feel like the tourists in places like Cancun. Cancun is a beautiful place, with a fantastic waterfront, ancient ruins nearby, and given a chance I will go there again. However, tourists are constantly live-spammed: from the moment you leave the arrival gate at the airport people want to sell you phone cards and blankets, taxis honk to offer a ride, hotel representatives intercept you on the street and offer drinks and try to make you switch accommodations mid-vacation, etc. The pressure is constant. You quickly adapt and start ignoring the offers, but they take some of your freedom away - e.g. you can’t just let your gaze wander for fear of making eye contact with a street vendor. In a world where RFID is abused for commercial gain, and other purposes, we’ll also have to develop evasive maneuvers and this too will take some of our freedom away.

Andrew Kun

Here’s a great article on how computer science education is being damaged by catering to students who value fun over hard work. Note the publication where this appeared: the STSS is a US Air Force organization.  Why would the USAF care? Because poor CS education will have an effect on US national security, since we won’t have the manpower to design new weapons systems. Of course, poor CS education will have a negative effect on the whole US economy.

Andrew Kun

Between our new eye tracker and not so new dirving simulator the total number of computers that we use during our experiments goes to 8. We are analyzing data logs from at least 3 different computers. So here we go:

Situation: There are three computers connected to a local network, let’s call them m1, p1, and p2. m1 and p1 are running Windows XP (SP2), p2 is running Linux FedoreCore 3. Programs running on these computers provide us with log files. We are interested in the relationship between timestamps of log entries in these files.

Problem: Time stamps in log files are not synchronized.

Previous solution: Ad-hoc methods with low accuracy, for example, sending a message with a time stamp from one computer to be logged on another.

New solution: Use ntp (Network Time Protocol) to synchronize all the computers with accuracy better than 500 micro seconds. Ntp implementation is available for both Windows and Linux. In our case m1 has two network cards and is connected to the local network and to the Internet. This allows us to synchronize m1 with public time servers, though we are interested in time synchronization between the local computers.

The following is the procedure one can follow to synchronize thier computers if they have similar situation:

For m1 (Windows, connected to the Internet and local network) and p1 (Windows, local network):

1. Download binaries for ntp server.
2. Unzip to some directory. I used C:\Program files\ntp-4.2.0.
3. Install ntp server as service by calling Instsrv.exe with the full path to ntpd.exe. In my case it was: Instsrv.exe “C:\Program Files\ntp-4.2.0\ntpd.exe”.
4. Make sure that your firewall allows ntpd to connect to the Internet and that port 123 UDP can be accessed by other computers.
5. Create configuration file C:\Windows\system32\drivers\etc\ntp.conf (see bellow).
6. Go to Control Panel->Date and Time Properties->Internet Time tab and uncheck Automatically synchronize with an Internet time server.
7. Go to Control Panel->Administrative tools->Services applet and
   1. Disable Windows Time service.
   2. Set Network Time Protocol to start automatically.
   3. Start Network Time Protocol service.
8. Check the status of ntp server (see below).

For p2 (Linux, connected to the local network):

1. Check if you already have ntp daemon installed. In our case it was /etc/init.d/ntpd. Chances are that you have it, otherwise you need to install it (let your package manger or google help you).
2. Edit configuration file /etc/ntp.conf (sudo vi /etc/ntp.conf) (see below).
3. Restart the daemon: sudo /etc/init.d/ntpd restart. You should get “Starting ntpd: [OK]”.
4. Make sure that the daemon starts automatically.
   1. Look at your /etc/inittab to see what init level your system starts in. Line id:X:initdefault: says that the level is X. It was 3 for us. If it’s different for you replace 3 in rc3.d to whatever your number is.
   2. Create (or make sure it’s already there) a symbolic link to the ntpd server in /etc/rc3.d by running ln -s /etc/init.d/ntpd /etc/rc3.d/S96ntpd.
5. Check the status of ntp server (see below).

