This is a JOINT POST with Evangelos Georgiadis who brought this problem to my attention.)
In 2010 I posted about how dense a set of integers has to be before you know there is a 3-AP in it (a 3-AP is a set of three numbers equally spaced). Such results were motivated by and are applied to getting upper bounds on
W(3,r) = the least W such that any r-coloring of {1,...,W} has a monochromatic 3-AP.
That blog, which also has history and context, is here
At the time of that post the following was known (and had just been proven by Sanders see here)
If A ⊂ {1,...,N} and |A| ≥ N*(log log N)5 / log N then A has a 3-AP
and hence
W(3,r) ≤ 2r(log r)5
(Added later: Bloom's paper (see link on next line) reports that Sanders result is a bit weaker than claimed- the 5's should be 6's, calculation error.)
This has now been improved!. Thomas Bloom, in this paper has shown
If A ⊂ {1,...,N} and |A| ≥ N*(log log N)4 / log N then A has a 3-AP
and hence
W(3,r) ≤ 2r(log r)4
An easy prob argument gives
W(3,r) ≥ r3/2
So is W(3,r)'s growth rate poly? Exp? in between? Is there a connection to SAT?
One can phrase the question W(k,r)=m as asking of a certain Boolean Formula is it satisfiable. IF we can get theorems about that kind of formula, that might help.
However, I doubt that it has much real connection to the general SAT problem.
I don't know if the consensus of the community is that W(3,r) is poly or exp.
(Warning: I've seen W(3,r) to mean something else: W(3,r) is the least W such that any 2-coloring of {1,...,W} with colors 0 and 1 has either a 0-colored 3-AP or a 1-colored k-AP. This is NOT the function we are talking about, though that one is also interesting. )
ADDED LATER: Knuth's Volume 4, Fascicle 6 has theorems about this W(3,r)
ADDED LATER: A commenter said that a simple prob greedy algorithm using salem-spencer sets
yields
W(3,r) ≥ exp(C (ln r)2)
I reconstructed the proof from these comments, though using Behrends sets instead of SS
The proof is here: here
Monday, June 05, 2017
Thursday, June 01, 2017
Who Sets Policy?
In April the New York Times Magazine ran an article Is it O.K. to Tinker with the Environment to Fight Climate Change? The article asks about the ethics of even running tests on such methods and has this quote froms David Battisti, an atmospheric scientist at UW.
As computer scientists we start to face these questions as software in our hyper-connected world starts to change society in unpredictable ways. How do we balance privacy, security, usability and fairness in communications and machine learning? What about net neutrality, self-driving cars, autonomous military robots? Job disruption from automation?
We have governments to deal with these challenges. But the world seems to have lost trust in its politicians and governments don't agree. How does one set different rules across states and countries which apply to software services over the Internet?
All too often companies set these policies, at least the default policies until government steps in. Uber didn't ask permission to completely change the paid-ride business and only a few places pushed back. Google, Facebook, etc. use machine learning with abandon, until some governments try and reign them in. The Department of Defense and the NSA, in some sense industries within government, set their policies often without public debate.
What is our role as computer scientists? It's not wrong to create the technologies, but we should acknowledge the ethical questions that come with them and what we technically can and cannot do to address them. Keep people informed so the decision makers, whomever they be, at least have the right knowledge to make their choices.
Name a technology humans have developed that they haven't used. I can't think of any. So we can work on this for sure. But we are in this dilemma: Once we do develop this technology, it will be tempting to use it.The article skirts the question on who makes this decision. Maybe the United Nations after some unlikely agreement among major powers. But what if the UN doesn't act and some billionaire decides to fund a project?
As computer scientists we start to face these questions as software in our hyper-connected world starts to change society in unpredictable ways. How do we balance privacy, security, usability and fairness in communications and machine learning? What about net neutrality, self-driving cars, autonomous military robots? Job disruption from automation?
We have governments to deal with these challenges. But the world seems to have lost trust in its politicians and governments don't agree. How does one set different rules across states and countries which apply to software services over the Internet?
All too often companies set these policies, at least the default policies until government steps in. Uber didn't ask permission to completely change the paid-ride business and only a few places pushed back. Google, Facebook, etc. use machine learning with abandon, until some governments try and reign them in. The Department of Defense and the NSA, in some sense industries within government, set their policies often without public debate.
What is our role as computer scientists? It's not wrong to create the technologies, but we should acknowledge the ethical questions that come with them and what we technically can and cannot do to address them. Keep people informed so the decision makers, whomever they be, at least have the right knowledge to make their choices.
Tuesday, May 30, 2017
Google Scholar thinks my Hilbert Number is 1
(I want to thank Lane Hemaspaandra for bringing this to my attention.)
When I google:
google scholar William Gasarch
I get this
(I wonder if what you get depends on who you are- I describe below what I get.)
The first entry on it is
Methods of Mathematical Physics by Courant and Hilbert.
The first page page that Google brings up has all papers with Hilbert as a co-author except the book
Handbook of discrete and combinatorial mathematics edited by Rosen
I have a short article in that book.
The second page has the paper of Hilbert:
Uber die irreducible ganzer rationaler blah blah
which I have a connection to since Mark Villarino, Bill Gasarch, and Ken Regan have a paper explaining this paper in modern terms here. Could this have confused google scholar into thinking I am Hilbert? Seems unlikely.
In fact, the first two pages are all papers with Hilbert as an author. The third page has
Some connections between bounded query classes and non-uniform complexity by Amir, Beigel, Gasarch.
And after that the pages are a mix of Hilbert's papers and mine, though mostly his since he had so many more papers than I did.
This raises some questions
1) How did this happen? Did I hack google scholar? Many theorists at one time in their lives were excellent programmers- however, I am not one of them.
2) How common is it for google scholar to be this far wrong?
3) If I wanted to get this fixed who would I tell? In the past whenever I've tried to tell a company something FOR THEIR OWN GOOD I get the same run-around as when I am trying to get information not on their website, so I have STOPPED even trying. Same here?
4) Should I want to get it fixed? I am proud to be affiliated with Hilbert!
When I google:
google scholar William Gasarch
I get this
(I wonder if what you get depends on who you are- I describe below what I get.)
The first entry on it is
Methods of Mathematical Physics by Courant and Hilbert.
The first page page that Google brings up has all papers with Hilbert as a co-author except the book
Handbook of discrete and combinatorial mathematics edited by Rosen
I have a short article in that book.
The second page has the paper of Hilbert:
Uber die irreducible ganzer rationaler blah blah
which I have a connection to since Mark Villarino, Bill Gasarch, and Ken Regan have a paper explaining this paper in modern terms here. Could this have confused google scholar into thinking I am Hilbert? Seems unlikely.
In fact, the first two pages are all papers with Hilbert as an author. The third page has
Some connections between bounded query classes and non-uniform complexity by Amir, Beigel, Gasarch.
And after that the pages are a mix of Hilbert's papers and mine, though mostly his since he had so many more papers than I did.
This raises some questions
1) How did this happen? Did I hack google scholar? Many theorists at one time in their lives were excellent programmers- however, I am not one of them.
2) How common is it for google scholar to be this far wrong?
3) If I wanted to get this fixed who would I tell? In the past whenever I've tried to tell a company something FOR THEIR OWN GOOD I get the same run-around as when I am trying to get information not on their website, so I have STOPPED even trying. Same here?
4) Should I want to get it fixed? I am proud to be affiliated with Hilbert!
Thursday, May 25, 2017
Graduation from the Other Side
I've attended many graduations in my time, mostly as faculty, a couple of times as a student or a brother. This last weekend I attended my first university graduation as a parent as my daughter Annie graduated from Brandeis University in Waltham, Massachusetts. Brandeis has a big graduation ceremony with lots of speeches and then different departments or groups of departments have their own diploma ceremonies with their own speakers and where they give out the actual diplomas.
