In this course we will have group homework and peer review. See the syllabus for more information.

I want to discuss my reasoning for these two decisions. Warning: much preachiness lies ahead (but I preach it 'cause I believe it).

Group homework goal:

Encouraging discussion: Many people find it easier to solve a problem by talking through it with another person. This is how most scientific work happens. I want to encourage discussion without causing worry about whether the discussion is too extensive and veers into cheating/plagiarism.

Within a group, any amount of discussion is permitted, and you will only submit a single homework submission from the whole group. Between different groups, the normal discussion rules will apply: high-level discussion is allowed, but not enough to enable one group simply to copy from another. See the Academic Dishonesty portion of the syllabus for more details.

Peer review goals:

Goal 1: Learn to evaluate others' technical work: This is itself an important technical skill to learn.

Goal 2: Learn to write for target audiences who don't know the solution already: When writing up your homework solution, it is best not to think of the target audience as the instructor or TAs, but rather other students in the class. That is to say, your goal is not to convince another person, who knows the solution, that you also know it. Instead, your goal is to convince another person, who does not already know the solution and may not be an expert in the field, that your solution is correct. This helps to clarify an important lesson: it is useless to solve a problem if you cannot teach the solution to others.

Most of your school experience has involved writing for teachers who know more about a subject than you. However, for the remainder of your career, your technical writing will be primarily targeted at people who know less about the topic than you. (This is why you are doing the writing and they are doing the reading.)

Unfortunately, most class-based homework and exam work is less an exercise in problem-solving and solution-communicating than a choreographed performance similar to a figure-skating routine or a piano recital. Each performer may mix and match techniques in different orders, but they are otherwise not doing anything fundamentally new: they are merely reproducing techniques well-known already to the judges, who score them by mentally comparing to a hypothetical idealized performance. However, this is misleading as a model of how computer scientists actually solve real problems and then communicate their solutions.

A major goal of the homework is to practice effective communication. The following phenomenon is often a sign that this goal has been forgotten: questions about the precise rules and boundaries for what constitutes a full-credit solution:

"Do we have to justify the reduction, or can we just give it?"
"Does it count if we don't write down the transitions?"
"If we give an example is that enough, or do we have to describe the whole construction?"

My hope is that peer review will illustrate that there is a single answer to all of these questions, and countless others like them: if it makes your solution easier to understand, then it is worth doing. Homework in college/graduate school will be your last experience with precise, unambiguous rules clearly delineating what constitutes "success". (If a job's duties can be codified into an unambiguous objective function, then this job will likely soon be done by an artificial neural net.) In the long run, it is beneficial to wean yourself from the comfort of certainty and treat the homework like every other problem you will ever encounter: where success is measured by how convincing you are to other people. So practice asking yourself the following kinds of questions:

"Is the correctness of the reduction obvious from its definition, or can we imagine ways it might appear to fail, so that we need to argue that those cannot happen?"
"Is it clear what the transitions are, or should we write them down explicitly to be sure they are unambiguous?"
"Is it obvious how to generalize this example to the whole construction, or would it help to describe it in more general terms?"

All of this has the potential to frustrate students who mistakenly believe that the purpose of grades and coursework, as though they were some intellectual version of a purebred dog show, is to evaluate the students: what they know and how smart they are. It isn't. The goal is to evaluate the work itself, and how well it communicates ideas.

Goal 3: Learn to write for a target audience with varied levels of ability/time to read your work: In reviewing the work of others, you may find yourself frustrated in trying to read their solution (and your review comments should indeed mention, politely, what frustrates you). However, a moment's self-reflection will often reveal that you have maybe done the same thing yourself, but you didn't think about it when your only job was to write down a solution, but not to read one. I have significantly improved my own ability to write papers because I have had to review so many other papers. Knowing what it feels like to be a reader of content is incredibly helpful in teaching you how better to craft your own content.

These lessons are particularly relevant to those of you intending to do research. I hope to dispel the notion that the only people who will ever evaluate your work are experts, with hours of time available to read and re-read and ponder and appreciate every nuance of your writing.

If you have not yet experienced the act of reading others' reviews of your research papers, a discomforting shock awaits you. You will soon learn that real peer review (of the kind used to judge research papers) necessitates an approach to writing that is "robust to the worst-case reviewer". You need to write in a way that impresses even lazy reviewers, who often don't have the requisite background knowledge to judge your work, who don't read the whole paper, or even the whole abstract, who get confused on page 4 and ask a question that you already answered on page 2 (but the reader forgot because that was two pages ago, or they skipped straight over pages 1-3 to page 4 because that's where the main results start).

Think of this class as a very gentle introduction to the social game called peer review that, for better or worse, all researchers must eventually learn to play.