My Blog Posts, in Reverse Chronological Order
subscribe via RSS
If you’re interested in taking a free online course, consider Coursera. It takes seconds to make an account and filter through the 700 or so classes currently in the database to find what interests you. Classes are generally affiliated with a university, and professors are often the ones lecturing in the videos online. In addition to video lectures, there are homework assignments and exams, which are submitted electronically, as well as user discussion forums where the students can discuss class concepts.
Coursera embodies the concept of the massive open online course (MOOC) which aims to have unlimited participation to allow (theoretically) anyone in the world to obtain an education for free. Founded in 2012 by Daphne Koller and Andrew Ng of Stanford University, Coursera now has over 7 million users and sports an impressive list of university partners. (Check out this paper for an interesting discussion about MOOCs.)
Coursera is similar to the well-known MIT OpenCourseWare, but it has several advantages. The biggest one is that courses on Coursera will have all class material eventually available to the students who sign up, whereas on MIT OpenCourseWare, you face the repeated problem of lack of video lectures, lack of exams and solutions, and other information, especially with the upper-level courses. Coursera’s website design is also vastly superior. On the other hand, Coursera classes requires the user to sign up in a certain date range, so if you go on Coursera right now, chances are high that some of the classes that you want to take aren’t offered in the near future (and you might have to add it to your “watch list” for the next session).
In the meantime, I’ve been checking out Andrew Ng’s machine learning class, which was what really started Coursera. It’s designed to be a ten-week course, with the following syllabus:
- Week 1: Introduction, Linear Algebra Review, Linear Regression with One Variable
- Week 2: Linear Regression with Multiple Variables
- Week 3: Logistic Regression and Regularization
- Week 4: Neural Networks (Representation)
- Week 5: Neural Networks (Learning)
- Week 6: Applying Machine Learning Algorithms
- Week 7: Support Vector Machines
- Week 8: Clustering, Dimensionality Reduction
- Week 9: Anomaly Detection, Recommender Systems
- Week 10: Large-Scale Machine Learning
A third of the grade is based on multiple-choice quizzes, and the rest is determined by programming assignments, to be done in MATLAB or Octave, the latter of which is an excellent free version of the former. Octave is one of the simplest programming languages out there, so it shouldn’t be too difficult for one to get used to it.
After going through the first few weeks of the course, here are some quick impressions:
- Advantages: The class doesn’t have many prerequisites (no calculus, no probability, etc.) and is accessible to a broad audience. Professor Ng’s video lectures are excellent. In fact, it’s nice to see that someone who can write complicated papers can clearly explain the basics. There seems to be a lot of collaboration among the students. The class covers most of the concepts I’d expect in a machine learning class, but for some reason doesn’t seem to cover the naive bayes and decision tree learning algorithms.
- Disadvantages: The simplicity of the class is also its major drawback — to someone like me who already knows machine learning, the class is too easy and I watch video lectures (for review purposes) at 1.5x or 1.75x the speed (a nice feature, by the way). Professor Ng often has to say “the discussion of this concept is beyond the scope of this course….” Consequently, a student at Stanford is better off taking Professor Ng’s “actual” machine learning course.
Again, if you’re interested in learning more about any subject, I encourage you to check out Coursera. There’s definitely a heavy focus on computer science — not surprising, given that the founders are computer science professors — but there are courses in subjects as diverse as health, law, engineering, and music.
Cholesterol, Saturated Fat, Grains, Meat, and Other Diet Controversies: why are There so Many People Challenging Conventional Wisdom?
As I mentioned in my recent post introducing Mark’s Daily Apple, I have become more interested in understanding diet, nutrition, and health. Sadly, this doesn’t come without challenges, and in this post, I’d like to discuss some of the current controversies that make it difficult for me to decide what to eat in order to maintain a healthy life.
First, let me provide some background. During elementary and middle school, I learned about the United States Department of Agriculture’s infamous food pyramid. Of course, like most Americans, I didn’t adhere to it exactly, but I at least kept it in mind, so it did impact the way I ate for most of my life.
After reading Fast Food Nation, I also avoided most forms of fast food starting in high school. On the surface, this diet approach seems to be excellent — just follow the food pyramid and avoid McDonald’s. Unfortunately, up until now, I had been unaware of the vast amount of misinformation, politics, and shoddy science of food that plague the country and are likely correlated with the shocking prevalence of obesity, heart disease, diabetes, and other chronic illnesses.
First, let’s go over a few hopefully non-controversial facts:
- In 2012, 34.9% of U.S. adults and 16.9% of U.S. youth were obese; in the early 1960s, obesity among U.S. adults was estimated at 13.4%. If we expand the pool of people to include those who are overweight (i.e., BMI of at least 25) then the percentage of overweight adults from 1962 to 2010 rose from less than 50% to more than 70%.
- Heart disease is now the leading cause of death for Americans, with the latest estimates pegging it in causing one out of every four deaths.
- In 2012, about 29.1 million Americans had diabetes, and almost two million new cases are diagnosed annually.
- The global pharmaceutical industry is expected to rake in $1,200 billion by 2016, of which the U.S. has the largest share.
- The estimated medicate costs of obesity in the U.S. is currently almost $150 billion.
I could go on and on, but I think the point is clear: the United States has a health crisis, and I’m pretty confident that we don’t need too many drugs to help us out. The human species, after all, cannot evolve so quickly over two generations to produce a much heavier population.
Fortunately, the rise in obesity and chronic illnesses has not gone unnoticed. The American Heart Association has established the following dietary recommendations to reduce the risk of heart disease:
- Choose lean meats and poultry without skin, saturated fat, and trans fat
- Eat fish at least twice a week
- Select low-fat or no-fat dairy products
- Cut back on food with partially hydrogenated vegetable oils (i.e., trans fat)
- Reduce consumption of saturated fat to lower cholesterol
- Avoid sugary beverages
- Prepare food without using too much salt to lower blood pressure
- Drink alcohol in moderation
- Pay attention to portion sizes
This is what I will refer to as conventional wisdom, and nowadays, the public perception is that diets high in fat result in weight gain, which subsequently leads to a whole host of other problems. Meanwhile, a healthy diet is low-fat, low-cholesterol, low-sodium, and plant-based (including grains). An example is the Ornish diet, which claims to reduce the incidence of heart disease by requring that no more than 10 percent of calories come from fat. It was created by Dr. Dean Ornish, Clinical Professor of Medicine at the University of California, San Francisco, whose books and nutrition program have won him widespread acclaim. After all, not everyone can say that they serve as a health advisor to Bill Clinton.
Unfortunately, the past few decades has suggested a paradox. Indeed, spurred by the government war on fat, Americans have been consuming less fat in recent decades, as shown by that 1998 paper (by the way, please let me know if you find a more recent reference). In 1965, the estimated daily intake of fat for American men and women was 139 grams and 83 grams, respectively, and in 1995, those figures were 101 grams and 65 grams, respectively. Furthermore, there is a consistent decrease in the percent of daily calories from fat, from 45% in 1965 to 34% in 1995. The caveat here is that the total caloric consumption of Americans has also increased, which might mitigate the “positive” effect of lowering fat, but if so, shouldn’t the rise in chronic illnesses be blamed to whatever else we’re eating to get those calories?