Brandeis gives out a number of honorary doctorates each year and for the first time gave one to a computer scientist, Turing Award Winner Leslie Lamport. Lamport received his PhD at Brandeis in math in 1972 before they had a CS deparment but now he has an (honorary) PhD in Computer Science. Lamport gave an eight-minute talk in the School of Science ceremony. But when you are a parent the weekend is about your child and my daughter didn't graduate from the school of science so I didn't see the Lamport talk.
In the main ceremony, Brandeis has not only an undergrad give a speech but also a grad student. Sounds like a crazy idea, but Vivekanand Vimal, Neuroscience PhD, gave what could be best described as a performance art. Since I can't find the video of Lamport and you probably don't want to see my videos of Annie, enjoy the new Dr. Vimal's ode to the craziness of the PhD and saving society.
Thursday, May 18, 2017
The Optimizers
In theoretical CS we often take NP-complete as a sign to stop searching for an efficient algorithm. Optimization people take NP-complete as a starting point, using powerful algorithmic ideas, clever heuristics and sheer computing power to solve or nearly optimize in many real-world cases.
Bill Cook talked about his adventures with the traveling salesman problem. Check out his British pub crawl and his tour through the nearly 50,000 US historic sites.
Michael Trick talked about his side job, schedule MLB baseball games, a surprisingly challenging problem. Like TSP, you want to minimize total travel distance but under a wide variety of constraints. "There's something satisfying about being at a bar, seeing a game on the TV and knowing those two teams are playing because you scheduled them." Can't say I've had that kind of satisfaction in my computational complexity research.
Tuesday, May 16, 2017
If an ugrad asks `is field X worth studying' the answer is almost always yes
An undergraduate Freshman recently emailed me that he was very interested in Quantum Computing and wanted to know
1) Who on the fCS aculty works in QC (Answer: Andrew Childs though you should ask him about postdocs, grad students, and Physics faulty in the area.)
2) What are good books on QC for a bright ugrad. I said the following:
QC since Democritus by Aaronson
QC-A gentle introduction by Rieffel and Polak
QC for CS by Yanofsy and Mannucci
QC and QI by Nielsen and Chuang
Some of Scott's blog posts.
Ask Andrew Childs for more.
my webpage of book reviews for SIGACT NEWS here and search for Quantum to get some other books- read the reviews and pick one.
on Amazon type in quantum computing and see what reviews say- though they might not be reliable.
There are likely other good books but I do not know of them. (You can leave comments.)
3) Is QC a good topic to get into? I said YES of course. My reasoning is that they would of course LEARN something by studying it.
But this raises the question: When would I say `that field is not worth studying' ?
1) If they really want to do RESEARCH and the topic is either too dead or too hard and they want to actually do research (as opposed to learning the topic without wanting to to research).
2) If there was nobody around to help them in that topic. Might still be okay if they are both highly motivated and very smart.
3) If the topic was bogus AND they would learn NOTHING from studying it. Are there topics that are bogus but you still learn from studying them? Does studying astrology seriously teach you some astronomy? Some history? How about Alchemy and Chemistry? Fine if the students KNOWS that Astrology is bogus and Alchemy is not correct.
The points is that I really do not want to dampen someone's enthusiasm for a topic.
SO- aside from the reasons above, can you think of any other reason to discourage a student from a topic they are interested in? I ask, as always, non-rhetorically.
1) Who on the fCS aculty works in QC (Answer: Andrew Childs though you should ask him about postdocs, grad students, and Physics faulty in the area.)
2) What are good books on QC for a bright ugrad. I said the following:
QC since Democritus by Aaronson
QC-A gentle introduction by Rieffel and Polak
QC for CS by Yanofsy and Mannucci
QC and QI by Nielsen and Chuang
Some of Scott's blog posts.
Ask Andrew Childs for more.
my webpage of book reviews for SIGACT NEWS here and search for Quantum to get some other books- read the reviews and pick one.
on Amazon type in quantum computing and see what reviews say- though they might not be reliable.
There are likely other good books but I do not know of them. (You can leave comments.)
3) Is QC a good topic to get into? I said YES of course. My reasoning is that they would of course LEARN something by studying it.
But this raises the question: When would I say `that field is not worth studying' ?
1) If they really want to do RESEARCH and the topic is either too dead or too hard and they want to actually do research (as opposed to learning the topic without wanting to to research).
2) If there was nobody around to help them in that topic. Might still be okay if they are both highly motivated and very smart.
3) If the topic was bogus AND they would learn NOTHING from studying it. Are there topics that are bogus but you still learn from studying them? Does studying astrology seriously teach you some astronomy? Some history? How about Alchemy and Chemistry? Fine if the students KNOWS that Astrology is bogus and Alchemy is not correct.
The points is that I really do not want to dampen someone's enthusiasm for a topic.
SO- aside from the reasons above, can you think of any other reason to discourage a student from a topic they are interested in? I ask, as always, non-rhetorically.
Sunday, May 14, 2017
William Tutte (1917-2002)
Today we celebrate our mothers of course, but also the 100th anniversary of the birth of Bill Tutte, best known for his role in decrypting the Lorenz cipher used by the Nazi high command. Tutte also made many important advances in graph theory and algorithms.For this post, let's look at one very powerful concept, the Tutte Polynomial, with a rather technical looking definition. Fix a graph G with vertex set V and edge set E with n = |V|. For a subset A of E, let kA be the number of connected components of A and nA be the number of vertices of the vertices of G induced by A, and k=kE the number of connected components of G.
The Tutte polynomial T(x,y) is the sum over all subsets A of E of the quantity
(x-1)kA-k(y-1)KA+nA-n.
What makes this problem interesting? For some fixed values of x and y we get various properties of the graph.
T(2,1) is the number of forests of G.
T(1,2) number of spanning forests (or spanning trees if G is connected.
T(2,0) is the number of spanning subgraphs.
T(0,2) is the number of strongly connected orientations.
The value (-1)n-kqkT(1-q,0) counts the number of q-colorings of G.
Computing T can be difficult. Counting the number of 3-colorings is #P-complete, equivalent to counting the number of satisfying assignments of a Boolean formula. So even computing T(-2,0) for a given graph G is #P-complete. Leslie Goldberg and Mark Jerrum show that even computing the sign of a Tutte polynomial, just determining whether it is positive, zero or negative, on certain values is still #P-hard.
This is only a sampling of the many applications of the Tutte polynomial. Let's remember Tutte for creating a single function that captures so much information about a graph and helping to defeat the Nazis. Not a bad life. Must have had a good mother.
Computing T can be difficult. Counting the number of 3-colorings is #P-complete, equivalent to counting the number of satisfying assignments of a Boolean formula. So even computing T(-2,0) for a given graph G is #P-complete. Leslie Goldberg and Mark Jerrum show that even computing the sign of a Tutte polynomial, just determining whether it is positive, zero or negative, on certain values is still #P-hard.
This is only a sampling of the many applications of the Tutte polynomial. Let's remember Tutte for creating a single function that captures so much information about a graph and helping to defeat the Nazis. Not a bad life. Must have had a good mother.
Thursday, May 11, 2017
How to Solve It
Today a guest post from Periklis Papakonstantinou, coincidentally not unrelated to Bill's post earlier this week. I'll be back with a special post on Sunday.
I'm teaching in an undergrad program that is half computer science and half business at Rutgers, but the CS part taught there is the real thing (I assume for Business too). This term I taught a very theoretical course in cryptography and I realized that (1) the students enjoyed it and (2) that they were lacking basic reasoning skills. I ended up teaching for a few weeks how one can structure basic logic arguments. I am not sure if they appreciated things like the hybrid argument but I believe I convinced them that without rigorous thinking one cannot think clearly.