Configuration files:

1. For m1:
# iburst option is used to make sure that time is synchronized as quickly as possible
server some.public.server.1 iburst
server some.public.server.X iburst
driftfile /WINDOWS/system32/drivers/etc/ntp.drift
peer p1
peer p2
2. For p1:
server m1 iburst
driftfile /WINDOWS/system32/drivers/etc/ntp.drift
peer p1
peer p2
3. For p2:
server m1 iburst
driftfile /var/lib/ntp/drift
peer p1
peer p2

To check the status of your ntp server/client execute ntpq -c peers after a few minutes. (Not that ntpq could be located /usr/sbin directory in Linux) The result should be similar to the following output if you run it on p1 (time values are in milliseconds for 4.2.0 release):

remote refid st t when poll reach delay offset jitter
==============================================================================
*m1 192.168.10.50 2 u 26 1024 377 17.505 -30.113 6.225
p1 .STEP. 16 u - 1024 0 0.000 0.000 4000.00
+p2 192.168.10.52 3 u 86 1024 207 8.254 3.189 1.473
Now you should have your m1, p1, and p2 synchronized. I used Getting Started tutorial to get ntp running on Windows. All credit for Linux side goes to Andras Fekete.

Alexander Shyrokov

Last week we received our newest research tool, a state-of-the-art eye-tracker, called faceLAB, product of Seeing Machines Inc. It uses a pair of stereo cameras to track the direction of the subject’s gaze. Not only the eyes are tracked, but head position and orientation too. Based on the collected data, it calculates fatigue measures (for example, PERCLOS) . It can be said that faceLAB is a head, gaze and fatigue analysis tool.

The stereo cameras came together with a bunch of hardware needed for faceLAB to run: a laptop, lenses, camera stands, video overlaying devices, calibration (chess) board, etc.

Contents of the faceLAB package.

The software is very nicely designed. I worked with stereo imaging before, so I can say that the mathematic calculations behind the the user interface are very complex and fascinating. For example, the system calculates the position and orientation of the cameras in 3D based only on showing the cameras the checkered calibration board.

The user interface is divided into four parts. The upper right window shows the views of the cameras and the extracted gaze direction, eye closure, head anchor points, etc. In the lower right quarter of faceLAB a virtual 3D world is shown.

In the virtual space, a model of the driving simulator can be created, with special areas, that will represent certain devices (GPS navigation device, radio, P54 screen, etc). The application can then log when the drivers are diverting their gaze from the road to the in-car devices.

After initial testing on the test stand, the cameras were mounted into our driving simulator onto the dashboard.

Next, tests will be performed to find the optimal position of the cameras. Pilot runs will be conducted to see how the logging of data is done and to verify the setup.

The eye-tracker opens a whole new field of research possibilities. Among other things, now we can track the flow of visual attention of drivers, which is a very important measure for safety assessment.

If you have any questions or comments about the eye-tracker, feel free to post them below. Thank you for your attention,

Oszkar Palinko

On the last day of the past semester (12.21.07), Alex Shyrokov and I paid a visit to Northeastern University’s Intelligent Human-Machine Systems Lab. This was a follow-up meeting to the contact we established with fellow researchers last fall on the NECHFES Student Conference. We were very interested in finding out about the sensored SmartWheel that they developed. It has many similarities with our sensored steering wheel. Our hosts were Hua Cai and Hongjie Leng, who kindly demonstrated us their system.

Hua Cai (NEU), Hongjie Leng (NEU), Oskar Palinko (UNH)

As it can be seen on the above image, the simulator consists of a driving seat, steering wheel, pedals and a computer screen. Both the software and hardware were developed at Northeastern. The university has a larger, higher fidelity simulator too in a neighboring lab, which was inaccessible at the time. The two simulators can be connected, so two drivers can interact in the same scenario.

The steering wheel has a number of sensors on its surface: blood volume pulse (infra-red), skin conductance, skin temperature and gripping force sensors. To be able to measure these variables, it is necessary that the drivers hold the steering wheel at exact positions, i.e. to align the fingertips with the sensor positions. The grip must not be too hard, otherwise the data gets noisy.

The multi-modal system even incorporates sensors in the seat belt: one for strain and another one for bending. The strain sensor picks up breathing cycle signals while the bending sensors can indicate possible accidents.

I would like to thank our hosts for letting us explore their system and for answering all of our questions. The exchanged knowledge will surely benefit both sides.

Oszkar Palinko

Here is an unnerving article about a report by Privacy International detailing the lack of privacy protection around the world, including here in the US (yeah, we didn’t rank very highly). PI also runs the Big Brother Awards. Check out this year’s winners in the Netherlands.

Andrew Kun