In addition, the percentage of American adult smokers dropped from 42.4% in 1965 to 19.0% in 2011. So something must be counteracting this beneficial effect because all signs point to an increase in chronic illnesses. Also, while Americans are living longer, that doesn’t mean our final years are that great. Our extended lifespan is largely due to better medical treatment that wasn’t available in earlier eras, and not due to an improved diet.
Good Calories, Bad Calories
In Good Calories, Bad Calories, Gary Taubes argues that United States government and health organizations have given us dietary advice that contradict the science. Taubes starts by explaining some of the earliest observational studies of the health of native populations before and after dietary changes (i.e., “Westernization” of diet). He then moves his way towards the mid-1900s, which coincided with the prominence of Ancel Keys and a new era of dietary advice that encouraged consumption of carbohydrates (including “white” bread/rice/cereal/pasta) and demonized saturated fat as the cause of heart disease. According to Taubes, the science showed, and continues to do so today, that saturated fat has little correlation with heart disease while the link between refined carbohydrates, sugars, and chronic illnesses is much stronger.
Good Calories, Bad Calories is ultimately a brutal attack on conventional wisdom. (By the way, I would like to point out that there’s a fair amount of support for following a vegan diet to obtain optimal health, so these people are also technically challenging conventional wisdom, but for the purposes of this post, as you have probably determined, I am mainly going to discuss the low-carb paradigm.)
In addition to what I mentioned earlier about the AHA and the Dean Ornish diet, conventional wisdom also proclaims that people can obtain a healthy body weight by “eating less and exercising more.” But this is problematic, Taubes says, because exercising more tends to cause an increase in appetite. For instance, athletes are known to require more calories than the average sedentary person.
When considering the totality of conventional wisdom, here is one of my attempts to sum up Taubes’ advice in one sentence:
“To be healthy, be sure to eat the right kind of calories from unprocessed meats and vegetables and avoid refined carbohydrates and sugars (including whole grains and processed foods); good exercise, while beneficial, cannot make up for a terrible diet.”
What is my opinion on Good Calories, Bad Calories? I have mixed feelings. Taubes is right on diet in many respects, but I’m not sure if people should be switching to a meat-heavy diet, which is what Taubes appears to advocate (but he never explicitly says so). Before reading the book, I was already aware of how saturated fat and cholesterol probably aren’t as bad as we (by that, I mean “conventional wisdom thinkers”) think they are. I had read Chris Kresser, Mark Sisson, and Zoe Harcombe, among others, give their take on cholesterol and similar topics, but I still tried to read the book with an open mind and a healthy level of skepticism, as none of the three people I just mentioned are true medical researchers. Good Calories, Bad Calories is definitely on the dense side in terms of writing style, but a lot of his analysis makes sense, and I have to say that the history of nutrition science and advice is interesting. My conclusion is that I think that anyone who has serious interest in diet and health should take a look at this book. It’s dense and has sixty six (!) pages of references, which is necessary to ensure that, as Taubes would later say, “to never take what I say on trust alone.” In fact, Taubes is not a nutritionist, but a science journalist; he got a Bachelor’s degree in physics from Harvard University.
Please don’t interpret the preceding paragraph as a full-on support of Taubes. I knew after reading this book that I needed to see if there was any legitimate criticism that would make me second-guess his advice. And by far, by far, the best review I’ve found of Good Calories, Bad Calories is a series of blog posts by a nutrition guy named Seth. And … wow, if Taubes relentlessly criticized the government in his book, Seth takes that kind of criticism, multiples it by ten, and levels it back at Taubes! It’s definitely a good read just to make sure that you don’t get too trapped into the whole “low-carb” ideology.
Just to be clear, Taubes doesn’t disagree with all of conventional wisdom. No one is out there advocating that Coke and Pepsi belong in a healthy diet, and that we should eschew non-starchy vegetables. The controversy is on the role that saturated fat, cholesterol, grains, and meat play in a healthy diet. Taubes never gives specific dietary advice, but his reader-friendly version (Why We Get Fat) does give a diet plan, which allows mainly unprocessed meats and non-starchy vegetables. He also has a cholesterol blog post, where he boasts about his cholesterol numbers while describing his diet as “eggs, sausage, cheese, cheeseburgers (no bun), steaks … high in fat, low in carbohydrates.”
Now, I know that the obvious reaction is to dismiss Taubes as a outsider to nutrition who doesn’t know what he’s talking about. But let’s start looking at some other sources that clearly have credibility.
Here’s one: the American Diabetes Association (ADA). They have to have some dietary advice, right? Good Calories, Bad Calories goes at length to explain how refined carbohydrates and sugars can induce diabetes, so it will be interesting to see how Taubes’ ideas match up.
I went to their page on Grains and Starchy Vegetables, where I saw this:
There is no end in sight to the debate as to whether grains help you lose weight, or if they promote weight gain. Even more importantly, do they help or hinder blood glucose management?
One thing is for sure. If you are going to eat grain foods, pick the ones that are the most nutritious. Choose whole grains. Whole grains are rich in vitamins, minerals, phytochemicals and fiber.
Reading labels is essential for this food group to make sure you are making the best choices.
Every time you choose to eat a starchy food, make it count! Leave the processed white flour-based products, especially the ones with added sugar, on the shelves or use them only for special occasion treats.
With that, the ADA just made me worried. I have to be honest: if I had diabetes, how could I feel comfortable eating grains — even whole grains — if the ADA can’t take a definitive stance on this, and suggests the possibility of “promote weight gain” and “hinder blood glucose management” as side effects? They do gently suggest eating whole grains on other parts of their website … but why not here, on the page that actually discusses it?
Here’s a second source that should be credible: the Harvard School of Public Health Nutrition Guidelines. But unfortunately, I remain worried. Here’s what they have to say in their 2011 article:
Nearly two decades ago, the U.S. Department of Agriculture (USDA) created a powerful icon: the Food Guide Pyramid. This simple illustration conveyed in a flash what the USDA said were the elements of a healthy diet. The Pyramid was taught in schools, appeared in countless media articles and brochures, and was plastered on cereal boxes and food labels.
Tragically, the information embodied in this pyramid didn’t point the way to healthy eating. Why not? Its blueprint was based on shaky scientific evidence, and it barely changed over the years to reflect major advances in our understanding of the connection between diet and health.
Wait … “shaky scientific evidence?” That’s not what I want to hear! Could Taubes be onto something after all? They continue their description by criticizing how current guidelines don’t penalize white/refined grains enough (so this is a point for Taubes), don’t penalize red meat enough (so this is a point against Taubes) and recommend too much dairy (I don’t think Taubes talks too much about dairy, so I’ll consider this a wash). The Harvard pyramid and associated article is an interesting read, and makes me feel much better that I base my diet on vegetables and have never been a huge milk drinker despite how doctors and others told me to drink more milk when I was young (thank goodness I didn’t!).