So, I decided to teach a much more fun class (hopefully next year) titled "How to solve it" -- à la Pólya. The goal is students to develop rigorous problem-solving skills. At the same time, I'd like to use this course as an excuse to introduce basic concepts in combinatorics, linear algebra, and theoretical stats. I'm not sure whether the original book by Polya is appropriate for this and that's why I thought of reaching out to my peers for suggestions. Any ideas and thoughts on possible texts, topics, or notes would be greatly appreciated.
I'm teaching in an undergrad program that is half computer science and half business at Rutgers, but the CS part taught there is the real thing (I assume for Business too). This term I taught a very theoretical course in cryptography and I realized that (1) the students enjoyed it and (2) that they were lacking basic reasoning skills. I ended up teaching for a few weeks how one can structure basic logic arguments. I am not sure if they appreciated things like the hybrid argument but I believe I convinced them that without rigorous thinking one cannot think clearly.
So, I decided to teach a much more fun class (hopefully next year) titled "How to solve it" -- à la Pólya. The goal is students to develop rigorous problem-solving skills. At the same time, I'd like to use this course as an excuse to introduce basic concepts in combinatorics, linear algebra, and theoretical stats. I'm not sure whether the original book by Polya is appropriate for this and that's why I thought of reaching out to my peers for suggestions. Any ideas and thoughts on possible texts, topics, or notes would be greatly appreciated.
Sunday, May 07, 2017
Students try to memorize rather than understand! Who knew! (everyone)
Discrete Math. Required for CS majors, taken mostly by Sophmores. Goal is to teach them how to think rigorously. Topics are logic, number theory (not much), induction, sets, functions, relations, combinatorics (includes Pigeon hole prin, henceoforth PNP), prob, countability, uncountability.
We taught the Pigeon Hole Principle and gave MANY examples and HW of the following type:
Let A be a subset of {1,...,50} of size 10. Show there are two subsets of A that have the same sum.
ANSWER: There are 2^{10} = 1024 possible subsets.
MAXSUM is 41+..+50 = (1+...+ 50 )-(1 +...+ 40) = 50*51/2 - 40*41/2 = 455
MINSUM is 0 (the empty set). So the NUMBER OF SUMS is 456
Since 1024 > 456 there are two subsets of A of the same size.
The EXAM covered PHP, combs, prob, and induction. Hence they should know n choose k
On the HW and in class we NEVER did a problem where we only cared about the subsets of a fixed size. Conceptually this is really the same problem but if you had MEMORIZED the proof template and tried to apply it you would get it wrong.
I asked the following on the exam: which was worth 20 points.
(20 points) Let A be a subset of {1,...,21} of size 8. Show that A has at least two subsets of size 3 which have the same sum.
ANSWER: There are (8 choose 3) = 56 subsets of A of size 3.
MAXSUM = 19+20+21 = 60, MINSUM = 1+2+ 3 = 6,
So the NUMBER OF SUMS is 60-5 = 55.
Since 56> 55 there are two subsets of A of size 3 that have the same size.
Grading rubric:
If they got (8 choose 3) thats 3 points.(Many said 2^8- I suspect incorrect memorization)
If they got the MAXSUM and the MINSUM both right (aritj errors- NO penalty) then 3 points
(Many had MINSUM=0- I suspect incorrect memorization).
If they knew to use these PHP then 3 points.
If they got all three right then 20 points
So they got 0,3,6, or 20.
Here is the final tally:
0 points: 85
3 points: 190
6 points: 71
20 points: 167
(For the entire exam: 39 100's, 55 90-99, 77 80-89, 77 70-79, 76 60-69, 70 50-59, 47 40-49, 24 30-39, 24 30-39, 17 20-29, 3 10-19, 3 1-9)
This all raises the much harder question- how can we get students to UNDERSTAND rather than MEMORIZE
Telling them: DO NOT MEMORIZE! TRY TO UNDERSTAND!- I did this. Oh well.
Allowing a cheat sheet (which I did) is both good and bad for this issue.
Giving them a much wider variety of problems of this type. Either they would understand OR they would memorize several different templates.
I WELCOME your thoughts on either my grading or on how to get them to try to UNDERSTAND rather than MEMORIZE.
Thursday, May 04, 2017
Summer Conferences
Ahh summer. No Classes. Baseball. Opera Festivals. Time to focus on research and starting a new book. But, of course, many computer scientists travel the world to various conferences. I went to too many last year and trying to cut down but many great options abound.
The STOC 2017 Theory Fest, June 19-23 in Montreal, five days of conference talks, tutorials, invited lectures and so much more. Sanjeev Arora has the details over at Windows on Theory.
The ACM celebrates 50 years of Turing Awards with a special conference June 23-24 in San Francisco. Tim Berners-Lee takes home this year's prize.
The Computational Complexity Conference, that meeting that shares its domain with this blog, holds its annual get together July 6-9 for the first time in Latvia. Latvia gave us Juris Hartmanis, one of the founders of the field. Travel grants available for students and "needy researchers", you don't have to be an author to apply.
Computability in Europe, June 12-16 in Turku, Finland. Economics and Computation, June 26-30 at MIT. Computational Geometry, July 4-7 in Brisbane. ICALP, July 10-14 in Warsaw. Random/Approx, August 16-18 in Berkeley.
If I missed your favorite events, well that's why we have comments.
The STOC 2017 Theory Fest, June 19-23 in Montreal, five days of conference talks, tutorials, invited lectures and so much more. Sanjeev Arora has the details over at Windows on Theory.
The ACM celebrates 50 years of Turing Awards with a special conference June 23-24 in San Francisco. Tim Berners-Lee takes home this year's prize.
The Computational Complexity Conference, that meeting that shares its domain with this blog, holds its annual get together July 6-9 for the first time in Latvia. Latvia gave us Juris Hartmanis, one of the founders of the field. Travel grants available for students and "needy researchers", you don't have to be an author to apply.
Computability in Europe, June 12-16 in Turku, Finland. Economics and Computation, June 26-30 at MIT. Computational Geometry, July 4-7 in Brisbane. ICALP, July 10-14 in Warsaw. Random/Approx, August 16-18 in Berkeley.
If I missed your favorite events, well that's why we have comments.
Monday, May 01, 2017
A Celebration of Computer Science at Harvard in Honor of Harry Lewis's 70th Bday
My adviser Harry Lewis turned 70 recently. I blogged about how things have changed since I got my Phd in this post. I now post on
A celebration of Computer Science at Harvard in Honor of Harry Lewis's 70th Birthday
(for video of all talks in order see: here)
The title was accurate: most of the speakers (1) were Harvard ugrads, (2) went on to do great things, and (3) Harry Lewis had inspired them. The talks were mostly non-technical and fun!
went on to do great things. Margo Seltzer, a prof at Harvard now (who I TAed many years ago in Aut Theory) orgnaized the event, though she gave lots of credit to her helpers.
0) Marty Chavez was one of Harry Lewis's teaching assistants for a CS programming course and recalled Harry's harsh (but fair) grading polices on code which he later saw the wisdom of.
1) Marty Chavez never thought he would use that HALT is undecidable (I think I might have been his TA for that course). But he found himself telling an egghead of economists who wanted to VERIFY all code to avoid future crashes that... Can't be done. Actually, while that is true, attempts to verify some of it might be a good idea.
2) James Gwertzman noted that:
in 1991 10% of all ugrads at Harvard had email, and there was no web
in 1995 100%of all ugrads at Harvard had email, and there was web (though primitive).
He then pointed out that a company can do very well by using LOTS of packages that are already out there to use. He named #slack, salesforce, trello, jeaking, mailchip, greenhouse, phabricator, pingdom (just deals with pings- really!), datadog, strips, statuspage, zendeski.
The future will be serverless and codeless.
4) Guy Steele gave the most technical talk and it was, as the kids say, awesome (do adults still say `as the kids say' ?) Here is a version of the talk:
A Logial Concern
Its about how papers at POPL and some other conference have been informally using a language to specify protocols and by now its all bent out of shape. There is also some nice history of math embedded in the talk of which I'll say one thing: one way to group terms together is by placing a bar over them. The most common use of this now is the squareroot sign which didn't always have that bar over the quantity.
Guy's talk even had some slides about his notebooks from Harvard, from a course Harry taughtback in 1974 (the first course Harry taught at Harvard). Part of the course was on the sequent calculus which relates to Guys work and the current paper. Guy's notebook had both material relevent to the current paper and doodles of things like a picture of a Church next to Church's thesis.
The paper was very labor intensive since you can't just use a search program to search for some of the notations he was talking about. For example overbar and underbar. So he had to go through ALL of the POPL proceedings (and a few others) by hand. In 2017. Will that ever be easier?
He also had a two quotes about proofs:
Its not enough to prove something. You must seduce people into believing it
One man's truth is another man's cold broccoli
I leave it to you to figure out who these quotes are credited to (different people).
5) Stuart Shieber's talk was WWHD (What Would Harry Do).
Care
Promote Character over knowledge (see Harry's book Excellence without a soul- How a great university forgot education)
Pursue the right over the popular
A late talk by Rebecca Nessin told of some things Harry did as Dean that were RIGHT but NOT POPULAR:
The housing at Harvard used to be you chose the house (dorm complex) you lived in. When I was there Dunster was KNOWN to be the Math-house, and others had other reputations that were somewhat accurate. Harry made housing RANDOMIZED (did he use a hardness result to derive a pseudo random generation?) His goal was to increase diversity- people should get to know other kinds of people that are not like themselves.
He made polices do curb underage drinking.
He raised standards for when students get WARNINGS about their performance in classes.
These were all unpopular BUT the right thing to do.
6) There was a panel discussion on teaching. I'll save this for a later blog post since my random thoughts on this may make this post longer than it should be. I WILL say it was excellent.
7) Rebecca Nessin is the head online course development at Harvard. The courses are (1) open enrollment (2) No faculty- all are borrowed from the usual faculty, (3) some courses are online. She developed a course where the students ARE avatars. Helps with shy students. And text based conversation allows students to get out coherent complete thoughts (CONTRAST- I find myself saying to my students questions ``that was a random sequence of math words'')
That was the first part of her talk.
THEN she began talking about her journey through Harvard and Harry's place in it. Unlike her fellow students she did not what she wanted to do. She took random courses (ancient greek! Multivar calc!) After graduating she still did not know what she wanted to do so she went to... Harvard Law School. While there she took a CS course (what! You can do that?) and soon after had Aut Theory with Harry. Her PhD was with Stuart Shieber with Harry on the committee and lots of Grammars in it.
Then she told a great story: There was a discussion of raising the min age that someone can get an ugrad degree at the Harvard extension school. Harry asked who this would affect. The answer was
A small number. Students who can't go to a residential full time school for some reason. This includes competitive athletes, performance artists, deployed military personal, youthful entrepreneurs, and people with disabilities.
to which Harry replied:
These are the oddballs. Are we trying to say there is no room for oddballs at Harvard?
Rebecca ended her talk by pointing out that with her crooked path to where she is now she is an oddball and
that all of the oddballs should celebrate that they will also have a place at at Harry Lewis's Harvard.
8) Cliff Young declared Moore's Law Dead (some disagree- see here) - and the solution is to go back to special purpose machines- which, contrary to popular belief, Do NOT just do one thing and
ARE programmable, He also talked about Amdahl's law which is about the limits of parallelism and about how parallelism research seems to fight the same battle over and over again (RISC vs CISC,
SIMD vs MIMD, and VLIW)
9) Danielle Feinberg from Pixar had the following quote about animation:
Long hair is an unsolved problem
But they did solve it (for the movie The Incredibles). She also pointed out that they can sometimes spend lots of time and energy and creativity on a scene that will take 3 second, or on something just in the background.
Much like Rebecca, Danielle also appreciated Harrys appreciating for oddballs.
10) Harry Lewis- He spoke some about his career but also about CS in general.
His career and where his now is sort-of an accident. He was originally going to get his PhD in Systems but Theorists got out faster.
Computer Science has changed a lot in the last X years- but the change he remarked on the most is that it CS is at
The twilight of the Amateur Era
I'll let you debate what that means.
11) Later at the reception Bill Gates and Mark Zuckerberg send their recorded greetings, though only Mark Z's is on the you tube video- towards the end. Its short so rather than summarize it- I urge to to view it yourself.
A celebration of Computer Science at Harvard in Honor of Harry Lewis's 70th Birthday
(for video of all talks in order see: here)
The title was accurate: most of the speakers (1) were Harvard ugrads, (2) went on to do great things, and (3) Harry Lewis had inspired them. The talks were mostly non-technical and fun!
went on to do great things. Margo Seltzer, a prof at Harvard now (who I TAed many years ago in Aut Theory) orgnaized the event, though she gave lots of credit to her helpers.
0) Marty Chavez was one of Harry Lewis's teaching assistants for a CS programming course and recalled Harry's harsh (but fair) grading polices on code which he later saw the wisdom of.
1) Marty Chavez never thought he would use that HALT is undecidable (I think I might have been his TA for that course). But he found himself telling an egghead of economists who wanted to VERIFY all code to avoid future crashes that... Can't be done. Actually, while that is true, attempts to verify some of it might be a good idea.
2) James Gwertzman noted that:
in 1991 10% of all ugrads at Harvard had email, and there was no web
in 1995 100%of all ugrads at Harvard had email, and there was web (though primitive).
He then pointed out that a company can do very well by using LOTS of packages that are already out there to use. He named #slack, salesforce, trello, jeaking, mailchip, greenhouse, phabricator, pingdom (just deals with pings- really!), datadog, strips, statuspage, zendeski.
The future will be serverless and codeless.
4) Guy Steele gave the most technical talk and it was, as the kids say, awesome (do adults still say `as the kids say' ?) Here is a version of the talk:
A Logial Concern
Its about how papers at POPL and some other conference have been informally using a language to specify protocols and by now its all bent out of shape. There is also some nice history of math embedded in the talk of which I'll say one thing: one way to group terms together is by placing a bar over them. The most common use of this now is the squareroot sign which didn't always have that bar over the quantity.
Guy's talk even had some slides about his notebooks from Harvard, from a course Harry taughtback in 1974 (the first course Harry taught at Harvard). Part of the course was on the sequent calculus which relates to Guys work and the current paper. Guy's notebook had both material relevent to the current paper and doodles of things like a picture of a Church next to Church's thesis.
The paper was very labor intensive since you can't just use a search program to search for some of the notations he was talking about. For example overbar and underbar. So he had to go through ALL of the POPL proceedings (and a few others) by hand. In 2017. Will that ever be easier?
He also had a two quotes about proofs:
Its not enough to prove something. You must seduce people into believing it
One man's truth is another man's cold broccoli
I leave it to you to figure out who these quotes are credited to (different people).
5) Stuart Shieber's talk was WWHD (What Would Harry Do).
Care
Promote Character over knowledge (see Harry's book Excellence without a soul- How a great university forgot education)
Pursue the right over the popular
A late talk by Rebecca Nessin told of some things Harry did as Dean that were RIGHT but NOT POPULAR:
The housing at Harvard used to be you chose the house (dorm complex) you lived in. When I was there Dunster was KNOWN to be the Math-house, and others had other reputations that were somewhat accurate. Harry made housing RANDOMIZED (did he use a hardness result to derive a pseudo random generation?) His goal was to increase diversity- people should get to know other kinds of people that are not like themselves.