So, from the Harvard guidelines, whole grains play a foundational role in a healthy diet, but refined grains are almost as bad as you can get in terms of food! Is this the right answer? Are whole grains really that much better? Well, I obviously don’t know the answer. I’ll have to be honest, I’m leaning towards supporting the Harvard pyramid, but again, there are many experts who would disagree; Dean Ornish, for instance, would oppose the allowance of egg yolks, which he classifies as among the worst foods to eat. And he advised Bill Clinton on his diet! But here’s something else that’s interesting … Bill Clinton also has a second dietician, Mark Hyman, who has encouraged Clinton to eat more fat! Thus, Clinton has two major medical minds giving him polarizing advice on the amount of fat (and eggs) to eat. As the article suggests, if Clinton can’t come to a consensus, what hope is there for the rest of us? The best-case scenario, of course, is if both diets are great. And they obviously are, when the baseline is a diet of McDonald’s and Pizza Hut. But how do they compare against each other? That’s the major question.
Going Against Conventional Wisdom
Given that there are some aspects of diet that don’t seem to have a consensus, I thought I’d just try and see what people are saying. My goal is to synthesize some of the well-known books that advocate at least one of the prevalent themes from Good Calories, Bad Calories, which among them are “fats are fine, refined carbohydrates and sugars far worse,” “low-carb living,” and “governmental impact on nutritional science.” Most or all of these books will substantially challenge conventional wisdom.
This in no way means I support these arguments — the purpose of seeing all these books is that it raises doubt about conventional wisdom. While it does sound disconcerting, I ultimately think it’s best if we know all this information, because then we can do our own independent research and can make informed decisions. And again, a healthy level of skepticism (but not too much) is needed as part of science, and this is what makes nutrition science so great — it’s generally accessible to people who are outsiders, at least way more than computer science.
I tried to ignore books that were diet or recipe-related (there are a lot of “Paleo Recipe” books out there) in favor of ones that take at least a scientific approach to nutrition by citing studies and forming logical arguments. The one exception might be the Atkins diet book, but that was the one that really started the whole low-carb movement, and I think Atkins (who was a cardiologist) did have some science to back him up beyond his trials on himself and his co-workers.
The books are ordered by their original publication date, though many have been updated at least once. A number of them, such as Taubes’ books, are New York Times or National bestsellers. You’ll also notice that the majority of them are from the recent decade. This is not surprising; Taubes explicitly mentions at the end of Good Calories, Bad Calories, that the Internet was the main reason why he was able to find all the sources he did.
So, here’s a list of books I found:
- Pure, White, and Deadly: How Sugar is Killing Us, and What We Can Do to Stop It, by John Yudkin. (1972, updated in 1986 and 2012)
- Dr. Atkins’ Diet Revolution, by Robert Atkins. (1972, updated 2009)
- Protein Power, by Michael and Mary Dan Eades. (1997)
- The Great Cholesterol Con, by Anthony Colop. (2006)
- Good Calories, Bad Calories: Fats, Carbs, and the Controversial Science of Diet and Health, by Gary Taubes. (2007)
- The Great Cholesterol Con: The Truth About What Really Causes Heart Disease and How to Avoid It, by Malcolm Kendrick. (2008)
- The Primal Blueprint: Reprogram Your Genes for Effortless Weight Loss, Vibrant Health, and Boundless Energy, by Mark Sisson. (2009, updated 2013)
- The Paleo Solution: The Original Human Diet, by Robb Wolf. (2010)
- Wheat Belly: Lose the Wheat, Lose the Weight, and Find Your Path Back to Health, by William Davis. (2011, updated 2014)
- Why We Get Fat: And What to Do About It, by Gary Taubes. (2011)
- The Art and Science of Low Carbohydrate Living, by Stephen Pinney and Jeff Volek. (2011)
- The Art and Science of Low Carbohydrate Performance, by Stephen Pinney and Jeff Volek. (2011)
- The Great Cholesterol Myth: Why Lowering Your Cholesterol Won’t Prevent Heart Disease — and the Statin-Free Plan That Will, by Johnny Bowden and Stephen Sinatra (2012)
- Fat Chance: Beating the Odds Against Sugar, Processed Foods, Obseity, and Disease, by Robert Lustig. (2012)
- Grain Brain: The Surprising Truth about Wheat, Carbs, and Sugar — Your Brain’s Silent Killers, by David Perlmutter. (2013)
- Death by Food Pyramid: How Shoddy Science, Sketchy Politics, and Shady Special Interests Have Ruined Our Health, by Denise Minger. (2013)
- Eat the Yolks, by Liz Wolfe. (2014)
- The Big Fat Surprise: Why Butter, Meat, and Cheese Belong in a Healthy Diet, by Nina Teicholz. (2014)
- Keto Clarity: Your Definitive Guide to the Benefits of a Low-Carb, High-Fat Diet, by Eric Westman and Jimmy Moore (coming soon!)
Wow. That’s a lot of books, and honestly, it didn’t take me a long time to find these. And as you can see by looking at the authors, we’re seeing more and more people with medical degrees support the general idea of following a low-carb diet.
One major problem with these books, and indeed, this entire low-carb argument, is the “cherry-picking” involved, when authors look at research studies and selectively choose the ones that fit their hypothesis while ignoring other studies that don’t. Then again, who’s to say that the same isn’t happening for people who advocate eating lots of starch? And even books that take a scientific-based approach to suggesting a high-starch or low-meat diet, such as The China Study, have their own critics. I haven’t read The China Study, but it’s on my agenda.
By the way, I should mention that the author of The China Study, T. Colin Campbell, is Professor Emeritus of Nutritional Biochemistry at Cornell University. He advocates a low-fat, vegan diet, but that’s not what the Harvard food pyramid implies. In fact, even Dean Ornish doesn’t advocate a vegan diet. Look at his spectrum of food choices again, and you’ll see egg whites and fat free milk in his “Group 1: Most Healthful” foods. (And, uh … beer is also in the “Most Healthful” category. That’s interesting … I don’t know how that ended up on there.)
If there is this much out there that seems to contradict what the American Heart Assocation and conventional wisdom dictate, then doesn’t this at least raise some doubt? (Don’t forget to also consider books that suggest a vegan diet.)
I guess my point is that if we are getting advice about nutrition from the government, it should be established beyond a reasonable doubt. Otherwise, I would rather see phrases like “we are still considering all evidence and are unsure about this….”
Dr. Yudkin and Sugar
Unfortunately, one of the biggest takeaways that I got from Good Calories, Bad Calories and my own brief research is that, after fat became demonized, people and industry looked for alternatives, and they found that in sugar. In fact, Taubes said that prominent nutritionists and professors at elite universities were recommending sugar, and that it was a safe alternative to fat, even as late as the 1980s. But Dr. John Yudkin disagreed, and published Pure, White, and Deadly: How Sugar is Killing us, and what we can do to Stop it. So perhaps sugar and some forms of fat (trans, saturated?) are bad for us, but sugar is a stronger risk?
According to a 2014 article on The Telegraph, Ancel Keys and the sugar industry ruthlessly attacked Dr. Yudkin, hindering his credibility. But as the past few decades has witnessed, Dr. Yudkin may have been right all along, at least with respect to the dangers of using sugar as a substitute for fat. I’m not confidently sure which one of sugar or saturated fat is more of a risk factor for chronic illnesses — they probably both are — but if you told me to pick the greater risk factor right now, I would say sugar. The World Health Organization — a credible source, I would like to add — has recommended lowering sugar intake, but is expecting a battle with the sugar industry, as highlighted by a 2014 Nature.com article. (I highlight these dates to show how recent this whole anti-sugar movement began.) In addition, Dr. Yudkin’s work was recently revived thanks to the efforts of Robert H. Lustig, a professor at the University of California, San Francisco medical school. You can see Dr. Lustig in his YouTube video, Sugar: The Bitter Truth. Dr. Yudkin’s book was then re-published in 2012 due to growing demand, which gives an idea of how it has stood the test of time.