He made polices do curb underage drinking.
He raised standards for when students get WARNINGS about their performance in classes.
These were all unpopular BUT the right thing to do.
6) There was a panel discussion on teaching. I'll save this for a later blog post since my random thoughts on this may make this post longer than it should be. I WILL say it was excellent.
7) Rebecca Nessin is the head online course development at Harvard. The courses are (1) open enrollment (2) No faculty- all are borrowed from the usual faculty, (3) some courses are online. She developed a course where the students ARE avatars. Helps with shy students. And text based conversation allows students to get out coherent complete thoughts (CONTRAST- I find myself saying to my students questions ``that was a random sequence of math words'')
That was the first part of her talk.
THEN she began talking about her journey through Harvard and Harry's place in it. Unlike her fellow students she did not what she wanted to do. She took random courses (ancient greek! Multivar calc!) After graduating she still did not know what she wanted to do so she went to... Harvard Law School. While there she took a CS course (what! You can do that?) and soon after had Aut Theory with Harry. Her PhD was with Stuart Shieber with Harry on the committee and lots of Grammars in it.
Then she told a great story: There was a discussion of raising the min age that someone can get an ugrad degree at the Harvard extension school. Harry asked who this would affect. The answer was
A small number. Students who can't go to a residential full time school for some reason. This includes competitive athletes, performance artists, deployed military personal, youthful entrepreneurs, and people with disabilities.
to which Harry replied:
These are the oddballs. Are we trying to say there is no room for oddballs at Harvard?
Rebecca ended her talk by pointing out that with her crooked path to where she is now she is an oddball and
that all of the oddballs should celebrate that they will also have a place at at Harry Lewis's Harvard.
8) Cliff Young declared Moore's Law Dead (some disagree- see here) - and the solution is to go back to special purpose machines- which, contrary to popular belief, Do NOT just do one thing and
ARE programmable, He also talked about Amdahl's law which is about the limits of parallelism and about how parallelism research seems to fight the same battle over and over again (RISC vs CISC,
SIMD vs MIMD, and VLIW)
9) Danielle Feinberg from Pixar had the following quote about animation:
Long hair is an unsolved problem
But they did solve it (for the movie The Incredibles). She also pointed out that they can sometimes spend lots of time and energy and creativity on a scene that will take 3 second, or on something just in the background.
Much like Rebecca, Danielle also appreciated Harrys appreciating for oddballs.
10) Harry Lewis- He spoke some about his career but also about CS in general.
His career and where his now is sort-of an accident. He was originally going to get his PhD in Systems but Theorists got out faster.
Computer Science has changed a lot in the last X years- but the change he remarked on the most is that it CS is at
The twilight of the Amateur Era
I'll let you debate what that means.
11) Later at the reception Bill Gates and Mark Zuckerberg send their recorded greetings, though only Mark Z's is on the you tube video- towards the end. Its short so rather than summarize it- I urge to to view it yourself.
Thursday, April 27, 2017
So Was I
Most of the marchers I saw did not come from the ranks of academia or professional scientists. Rather people from all walks of life who believe in the important role science and scientists have in shaping our future. Parents dragged their kids. Kids dragged their parents.
There have been some worry about politicizing science and whether the march would be a bad idea. The march won't have much effect on policy positively or negatively. But we mustn't forget that scientists need to deal with politics, as long as the government continues its missions of funding science and using proper science to help guide policies that require understanding of the world.
If there's one positive sign of a Trump presidency, as Molly explained to me, it's inspiring a generation. We would not have had a March for Science if Trump wasn't president, but what an wonderful movement and we should march every year around Earth Day no matter who sits in the oval office.
Monday, April 24, 2017
I was at the March for Science on Saturday
(Will blog on Harry Lewis's 70th Bday next week-- Today's post is more time sensitive.)
I was on the March for Science on April 22. Here are some Kolmogorov random comments
1) Why should I go to it? One less person there would not have matters. AH- but if they all think that then nobody goes. The Classic Voting Paradox- why vote if the chance that your vote matters is so small (even less so in my state- Maryland is one of the Bluest States). In the case of the March For Science there is another factor- since I live in Maryland I really CAN go at minimal effort. Most of the readers of this blog cannot (Though there were some other marches in other cities. Scott was at a March in Austin Texas.)
2) One of the speakers said something like `and the fact that you are all here in the rain shows how much you believe in our cause!' While the rain might have made our being there more impressive, I wish it had been better weather.
3) Here are some of the Signs I saw:
What do we Want!
Empirical Based Science!
When do we Want it!
After Peer Review!
Trump- where's your PhD? Trump University?
(This one is not fair- most presidents have not been scientists and have funded science. Trump himself not have a PhD is not relevant here.)
A sign had in a circle: pi, sqrt(2) and Trump and said: These are all irrational.
A 6-year old had a sign: Light travels faster than sound which is why Trump looks bright until he talks (I think her mother, who was there, made it for her).
Science is the Solution (with a picture of a chemical Flask)
If you are not part of the solution you are part of the precipitate
Truth is sometimes inconvenient.
So severe even the nerds are here
I can't believe I'm marching for facts!
There is no planet B (this refers to if Global Warming kills the planet we can't go elsewhere- a play off of `Plan B')
I'm with her (pointing to the earth) (The person with this sign told me she used the same sign for the Women's March- so recycling!)
Science has no borders
Science doesn't care what you think.
Its not Rocket Science- well, some of it is.
4) The March For Science was the same day as Earth Day and many of the talks mentioned global warming and pollution. Many of the talks mentioned the contributions of women and minorities. One of the speakers was transgender .Hence the March had a liberal slant. BUT- if believing in Global Warming and wanting to open science up to all people (e.g., women and minorities) are Liberal positions, this speaks very badly of conservatives. First ACCEPT that Global Warming is TRUE- then one can debate what to do about it--and that debate could be a constructive political debate. One talk was about Indigenous Science-- I can't tell if its a healthy alternative view or ... not.
A more telling point about the march having a liberal slant is the OMISSION of the following topics:
Technology has
(a) helped Oil people extract more oil, and fracking to be cost effective
(b) GMO's have helped feed the world and have had no ill effects (I think anti-GMO in America is a fringe view-- I don't know of any elected democrat who is anti-GMO, though I could be wrong. I think its a more mainstream view in Europe.)
(c) make the weapons that keep us safe (that's a positive spin on it)
(d) DNA used to prove people GUILTY (they did mention DNA used to prove people INNOCENT).
So the March LOOKED like it was a bunch of Liberal Scientists. Does this make it less effective and easy for Trump and others to dismiss? Or are we so far past any hope of intelligent conversation that it doesn't matter?
5) Many of the machers, including Darling and me, had lunch at the Ronald Reagan Center. Is this an IRONY?
NO: Reagan funded the NSF as well as other presidents, see this blog post of Lance's from 2004. That post is interesting for other reasons: at the time Dems and Reps seemed to both RESPECT science. Trump may be the first one not to- though its early in his term so we'll see how it all pans out. Second, Lance has been blogging for a LONG time! (since 2003, and me since 2007).
YES: See these quotes by the Gipper (ask your grandparents why Reagan is called that):here
6) Will it have any effect? Short term I doubt it, Long term probably yes. An article about the impact of the the Women's March: here
7) There have been Women's Marches, The Million Man March, Civil Rights Marchs, pro-life, pro-choice, anti-war, pro-gay, anti-gay marches before. Has there ever been a March for Science before? Has there ever been a need before? I don't think so but I am asking non-rhetorically.
Cutting EPA because you don't believe in Global warming is appalling, (see here) but I understand politically where that comes from.
Not allowing funding of gun violence because you are pro-gun is appalling, (see here) but I understand politically where that comes from.
IF they cut funding on the study of evolution (Have republican presidents done that?) then that would be appalling but I would understand politically where it came from.
But cutting the NIH (see here) or the NSF (has he done that yet or is he just thinking of doing that?) I really DON"T understand- It does not even fit into the Republican Philosophy.
There should NOT be a NEED for a MARCH FOR SCIENCE, Or, to quote one of the signs
I can't believe I"m marching for facts!
I was on the March for Science on April 22. Here are some Kolmogorov random comments
1) Why should I go to it? One less person there would not have matters. AH- but if they all think that then nobody goes. The Classic Voting Paradox- why vote if the chance that your vote matters is so small (even less so in my state- Maryland is one of the Bluest States). In the case of the March For Science there is another factor- since I live in Maryland I really CAN go at minimal effort. Most of the readers of this blog cannot (Though there were some other marches in other cities. Scott was at a March in Austin Texas.)
2) One of the speakers said something like `and the fact that you are all here in the rain shows how much you believe in our cause!' While the rain might have made our being there more impressive, I wish it had been better weather.
3) Here are some of the Signs I saw:
What do we Want!
Empirical Based Science!
When do we Want it!
After Peer Review!
Trump- where's your PhD? Trump University?
(This one is not fair- most presidents have not been scientists and have funded science. Trump himself not have a PhD is not relevant here.)
A sign had in a circle: pi, sqrt(2) and Trump and said: These are all irrational.
A 6-year old had a sign: Light travels faster than sound which is why Trump looks bright until he talks (I think her mother, who was there, made it for her).
Science is the Solution (with a picture of a chemical Flask)
If you are not part of the solution you are part of the precipitate
Truth is sometimes inconvenient.
So severe even the nerds are here
I can't believe I'm marching for facts!
There is no planet B (this refers to if Global Warming kills the planet we can't go elsewhere- a play off of `Plan B')
I'm with her (pointing to the earth) (The person with this sign told me she used the same sign for the Women's March- so recycling!)
Science has no borders
Science doesn't care what you think.
Its not Rocket Science- well, some of it is.
4) The March For Science was the same day as Earth Day and many of the talks mentioned global warming and pollution. Many of the talks mentioned the contributions of women and minorities. One of the speakers was transgender .Hence the March had a liberal slant. BUT- if believing in Global Warming and wanting to open science up to all people (e.g., women and minorities) are Liberal positions, this speaks very badly of conservatives. First ACCEPT that Global Warming is TRUE- then one can debate what to do about it--and that debate could be a constructive political debate. One talk was about Indigenous Science-- I can't tell if its a healthy alternative view or ... not.
A more telling point about the march having a liberal slant is the OMISSION of the following topics:
Technology has
(a) helped Oil people extract more oil, and fracking to be cost effective
(b) GMO's have helped feed the world and have had no ill effects (I think anti-GMO in America is a fringe view-- I don't know of any elected democrat who is anti-GMO, though I could be wrong. I think its a more mainstream view in Europe.)
(c) make the weapons that keep us safe (that's a positive spin on it)
(d) DNA used to prove people GUILTY (they did mention DNA used to prove people INNOCENT).
So the March LOOKED like it was a bunch of Liberal Scientists. Does this make it less effective and easy for Trump and others to dismiss? Or are we so far past any hope of intelligent conversation that it doesn't matter?
5) Many of the machers, including Darling and me, had lunch at the Ronald Reagan Center. Is this an IRONY?
NO: Reagan funded the NSF as well as other presidents, see this blog post of Lance's from 2004. That post is interesting for other reasons: at the time Dems and Reps seemed to both RESPECT science. Trump may be the first one not to- though its early in his term so we'll see how it all pans out. Second, Lance has been blogging for a LONG time! (since 2003, and me since 2007).
YES: See these quotes by the Gipper (ask your grandparents why Reagan is called that):here
6) Will it have any effect? Short term I doubt it, Long term probably yes. An article about the impact of the the Women's March: here
7) There have been Women's Marches, The Million Man March, Civil Rights Marchs, pro-life, pro-choice, anti-war, pro-gay, anti-gay marches before. Has there ever been a March for Science before? Has there ever been a need before? I don't think so but I am asking non-rhetorically.
Cutting EPA because you don't believe in Global warming is appalling, (see here) but I understand politically where that comes from.
Not allowing funding of gun violence because you are pro-gun is appalling, (see here) but I understand politically where that comes from.
IF they cut funding on the study of evolution (Have republican presidents done that?) then that would be appalling but I would understand politically where it came from.
But cutting the NIH (see here) or the NSF (has he done that yet or is he just thinking of doing that?) I really DON"T understand- It does not even fit into the Republican Philosophy.
There should NOT be a NEED for a MARCH FOR SCIENCE, Or, to quote one of the signs
I can't believe I"m marching for facts!
Thursday, April 20, 2017
Will talk about Harry Lewis 70th bday conference later but for now- that was then/this is now
On Wed April 19 I was at the Harry Lewis 70th birthday celebration!
I will blog on that later.
Harry Lewis was my thesis adviser. Odd to use the past tense- I DID finish my thesis with him
and so he IS my adviser? Anyway, I will do a blog about the celebration next week.
This week I ponder- what was different then and now (I got my PhD in 1985).
False predictions that I made in 1985:
1) CS depts all have different views of what a CS major should know. By the year 2017 they will have figured out EVERY CS MAJOR SHOULD KNOW XXX and I will still write questions for the CS GRE. DID NOT HAPPEN. And a MINOR source of income for me has been cut off.
2) CS will be about 45% or more female. After all, the old guard is dying, its a new field without a tradition of sexism (this may have been false even then). Actually Women in CS has DECLINED since 1985. I'm still surprised since people in computing tend to be progressive. One could do several blog posts on this, but lacking the expertise I won't. (Gee bill- since when has lacking expertise stopped you before :-)
3) There will be some progress on P vs NP. Maybe an n^2 lower bound on SAT. Saying we've made NO progress is perhaps pessimistic, but we haven't made much.
4) in 2017 when Jet Blue emails me `CLICK HERE TO PRINT YOUR BOARDING PASS' the previous night then it will always work, and if it doesn't then I can call them and after 9 minutes on hold (not too bad) be able to fix the problem. They were not able to, though at the airport they fixed it and got me onto the plane fast as compensation.
OTHER CHANGES
1) Theory was more centralized. STOC and FOCS were the only prestige conferences, and everyone went to them.
2) A grad student could get a PhD and only have 2 papers published and get a Tenure Track Job.
3) One could learn all that was known in complexity theory in about two years.
4) You didn't have to do ugrad research to get into grad school (I don't think you HAVE TO now either, but many more do it so I PREDICT in the future you'll have to. Though my other predictions were not correct so .... there's that)
5) Complexity was more based in Logic then Combinatorics.
6) Complexity theory was easier! Gee, when did it get so hard and use so much hard math!
7) It seemed feasible that P vs NP would be solved within twenty years. I've heard it said that the Graph Minor Theorem was when P lost its innocence- there were now problems in P that used VERY HARD math--- techniques that were hard to pin down and hence hard to show would not work.
8) The number of complexity classes was reasonable. (I don't count Sigma_i as an infinite number of classes)
9) Grad students were just beginning to NOT learn the Blum Speed Up Theorem. It would take a while before they began to NOT learn finite injury priority arguments in recursion theory. OH- speaking of which...
10) Computability theory was called recursion theory.
11) Some schools had this odd idea that in FRESHMAN programming one should teach proofs of program correctness.
12) Some schools (SUNY Stonybrook and Harvard were among them) did not have a discrete math course. Hence the course in automata theory spend some of its time teaching how to prove things. (Both schools now have such a course. For Maryland I don't recall- either it didn't have one and I invented it OR it did have one and I revamped it.)