Summers are nice because they offer me a break from an intense academic environment. As a result, I’ve had the chance to explore other fields that interest me, and one of them concerns the human diet. Simply put, I’m trying to figure out what I should eat in order to maintain a healthy life.
The Original Food Pyramid
There’s a lot of information available that we can use for diet advice. For instance, the United States Department of Agriculture has their famous (or infamous, as I’ll get to shortly) 1992 food pyramid:
Let’s suppose we use this as a guide to optimal health, which seems reasonable because it’s from a United States government organization. (It shouldn’t be, because that pyramid has already been scrapped in favor of new dietary guidelines, but it’s good to discuss it to see the historical perspective on food.)
Unfortunately, even without consulting outside sources, I can already see several problems:
- It makes no distinction between whole or minimally processed foods and heavily processed foods. (I’ll throw in whole grains in the “minimally processed foods” category.) The former group includes fruits, vegetables, and animal products obtained from their natural state. The latter group would include pizza, chemically-laden meats, and so on.
- It suggests consuming fats, oils, and sweets sparingly, but the dairy and protein groups already include substantial amounts of fat. And my understanding is that fat has long been essential for human health. Our early ancestors ate lots of plants, but they would also eat the complete carcass of animals, including fat-dense organs that we shun today.
- It suggests that the serving counts should not be exceeded, which might impose unnecessary restrictions. Consider my situation: I love eating huge salads, and I also have a habit of downing an entire bag of baby carrots as an afternoon snack. This means that I easily rack up 7-10 servings of vegetables daily (depending on how you define a serving), but according to this pyramid, I shouldn’t be eating so many vegetables.
I know that no pyramid can disseminate detailed information in such a small amount of space, but such simple modifications could go a long way.
This brings me to the next part of this post.
Mark’s Daily Apple
My quest for learning more about health, diet, nutrition, and food led me to Mark’s Daily Apple. It’s an extensive blog written by Mark Sisson, a well-known advocate of eating the Paleo diet (though he calls it “Primal”) and preventing chronic diseases of civilization (e.g., diabetes and heart disease) by lifestyle choices. I didn’t think much of this at first, but the more I thought about the food I ate, the more I kept coming back to his blog. It also didn’t hurt that he’s another Williams alum, which might have piqued my curiosity.
Mark Sisson advocates his own food pyramid, which emphasizes meats (including fish, eggs, and fowl), vegetables, fats, fruits, and some carbohydrates. Notice the distinct lack of bread, rice, cereal, and pasta! It’s a long story about why he excludes them, but Sisson explains this in his blog and has some decent (but in my opinion, not overwhelming) evidence to back up his claims. For the most part, I favor his food pyramid over the 1992 USDA food pyramid, but I think Sisson’s pyramid should have kept vegetables as the “base” group to reiterate how they should compose the bulk of the diet in terms of volume.
If you want more information about his philosophy towards food and life, I’ll refer you to his Start Here page. When reading Mark’s Daily Apple, realize that Mark Sisson’s focus is not just on nutrition, but indeed, on a lifestyle. His advice encompasses sleep, play, exercise, and many other factors that affect our health. There’s so much out there that Mark Sisson posts new entries daily and still has no shortage of topics to talk about.
In a future blog post, I’ll delve more deeply into diet controversies. (Don’t worry — this digression doesn’t mean that I’m turning into a nutritionist…)
About a year and a half ago, I wrote a blog entry about deaf computer science Ph.D.s. I recently revisited this topic and found out that I missed a few people from my earlier list, so this post is a continuation of my previous one. Here are the new Ph.D.s:
- Vinton Cerf (Ph.D., University of California, Los Angeles, 1972) is hard-of-hearing, though he’s on the board of trustees at Gallaudet University.
- Daniel Berry (Ph.D., Brown University, 1974).
- Dimitri Kanevsky (Ph.D., Moscow State University, in the 1970s), though again, if we’re picky with our criteria, he actually got his Ph.D. in math.
These three people are all established scientists with impressive resumes, even if two of them don’t fit clearly in the mold of a “deaf” person.
Vinton Cerf is known as one of the “fathers of the Internet,” which should say a lot about his contributions to computer science. For instance, he helped to form the Internet Corporation for Assigned Names and Numbers (ICANN), which manages the global domain naming system (its headquarters is in the U.S. … I’m not sure how other countries feel about that). Not surprisingly, Dr. Cerf has an array of awards and honors, including the Turing Award, the Presidential Medal of Freedom, and the National Medal of Technology. In 1997, he joined the board of trustees at Gallaudet University. Nowadays, he works at Google.
Daniel Berry has been deaf in both ears since birth, and can only use a hearing aid in one ear. He does not sign because his parents spoke English and he picked up lipreading, which may be one reason why I didn’t know of him until now (and he mentions on his website that he doesn’t have many hearing impaired acquaintances). In terms of academics, he got his computer science Ph.D. from Brown back in 1974. He then joined the faculty at UCLA from 1972 to 1987, The Israel Institute of Technology from 1987 to 1998, and then at the University of Waterloo from 1988 to now. Note that all three of these schools have outstanding computer science departments, which should give you an idea of his research ability. Professor Berry provides a brief paper on his website that describes his background in more detail, as well as his recommendations for making the web more accessible.
Dimitri Kanevsky got his Ph.D. in Russia in the 1970s (I can’t find the exact date) and held research positions at the Max Planck Institute in Germany and the Institute for Advanced Study in Princeton. Armed with mathematics background but a desire to make practical use of it, he joined IBM in 1986, where he still works today. Dr. Kanevsky specializes in speech recognition, so there’s a good chance that any recognition software today traces its origin to him. Dr. Kanevsky’s resume includes being named an IBM Master and more than 15o U.S. patents.
Unfortunately, I’ve never personally met any of these people, but it would be nice to do so someday.
In deaf-related news, I just found that Carol Padden, professor of communication at the University of California, San Diego, has been named dean of the social sciences at her school. Her tenure will start in October.
I didn’t know her before this announcement, so it was interesting to read about her background. She was born as the second deaf child of two deaf faculty members at Gallaudet University. She started her education in a deaf school and then became a mainstreamed student at a public school system.
After graduating from Georgetown in 1978 with a Bachelor of Science in Linguistics, she went to UC San Diego to pursue a PhD in Linguistics, which she obtained in 1983. She has since been on the faculty (and, of course, will be a dean) at the same school, specializing in the study of American Sign Language. Professor Padden seems particularly interested about understanding the variety of sign languages that are developing throughout the world. For instance, what properties of sign languages develop after only one or two generations of use? Which ones require more time to evolve? While I can’t judge her research, it must be top-tier, since she was a MacArthur Fellow in 2010.
I’d been meaning to post this earlier, but I got sidetracked by Project Euler (more on that later). In any case, I want to list a few computer science textbooks that, in my opinion, are written in a friendly style and are easy for someone to read like a novel. These are my favorite kind of textbooks, because they often incorporate two important aspects: elaboration and examples. Notice that this does not mean that mastering the corresponding subject is easy! It just makes it easier for an experienced and educated reader to do so.