13) No Web. You had to go to a library to copy papers on a copier (Ask your grandparents what a copier is)
14) Copying cost far less than printing.
15) Someone who looked good on paper for MATH but had no real CS background could get into Harvard Applied science department for grad school and get a degree in ... speaking of which
16) In 1980 Harvard did not have a CS dept. So my Masters degree is formally in Applied Math, though I don't recall solving partial diff equations or other things that one associates with applied math. Sometime when I was there CS became officially something so I got my PhD in CS. (My students are surprised to hear this-- they think I got my PhD in Math.)
17) Harry Lewis had a moustache and smoked a pipe. He has shaved off one and gave up the other.
SO, what to make of this list? ONE THING- I DO NOT `yearn for the good old days' That was then, this is now. I am GLAD about everything on the list EXCEPT two area where NOT ENOUGH change has happened- (a) I wish there was more diversity in CS, and (b) I wish Jet Blue had better software for boarding passes.
I will blog on that later.
Harry Lewis was my thesis adviser. Odd to use the past tense- I DID finish my thesis with him
and so he IS my adviser? Anyway, I will do a blog about the celebration next week.
This week I ponder- what was different then and now (I got my PhD in 1985).
False predictions that I made in 1985:
1) CS depts all have different views of what a CS major should know. By the year 2017 they will have figured out EVERY CS MAJOR SHOULD KNOW XXX and I will still write questions for the CS GRE. DID NOT HAPPEN. And a MINOR source of income for me has been cut off.
2) CS will be about 45% or more female. After all, the old guard is dying, its a new field without a tradition of sexism (this may have been false even then). Actually Women in CS has DECLINED since 1985. I'm still surprised since people in computing tend to be progressive. One could do several blog posts on this, but lacking the expertise I won't. (Gee bill- since when has lacking expertise stopped you before :-)
3) There will be some progress on P vs NP. Maybe an n^2 lower bound on SAT. Saying we've made NO progress is perhaps pessimistic, but we haven't made much.
4) in 2017 when Jet Blue emails me `CLICK HERE TO PRINT YOUR BOARDING PASS' the previous night then it will always work, and if it doesn't then I can call them and after 9 minutes on hold (not too bad) be able to fix the problem. They were not able to, though at the airport they fixed it and got me onto the plane fast as compensation.
OTHER CHANGES
1) Theory was more centralized. STOC and FOCS were the only prestige conferences, and everyone went to them.
2) A grad student could get a PhD and only have 2 papers published and get a Tenure Track Job.
3) One could learn all that was known in complexity theory in about two years.
4) You didn't have to do ugrad research to get into grad school (I don't think you HAVE TO now either, but many more do it so I PREDICT in the future you'll have to. Though my other predictions were not correct so .... there's that)
5) Complexity was more based in Logic then Combinatorics.
6) Complexity theory was easier! Gee, when did it get so hard and use so much hard math!
7) It seemed feasible that P vs NP would be solved within twenty years. I've heard it said that the Graph Minor Theorem was when P lost its innocence- there were now problems in P that used VERY HARD math--- techniques that were hard to pin down and hence hard to show would not work.
8) The number of complexity classes was reasonable. (I don't count Sigma_i as an infinite number of classes)
9) Grad students were just beginning to NOT learn the Blum Speed Up Theorem. It would take a while before they began to NOT learn finite injury priority arguments in recursion theory. OH- speaking of which...
10) Computability theory was called recursion theory.
11) Some schools had this odd idea that in FRESHMAN programming one should teach proofs of program correctness.
12) Some schools (SUNY Stonybrook and Harvard were among them) did not have a discrete math course. Hence the course in automata theory spend some of its time teaching how to prove things. (Both schools now have such a course. For Maryland I don't recall- either it didn't have one and I invented it OR it did have one and I revamped it.)
13) No Web. You had to go to a library to copy papers on a copier (Ask your grandparents what a copier is)
14) Copying cost far less than printing.
15) Someone who looked good on paper for MATH but had no real CS background could get into Harvard Applied science department for grad school and get a degree in ... speaking of which
16) In 1980 Harvard did not have a CS dept. So my Masters degree is formally in Applied Math, though I don't recall solving partial diff equations or other things that one associates with applied math. Sometime when I was there CS became officially something so I got my PhD in CS. (My students are surprised to hear this-- they think I got my PhD in Math.)
17) Harry Lewis had a moustache and smoked a pipe. He has shaved off one and gave up the other.
SO, what to make of this list? ONE THING- I DO NOT `yearn for the good old days' That was then, this is now. I am GLAD about everything on the list EXCEPT two area where NOT ENOUGH change has happened- (a) I wish there was more diversity in CS, and (b) I wish Jet Blue had better software for boarding passes.
Monday, April 17, 2017
Understanding Machine Learning
Today Georgia Tech had the launch event for our new Machine Learning Center. A panel discussion talked about different challenges in machine learning across the whole university but one common theme emerged: Many machine learning algorithms seem to work very well but we don't know why. If you look at a neural net (basically a weighted circuit of threshold gates) trained for say voice recognition, it's very hard to understand why it makes the choices it makes. Obfuscation at its finest.
Why should we care? A few reasons:
Why should we care? A few reasons:
- Trust: How do we know that the neural net is acting correctly? Beyond checking input/output pairs we can't do any other analysis. Different applications have a different level of trust. It's okay if Netflix makes a bad movie recommendation, but if a self-driving car makes a mistake...
- Fairness: Many examples abound of algorithms trained on data will learn intended or unintended biases in that data. If you don't understand the program how do figure out the biases?
- Security: If you use machine learning to monitor systems for security, you won't know what exploits still might exist, especially if your adversary is being adaptive. If you can understand the code you could spot and fix security leaks. Of course if the adversary had the code, they might find exploits.
- Cause and Effect: Right now at best you can check that a machine learning algorithm only correlates with the kind of output you desire. Understanding the code might help us understan the causality in the data, leading to better science and medicine.
What if P = NP? Would that help. Actually it would makes things worse. If you had a quick algorithm for NP-complete problems, you could use it to find the smallest possible circuit for say matching or traveling salesman but you would have no clue why that circuit works.
Sometimes I feel we put to much pressure on the machines. When we deal with humans, for example when we hire people, we have to trust them, assume they are fair, play by the rules without at all understanding their internal thinking mechanisms. And we're a long way from figuring out cause and effect in people.
Thursday, April 13, 2017
Alice and Bob and Pat and Vanna
"The only useful thing computer science has given us is Alice and Bob" - A physicist at a 1999 quantum computing workshopAlice and Bob, great holders of secrets, seemed to pop into every cryptography talk and now you see them referenced anytime you have two parties who have something to share. Someone at Dagstuhl a few weeks back asked who first used Alice and Bob. What a great idea for a blog post, and I decided to do some binary searching through research papers to find that elusive first Alice and Bob paper. Turns out Wikipedia beat me to it, giving credit to Rivest, Shamir and Adleman in their paper A method for obtaining digital signatures and public-key cryptosystems, the paper that won them the Turing Award.
In grad school attending a square dance convention we ran into a married couple Alice and Bob and they couldn't figure out why we were laughing. Yes, I square danced in grad school, get over it.
The Wikipedia page lists a number of other common names used in protocols including perennial third wheel Charlie and nosy eavesdropper Eve. I can claim credit for two of those names, Pat and Vanna. In my first conference talk in 1987 I had to explain interactive proofs and for the prover and verifier I picked Pat and Vanna after the hosts, Pat Sajak and Vanna White, of the popular game show Wheel of Fortune. Vanna didn't trust Pat and spun the wheel to get random questions to challenge him. Half of the audience laughed hysterically, the other half had no clue what I was talking about. I heard the FOCS PC took a break by watching an episode of Wheel of Fortune to understand the joke.