I’m inspired to think about this because, as much as I enjoyed my complex analysis class last fall, the textbook we used glosses over so many details that it made analyzing some of the proofs excruciatingly difficult. At least, for me … I can’t speak for everyone in the class, but my professor did have to explain that the goal of his lectures was to emphasize why the authors/book did something in their proofs.
In the past few weeks, I read parts of Methods of Mathematical Economics, a textbook about the mathematics behind linear programming and other popular applied mathematics techniques. It is written in a conversational style (the author uses “I” instead of “we”), and it was very helpful to me for a final project.
Here are four books on the computer science side that I’ve found to be very readable.
- Algorithm Design, by Jon Kleinberg, and Eva Tardos, presents an introduction to the common themes underlying an algorithms course. I enjoyed it because it emphasizes the decision-making behind many of the proofs. In addition, it contains several sample problems with detailed solutions. These solutions also explain why certain approaches might not work or are suboptimal.
- Artificial Intelligence: A Modern Approach, by Stuart Russell and Peter Norvig, is a surprisingly readable “encyclopedia-like” book about AI. I do not recommend reading the entire thing, especially in one sitting! But if you pick out a single chapter, the book should serve you well. I talked with Stuart when I was visiting Berkeley, and he told me I needed to know more learning theory. I asked him how I could learn more, and he said: “read the book.” Good — I’ll do that!
- Distributed Systems: Principles and Paradigms, by Andrew S. Tanenbaum and Maarten Van Steen, is about concepts of distributed systems (i.e., those relying on multiple computers/machines). This book is filled with examples. Almost every concept is explained with an immediate real-life example.
- Introduction to the Theory of Computation, by Michael Sipser, overlaps somewhat with Algorithm Design, but emphasizes automata, computability, and complexity, rather than pure algorithms. It is a concise book, but somehow provides the impression that it’s detailed and expansive. Fortunately, each chapter contains problems with full solutions.
I’d be interested in knowing if there are other popular, readable computer science textbooks.
After finding out I made it into the Williams College Phi Beta Kappa and Sigma Xi honor societies, I went online looking for more information about the induction ceremonies. Unlike the commencement exercises, there does not seem to be much out there about these two ceremonies, so I figured I might explain what goes on during these events.
Just a quick background: Phi Beta Kappa and Sigma Xi are two honor societies that some colleges and universities offer to their students. Phi Beta Kappa tends to admit students based on GPA and breadth of coursework, while Sigma Xi admits those who have done and plan to continue doing science and math research. (At Williams College, Phi Beta Kappa is solely based off of GPA; roughly the top 12.5% of the class gets in.) In the Williams College class of 2014, I believe there were 68 Phi Beta Kappa inductees and 56 Sigma Xi inductees. Of course, there was significant overlap among the two groups.
Phi Beta Kappa
We had several speakers in this ceremony. Some provided administrative information, such as “welcome to the society” and “please do X to complete your application” type of stuff.
To me, the most notable speaker was the Williams professor who introduced us to the history of the Williams College Phi Beta Kappa charter. It started in 1776 at the College of William and Mary (the college that, according to the speaker, our grandmothers still think we go to) and then spread to Massachusetts when Harvard acquired a charter in 1779. Williams tried numerous times to acquire one from Harvard, but did not succeed until 1864. The charter we obtained, written in Latin, was shown on display throughout the ceremony. Sadly, it appears to be the only one of its kind in which the granting institution’s name (“Harvard”) was written in much larger text than the name of the school being awarded (“Williams”)!
After the history talk, another guest speaker highlighted the diversity of the majors studied and languages spoken among the Phi Beta Kappa members. He listed almost every major and language, except for computer science and American Sign Language. I am not sure how those two didn’t make it on his lists. I mean … maybe if one of my two “diversity contributions” was listed, I wouldn’t have made a big deal out of it, but both?
Near the end of the ceremony, each of us went on stage to receive Phi Beta Kappa materials.
This induction ceremony was much shorter than the one for Phi Beta Kappa. The Sigma Xi speaker spoke briefly about how he wished us all an excellent career in science, and encouraged us to try and get promoted to full membership. All the Sigma Xi inductees stood up and got to the stage to get our materials, and then we stood there while all the parents and other family members took pictures. I noticed at least one of the old class pictures is online — if the 2014 one goes up, you should try and find me.
Here’s an interesting question to consider: which one of “he/she/someone uses ASL” or “he/she/someone speaks ASL” should you use in writing and conversation? The answer isn’t obvious to most people without knowledge of ASL, because while speak is the default term for most languages, it’s not clear if ASL falls under that category due to its visual nature (and, I might add, its no-speaking requirement).
I never had someone tell me which of the two phrases to use, and I’ve flip-flopped on my usage. I would say speak when I was younger, then in recent years I switched to saying use, and now I’m starting to think I should go back to speak. My recent shift is due to a discussion on the Deaf Academics mailing list.
Here is a video of someone arguing on behalf of speak, and here is the transcript (the meaning of s5s is explained in the video):
“Hi, I’m Michele Westfall and I currently write for ASL Rose. John Clark asked me to talk with you about Speak vs Sign, and I’m happy to do so. It’s no problem at all. Why? For the past couple of years, I’ve been encouraging Deaf people … really, everyone to use the word “speak” in relation to ASL. Why? I’ve been noticing that the hearing society frequently sees us as “silent.” Yes, they regard ASL as “beautiful,” but when it really counts and really comes down to it, we are still SILENT in their eyes. I’ve noticed that both hearing and Deaf people tend to say “ASL users”…and that bothers me.
Think about it: hearing people never say “English users”…”French users”…”Spanish users”…”Chinese users.” Really, the contrast is huge. We say “ASL users”, “we sign”, “we do ASL”….which serves to emphasize the difference between ASL and all other voiced languages and puts way too much emphasis on the Voice. Hearing people get to say “speak”…and we can’t.
I disagree with that. I say, YES, we speak ASL. Understand, I’m not saying we have to say “We speak [using 4-handshape on chin] ASL.” That’s wrong. What’s the right way? “We speak [using s-5-s handshape] ASL.” Get it? s5s = speak…that’s our version of speak. Don’t say 4-handshape on chin/speak. That’s the hearing version (or the hearing-minded version). We s5s/speak ASL. Or…”I’m an ASL s5s/person [speaker].” Or “The ASL s5s/person [speaker] for the day is XXXXX.”
Just an example. You s5s/speak ASL. I s5s/speak ASL. To let you know, I’ve been saying this for the past several years. I even write it…for example “I speak ASL.” on paper. In English sentence….I speak ASL/you speak ASL. And hearing people always accept it and never object. They don’t say, “Wait a minute! What’s this?” They accept it because it seems natural to them. I think…it’s time. We’ve been saying all along that ASL is a language. It means we must change our words to reflect that reality. What’s our reality? ASL is our language, and therefore…what? We speak ASL! We don’t sign ASL.
Saying “We sign ASL” gives way too much credit/weight to Voice. No. Enough. It’s time to bring ASL on equal footing with all other voiced languages. You speak ASL. I speak ASL. We don’t “do” ASL. We don’t “use” ASL. We speak ASL. We are ASL speakers. I hope this makes sense to you. If not, let me know. Bye bye.”
One critical skill I acquired this past year was how to use git. It’s a distributed version control system that is often used by programmers and software companies to manage their code. The reason why git and other version/revision control systems are so useful is that they allow multiple people to easily work together on a project by means of a shared code repository. Rather than have one person work on a small part of a project and email any updates to everyone else, he or she can instead “push” their modifications to the repository, and other collaborators can “pull” the updated code so that they are working with the latest version. (I use “push” and “pull” since these are git-related terms, but other systems have different terminologies for adding and getting repository code.)
If the current code has serious issues, it is possible to roll back to a previous version, which is just like having a bunch of backup files. This is one of the main benefits of using version control systems, even for a project managed by just one person. For instance, I used git for my undergraduate thesis and the related code, so whenever I wanted to work on it, I would just pull from the current repository, make my changes, and then push at the end. Git turned out to be a lifesaver when, two weeks before I had to send my thesis to the rest of the department, I accidentally deleted a chapter. The solution? Pull the latest version of that chapter from the repository.
Git is very common among computer science students, and possibly even more popular among computer
science Ph.D. students, so it’s a good skill to have in one’s toolkit. I learned git
largely by trial and error and having other people tell me what to do, but one resource I found that
might be useful to beginners is Git Immersion. This is a tutorial that walks you through the
basic commands of git. It was mostly review for me, but I did learn a few things, such as
mv which will save me time when moving files in my repositories. In the past, I would move files,
then delete the original files from git, then add the new ones (in the different location), then
commit everything, but git mv does all that at once. My one qualm is that I wish they incorporated
GitHub in the tutorial, but it’s technically not part of git, so I can see the reason for
Say, I wonder how non-computer scientists write their theses, or more accurately, those who don’t employ version control systems. I would hate having to save countless backups for a variety of files. In addition, another key benefit of git is that each time you make a change, you can record the high-level idea of this modification, and it will appear in the log for future reference. This makes it easy to go back and search for an old version of a file.
As my final college grades come in, I once again reflect back on my undergraduate classes and their grading schemes. The key question: how much did my grades correspond to the amount of material I learned or the amount of the subject I mastered? This is a tricky question to consider. Obviously, the amount of “mastery” required to get an A varies from school to school, subject to subject, and even course to course within the same field. But I believe that everyone can give a rough interpretation of how much he or she learned from a class (at least, right after it finishes). This may or may not correspond to the actual grade.
For the sake of completeness, here are four cases that can occur, supposing for simplicity that an A is the standard for excellence:
- You get an A, and feel like you deserved it.
- You get an A, but don’t feel like you deserved it.
- You get less than an A, and feel like you deserved it.
- You get less than an A, and don’t feel like you deserved it.
I have had all four of these cases happen at Williams. Case 4 is obviously bad, since everyone feels slighted when this happens. But Case 2 can arguably be just as worse in the long run, since you know less about the subject than what might be suggested from your transcript, but employers may not see that until after you’re on the job. In an ideal grading scheme, only cases 1 and 3 would occur.
So how can classes be designed to reduce instances of the two undesirable cases? I have two suggestions, but keep in mind that these are aimed at computer science and/or mathematics courses. I don’t have enough experience with other majors, though these might work for corresponding classes anyway.
Suggestion 1: Require Individual Work
One of the main observations I’ve made while at Williams is that sometimes it is possible to “hide” your weaknesses by joining a group and earning the group’s collective grade. For instance, this might involve a computer science group project where everyone in the group gets the same grade. In these cases, your grade is largely determined by who you work with!
Learning how to work in groups is certainly an important skill, so I’m not suggesting that these projects be eliminated entirely. Instead, I urge professors to divide up projects in two categories: those that allow groups and those that must be done individually. Or during a group project, perhaps require that everyone give a self-evaluation of their peers. This happened in my African Studies class in the Spring 2013 semester. (But this tactic runs into problems if you work with shady people … again, it matters who you work with!)
In a typical computer science course, grading is determined by a combination of group work, homework, and exams. For mathematics courses, they typically use only homework and exams. This brings me to my next suggestion…
Suggestion 2: Make Exam Score Ranges Larger
I think this suggestion will be more helpful than requiring individual work, and in any case exams are (I hope!) an example of something in that category. The problem that I have experienced in Williams classes is that exams are often set so that the vast majority of students (say, 85%) get scores in the 85-100 range. In a ten question exam where all questions are weighted equally, the first nine might be minor variations of homework problems, and only the last question gets used to differentiate between those who really know the material.
But this doesn’t give enough discrimination among students, and it means more students might get As because they lucked out on that tricky question, and more students might get Bs because they happened to make a careless error on one of the easier questions.
Increasing the exam score range so that the median and mean are within the 70-75 range would give professors more ability in distinguishing the different categories of students. With an “85-100 exam,” if I get a 94 and another person gets a 95, should I consider myself equal to that other person in terms of knowing the material? If I lucked out on that last question, that class might end up giving me an A, but I might view it as Case 2. But if scores were distributed over a 40-100 range, all of a sudden that 94 starts looking a lot better. And if I end up with, say, a 70 on an exam where lots of students get 90s, I’ll be momentarily disappointed, but the grade I get will reflect that I didn’t know enough of the material to warrant a higher grade, and that others were more deserving of getting an A.
I think a lot of students won’t like larger score ranges, but really, this shouldn’t be the case. Professors should assign grade ranges appropriately, so that scores in the 80-100 range would be an A, rather than the “standard” 90-100 range. All these numbers are really arbitrary, and in the real world, no one knows “100 percent” of their field/profession anyway.
PS: It’s good that Williams’ final exam period ends five days before the last day of May. Otherwise, I would have broken my streak of having at least one blog post a month…
For one of my final projects this semester, I’m working on something that’s loosely based off of the traveling salesman problem (TSP) and the P vs NP question, the latter of which is arguably the most famous open problem in all of computer science. Whoever proves that P = NP, P =/= NP, or that the problem cannot be proved/solved for whatever reason, gets a $1,000,000 prize from the Clay Mathematics Institute.
As part of my pre-project research, I’ve had the chance to look at a few papers about the TSP and the P vs NP question. Some papers have been great to skim, if only to gather insights about the TSP’s integer programming formulation, but others (typically those that claim to solve P vs NP) are clearly mockeries. In this post, I’d like to discuss the ways one can quickly tell that papers claiming to solve this question — or in fact, any other famous open problem — are bogus. Let’s do this under the assumption that you’re looking at a pre-print of the paper before it has been reviewed by a preeminent conference or journal.
For reference and amusement, here’s a page that lists (as of this writing) one hundred and five papers/proofs that (mostly) claim to solve this question. The author of this page states that only one of these papers has appeared in a peer-reviewed journal. And of course, it doesn’t solve the problem:
Note: The following paragraphs list many papers that try to contribute to the P-versus-NP question. Among all these papers, there is only a single paper that has appeared in a peer-reviewed journal, that has thoroughly been verified by the experts in the area, and whose correctness is accepted by the general research community: The paper by Mihalis Yannakakis. (And this paper does not settle the P-versus-NP question, but “just” shows that a certain approach to settling this question will never work out.)