Howard Karloff insisted we use Pat and Vanna in the LFKN paper. Pat and Vanna have since retired from interactive proving but thirty years later they still host Wheel of Fortune.
Monday, April 10, 2017
What is William Rowan Hamilton know for- for us? for everyone else?
I found the song William Rowan Hamilton that I used in my April fools day post because I was working on a song about Hamiltonian Circuits to the tune of Alexander Hamilton
Circuit Hamiltonian
I want a Circuit Hamiltonian
And I'm run-ing a pro-GRAM for it
So I wait, so I wait
(Darling said: Don't quit your day job.)
I noticed that William Rowan Hamilton had the same cadence as Alexander Hamilton so I assumed that someone must have used that for a parody, and I was right
But
Listen to the son. They mention the following::
Kinetics, Quaternions (This is mentioned the most), An optimization view of light, Minimal action,
`your energy function generates the flow of time' ,Operators that Lie Commute with the symbol that bears your name, His versors(?) formed hyperspheres - see if you can plot 'em- invented vectors and scalars for when you dot 'em (Did he really invent vectors? Wikipedia says that in a sense he invented cross and dot products.) And Schrodinger sings that he adapted Hamilton's work for Quantum (I didn't know Schrodinger could sing!).
What do they NOT mention: Hamiltonian paths or circuits. His Wikipedia page does mention Hamiltonian circuits, but not much and you would have no idea they were important.
When a computer science theorists hears `Hamiltonian' she prob thinks `path' or `circuit' and not `an optimization view of light' or anything else in physics' She might think of Quaternions and if she does Quantum Computing she may very well think of some of the items above. But these are exception. She would likely think of the graph problems.
The rest of the world would think of the list above (or would think William Rowan Hamilton died in a dual over Quaternions and later had the best Hamilton Satire written about him- the second of course being the one about Batman,)
In his own time he was best know for Physics. Maybe also quaternions. I think Hamilton himself would be surprised that this problem became important. So here is my question:
When did the problem become important? Before NP-Completness or after?
Is he still better known for his physics and quats- I think yes.
When I say `Hamilton' what comes to YOUR mind?
Circuit Hamiltonian
I want a Circuit Hamiltonian
And I'm run-ing a pro-GRAM for it
So I wait, so I wait
(Darling said: Don't quit your day job.)
I noticed that William Rowan Hamilton had the same cadence as Alexander Hamilton so I assumed that someone must have used that for a parody, and I was right
But
Listen to the son. They mention the following::
Kinetics, Quaternions (This is mentioned the most), An optimization view of light, Minimal action,
`your energy function generates the flow of time' ,Operators that Lie Commute with the symbol that bears your name, His versors(?) formed hyperspheres - see if you can plot 'em- invented vectors and scalars for when you dot 'em (Did he really invent vectors? Wikipedia says that in a sense he invented cross and dot products.) And Schrodinger sings that he adapted Hamilton's work for Quantum (I didn't know Schrodinger could sing!).
What do they NOT mention: Hamiltonian paths or circuits. His Wikipedia page does mention Hamiltonian circuits, but not much and you would have no idea they were important.
When a computer science theorists hears `Hamiltonian' she prob thinks `path' or `circuit' and not `an optimization view of light' or anything else in physics' She might think of Quaternions and if she does Quantum Computing she may very well think of some of the items above. But these are exception. She would likely think of the graph problems.
The rest of the world would think of the list above (or would think William Rowan Hamilton died in a dual over Quaternions and later had the best Hamilton Satire written about him- the second of course being the one about Batman,)
In his own time he was best know for Physics. Maybe also quaternions. I think Hamilton himself would be surprised that this problem became important. So here is my question:
When did the problem become important? Before NP-Completness or after?
Is he still better known for his physics and quats- I think yes.
When I say `Hamilton' what comes to YOUR mind?
Thursday, April 06, 2017
A Bridge Too Far
In Atlanta last week a fire destroyed a major highway bridge right on my, and so many other's, commutes. I've been playing with different strategies, like coming in later or even working at home when I can, not so easy when a department chair. I expect at Georgia Tech, just South of the damaged highway, we'll see less people around for the next ten weeks or so.
Even before the bridge collapse faculty don't all come in every day. In the Chronicle last month Deborah Fitzgerald laments the empty hallways she sees in her department. Hallways became a victim of technology, particularly the Internet. We mostly communicate electronically, can access our files and academic papers on our laptops and iPads just as easily in a coffeehouse as in our office. If you use your mobile phone as your primary number the person calling you won't even know if you are in the office. The only reason to come into the office is to teach or to meet other people.
Of course meeting other people is a very good reason. Not only scheduled meeting with students but the random meeting with another colleague that turns into a research project. The times I've walked into a student's office with a crazy idea, or needed a combinatorial theorem from one of the local experts. As we even move our meetings to video conferences, we really start to lose those spontaneous connections that come from random conversations. Soon the technology may get so good that our online meetings and courses will become a better experience than meeting in person. What will happen to the universities then?
Tuesday, April 04, 2017
Proving Langs not Regular using Comm Complexity
(My notes on this are at my course website: here They are notes for my ugrad students so they may be longer and more detailed than you want.)
Using that { (x,y) : x=y, both of length n} has Communication Complexity \ge n+1 one can easily prove:
a) The Language \{ xx : x\in \Sigma^*} is NOT regular
b) For all n the language \{ xx : x \in \Sigma^n }, which is regular, requires a DFA on 2^{n+1} states.
I also used Comm Complexity to show that
{ w : the number of a's in w is a square} is not regular, from which one can get
{ a^{n^2} : n\in N} is not regular.
More generally, if A is any set such that there are arb large gaps in A, the set
{ w : the number of a's in w is in A} and {a^n : n \in A} are not regular.
This approach HAS TO BE KNOWN and in fact it IS- Ian Glaister and Jeffrey Shallit had a paper in 1996 that gave lower bounds on the size of NFA's using ideas from Comm Complexity (see here). They present their technique as a way to get lower bounds on the size of NFA's; however, their techniques can easily be adapted to get all of the results I have, with similar proofs to what I have.
(Jeffrey Shallit, in the comments, pointed me to an article that predates him that had similar ideas:here.)
(Added later- another early referene on applying comm comp to proving langs not regular is Communication Complexity. Advances in Computers Vol 44 Pages 331-360 (1997),
section 3.1, by Eyal Kushlevitz. (See here)
(Jeffrey Shallit, in the comments, pointed me to an article that predates him that had similar ideas:here.)
(Added later- another early referene on applying comm comp to proving langs not regular is Communication Complexity. Advances in Computers Vol 44 Pages 331-360 (1997),
section 3.1, by Eyal Kushlevitz. (See here)
Next time you teach Automata theory you may want to teach showing langs are NOT regular using Comm Complexity. Its a nice technique that also leads to lower bounds on the number of states for DFA's and NFA's.
Saturday, April 01, 2017
William Rowan Hamilton- The Musical!
With the success of Hamilton,the musical on broadway (for all of the songs and the lyrics to them see here- I wonder who would buy the CD since its here for free) Lin-Manuel Miranda looked around for other famous figures he could make a musical about. Per chance I know Lin's college roommates father and I suggested to him, more as a joke, that Lin-Manuel could make a musical about
William Rowan Hamilton
Well, Lin-Manuel heard about this and noticed that
William Rowan Hamilton
has the exact same number of syllabus as
Alexander Hamilton.
Hence some of the songs would be able to have the same cadence. He has gone ahead with the project! He has asked that I beta test the first song by posting it, so I will:
William Rowan Hamilton
Lin-Manuel will be reading the comments to this blog- so please leave constructive comments about the song and the idea.
Subscribe to:
Posts (Atom)