Before we begin, let me just start with a disclaimer. Yes … this problem may be solved throughout my lifetime. So the title of this blog post may be changed to “How to Quickly Tell That All but One P = NP or P =/= NP Papers are Bogus” … but for now it is safe, as the Clay Mathematics Institute has yet to recognize a solution. Furthermore, by definition, if P = NP, then the papers claiming that P =/= NP are erroneous, and vice versa. So, under the assumption that (for whatever reason) you’re skimming a paper claiming to solve the P vs NP question (or another famous open problem), what are three quick signs that the paper is not worth reading?
Reason #1: The author isn’t a top theoretical computer scientist (or, if there is more than one author, then none are top theoretical computer scientists) The logic is pretty simple. The top theoretical computer scientists are more likely to be at the forefront of the P vs NP question. Thus, they will know the relevant background work, understand why previous approaches failed, and may have ideas or new techniques to bring to the field. If you don’t know the name of an author, then a good proxy is the prestige of the university he or she resides. Again, the logic is simple: ranking of universities correlates with quality of professors.
But … an unknown mathematician made the biggest academic result in all of science, technology, engineering, and mathematics in 2013. I’m referring to when Yitang Zhang made groundbreaking progress on the Twin Primes Conjecture. (If I was writing this post last year, I’d be talking about the Higgs Boson result in 2012, which was from several established scientists. But I find Zhang’s case to be more interesting to discuss, as some may consider it as a counterexample to Reason #1.) On the other hand, Zhang’s paper, which is available on the Internet somewhere, satisfies my two other reasons.
Reason #2: The paper isn’t written using LaTeX. Oh yes … LaTeX is awesome. And it’s a shame that some people who write up these proofs do not even take the time to make their papers look nice. All top computer scientists will use LaTeX for their papers, and about 99% of the non-top ones will. Yet some of the papers I saw on that linked page were clearly written using a different (and inferior) word processor.
The use of LaTeX is important to give the impression that one is serious about his or her work by presenting it cleanly to readers. But there’s another easily-identifiable characteristic of bogus papers that can make it clear that the paper isn’t worth reading. That is …
Reason #3: The paper is short. This can vary from discipline to discipline, and whether the paper was written using LaTeX or not (with non-LaTeX papers tending to take up more pages due to poor formatting/spacing). I would suggest that any paper that is shorter than the equivalent of about 40 double–spaced, 12-pt font, normal-margin pages cannot contain enough information to really resolve the P vs NP question. For that kind of paper, the authors would almost certainly need a full section of more than 10 pages just to discuss the background work. Then it’s another 20 or more pages to discuss the new techniques or ideas the paper brings to the field. Then it’s 20 more to set up the lemmas to prove it. Then 20 more to prove the theorem. Then 20 more to show why common counter-arguments are false. And so on.
Only one of the papers I checked on that P vs NP page were long enough to my liking. I didn’t have the time to see all of them, so maybe I am missing a few others. It is worth noting that Zhang’s paper was around 50 pages.
This entry is a few days overdue — my apologies. (Excuse of the day: my thesis is due in a few weeks.) Admitted students had until April 15th to make their graduate school choices, and I made mine on April 13. As you can no doubt tell from the title of this post, I have committed to the University of California, Berkeley. I will start pursuing a Ph.D. in computer science there this fall.
It was not an easy decision, especially because I only applied to top-tier schools. I was debating for a full month between Berkeley and Washington (with each school seemingly have the edge on various days), but in the end, there were several factors that made me end up on Berkeley.
It’s an exciting time for me! I have already received an email about setting up my Berkeley ID, and I can’t wait to get started this fall. I hope that the next six years (five if I’m extraordinarily lucky) will bring fulfilling and rewarding experiences.
Recently, there’s been a flurry of emails in one of my subscribed email lists about some recent regulations that require employers to ask if their employees have a disability. I’ll list the main points of the linked Wall Street Journal article.
- U.S. regulations now require federal contractors to ask their employees if they have a disability. It is up to employees to determine if they wish to disclose any information. (So the title is actually misleading, as the bosses don’t need to know.)
- Those contractors that don’t employ a minimum of 7 percent, or can’t prove they are taking steps to achieve this goal, could face penalties.
- This applies to contractors with at least 50 employees or at least $50,000 in government money.
- The Labor Department issued these regulations to help combat the high jobless rate of the disabled population.
Some people responded to the email list saying that these regulations were long overdue, but a few were not satisfied or had reservations. (To put it briefly, I think these are “overdue” mainly because they help raise awareness of the challenges disabled people face in the workforce, but for now I’ll discuss what others have said.)
A number of people talked about how disclosing information is a difficult and sensitive topic for people with hidden disabilities. Should they tell their bosses or not? Consider the case of someone with a hidden disability applying for a job. Do they feel confident enough to disclose their disability to a hiring committee before any job offer? I suspect that if the job applicant knows that these contractors are trying to recruit people with disabilities, that will raise the probability of disclosure, but I doubt any information will be easily accessible.
Related to the visible/hidden disability discussion, some also worried that employers would give preference to people with visible disabilities to fill in the ranks if not enough employees were willing to disclose hidden disabilities. Because of how the ADA has expanded the definition of a disability, it’s very likely that contractors already have way more than 7% of employees who are disabled.
And of course, there is the possibility that managers and hiring committees of these contractors will protest, arguing that they now have to “lower the bar” to hire a specific group of people.
There didn’t seem to be much discussion about deaf people, which would have been related to the visible/hidden disability discussion because I would argue that deafness can fall in both categories.
I departed from San Francisco and landed in Seattle near midnight. Fortunately, Washington’s visit days didn’t start until the following day (Tuesday). They were the only school I visited that restricted their visit days to be two days, probably because they know they have lots of cross-admits with Berkeley. I rode a shuttle that brought me back to the Silver Cloud Inn near the university’s campus. I did request to live with graduate students, but I was told that Washington had over ninety students show up to visit days, and there were sadly not enough hosts to accommodate all of us.
The following morning, another shuttle took a group of students from the hotel to the Paul G. Allen Center. It was quite nostalgic coming back to the same building that I spent a lot of time in while at the 2011 Summer Academy, and I still remembered my way around it.
Unlike Berkeley, Washington’s schedule started super-early, with breakfast at 8:00 AM. Then the department chair gave us a slide show welcoming us to Washington and gave us reasons why we should attend the Ph.D. program. One of them was that Washington would be substantially expanding the size of their faculty, which increases the pool of potential advisors for incoming students. I remember back in 2012 when Washington shocked the computer science world by hiring seven faculty members. They followed that up with four new hires in 2013. Hiring eleven in the span of two years is is rare enough, but Washington isn’t done yet! The department will be hiring six faculty members in 2014 and then six more in 2015, for a grand total of twenty-three hires in four years! Indeed, the department is going to look very different soon. I also found out from several reliable sources that one of the new 2014 hires is a machine learning expert. The list of new faculty will probably be announced sometime in late May or early June.
After the chair’s presentation, four faculty members gave quick technical talks to highlight the breadth of research in the department. Then there was, as usual, a graduate student panel. According to the current students, the worst thing about being at Washington was (unsurprisingly) the cold weather. I don’t think there was anything super-surprising I learned from the panel.
Later that day, we had faculty and laboratory meetings. I went to the databases and machine learning meetings, which were each an hour long and involved the usual series of faculty talks. Washington’s style is to have grad students groups in the center of the building (usually) with professors on the outer side, but the Paul Allen center has a giant atrium inside of it so it doesn’t feel too isolating. To clarify, each floor has some conference rooms in the center of the building, and the shared offices for graduate students are shown on the right side in the following image:
The machine learning lab was packed with faculty and admits, which forced many people to stand due to the lack of chairs. It was clear to me that machine learning is one of Washington’s strengths. When I had lunch earlier, a current graduate student told me he estimated that more than half of the students were working on either machine learning or HCI.
Quite surprisingly, I was also scheduled to have one-on-one meetings with *graduate students. *These meetings were not a part of UT Austin, Cornell, or Berkeley’s visit days, and I consider that a plus for Washington. (I sent in requests to meet certain faculty members, but not graduate students.) All one-on-one meetings, by the way, lasted 20-30 minutes, which seems to be standard among the schools I visited.
At 5:30 PM, there was a reception in the Allen Atrium with beer, wine, soda, veggies, and cheese/crackers. It was a bit on the noisy side, but I had some worthwhile conversations. (The wine was also good.) Then the admits, current students, and faculty walked as a group to a nearby building where we all had a formal dinner. The food was great.
I had a few more meetings scheduled on Wednesday and also visited the HCI laboratory, but the highlights of the day took place after lunch, where research groups hosted different activities. I went on the AI/ML/Robotics boat tour. Sadly, it was cloudy, so the view wasn’t that great, but it was nice to be outside for a while.
After eating a quick dinner on “The Ave,” I found some time to catch up on email, turn in some late homework, etc., then I took the shuttle back to Seattle. I flew back Wednesday night (in Seattle time) and arrived on the east coast at roughly 9:00 AM. I barely made it back in time for my Distributed Systems lecture on Thursday. It was time for my life to get back to normal, but I admit that traveling was fun.
I also need to decide on which graduate school to attend! That’s what I’ve been doing and still will be doing for another few weeks.
After visiting Cornell, I had a short break at Williams and then traveled to San Francisco to visit the University of California at Berkeley.
I arrived at around 5:00 PM and one of my graduate student hosts drove me over to his house, where I would be staying at for the next two nights. I think Berkeley gave admits the option to reside in a hotel or stay with graduate students. Naturally, I chose to stay with the students, since I think one learns more by living with them. My host’s house was quite nice, and gave me a splendid aerial-like view of their new (and controversial) football field.
There was still plenty of time left in the evening, so my host drove me over to Soda Hall where the admits were getting together. And … wow, there were a lot of admits! One of the visit day coordinators told me that Berkeley had 130 admitted students show up! About half of the students were electrical engineering admits, though, since Berkeley has a joint EECS program.
By the time I had gotten to the lounge, the food was pretty much gone and I was left to stand awkwardly in a noisy, crowded room. Fortunately, I was saved by a group of current Berkeley students who took some of the late arrivals out to dinner. I ended up eating a salad meal from a Vietnamese place near campus, and then went back to my host’s house for the night.
The second day was action-packed, though surprisingly, it started quite late at 9:30 am (good news for those who are night owls!). I went over to Sutardja Dai Hall, which served us breakfast. I had a nice time eating and chatting with other prospective students, and Berkeley even gave free hoodies to us! One interesting thing about the way Berkeley handles the visit days is that they were the only school I visited that gave out name tags to people that also included their undergraduate institution and their research area. That’s very clever, as “Where do you go to school?” and “What are your research interests?” are the first two questions that the admits always ask each other!
At 11:00 AM, we all gathered in an auditorium, where the department chair gave an hour-long presentation introducing us to Berkeley. To the surprise of no one, the chair emphasized how Berkeley had the best placement of faculty members in top computer science departments. He also talked about the department ranking (#1 in computer science, #1 in electrical engineering) and gave us this table based on the 2014 rankings:
- Ranking in AI: (1) Stanford, (2) CMU, (3) MIT, (4) Berkeley
- Ranking in Programming Languages: (1) CMU, (2) Berkeley, (3) Stanford, (4) MIT
- Ranking in Systems: (1) Berkeley, (2) MIT, (3) Stanford, (4) CMU
- Ranking in Theory: (1) Berkeley, (2) MIT, (3) Stanford, (5) CMU (Princeton is #4)
- Average ranking (1) Berkeley = 2.0, (2) Stanford = 2.5, (3) MIT = 2.75, (4) CMU = 3
The chair also listed some of the incredible accomplishments made by Berkeley alums. He later gave us more reasons why they were better than MIT, starting with the warm weather. (At the time I visited there a few weeks ago, the high in Boston was around 26 degrees. Yikes.)
After the chair’s presentation, there were group meetings based on research area. My main interests are in AI, but my name tag actually said my research area was “HCI” and I think this was because the HCI professors were the ones that reached out to me the most. (One professor who talked to me was “cross-listed” in both AI and HCI.) So I went with the HCI group, and got to see some of their latest work while eating lunch. Then they gave me a quick tour across their facilities, followed by another long series of research presentations at the Visual Computing Lab. One graduate student had a really cool project I remember: he used data from images to infer crime rate in cities and applied that to a shortest-safest-path software. Thus, unlike Google maps, which suggests the most direct route when given two points to connect, his software would redirect the person to a longer but safer route. He collected data from San Francisco and inferred it on the Chicago maps, and confirmed its accuracy with real Chicago data. Amazing!
After all that was the campus tour at 4:00 PM. I went inside the Sather Tower and took the following photo while I was at the top. Another one of my photos is the one that starts this blog post.
The day wouldn’t have been complete without the graduate student panel, which took place right after the tour. It was similar to the ones at Cornell and UT Austin. I was genuinely interested to see what the graduate students would answer to the question of “What is the worst thing about Berkeley?” After all, they cannot use the excuses of either the weather being cold (it’s not) or the school not being the highest ranked (it has the best ranking). The grad students said: getting through the bureaucracy.
Right before dinner, there was a light reception at Soda Hall. I got to meet some more amazing admits (some of them are remarkably talented!) and current grad students, and I also had the pleasure of seeing a concert featuring David Culler (guitar), Michael I. Jordan (drums), and a few other graduate students. Then we went to dinner, where one of the professors on the admissions committee recognized me immediately and mentioned that he saw my blog. Say, many of the faculty members knew who I was before I arrived … I guess that’s a good sign.
The third day was when all the one-on-one, 30-minute faculty meetings occurred. I met with five professors and got a good sense of the kind of problems I might work on this fall should I accept their offer of admission. In my free time, I was touring the AI and HCI labs and chatting with some of the current students. My main regret from my visit to Berkeley is that I didn’t get to meet enough machine learning faculty and students, so I’ve been sending a few emails asking questions to the relevant people. At Berkeley, incidentally, the students have shared, open workspaces that are typically surrounded by faculty offices.
After my final faculty meeting of the day, I was able to sneak in another hour or two to do some late Operations Research homework. For dinner, I headed out with the HCI group, and then later took the train to San Francisco airport since, like many of the Berkeley admits, I had to head to Seattle for another visit days starting the next morning….