Which is the most difficult CS subject/theory that you studied but important to the field? And the reason please?
3I think this largely depends on the person learning the subject, and the person teaching it– Mahmoud HossamFeb 8, 2011 at 8:19
“There are 2 hard problems in computer science: caching, naming, and off-by-1 errors”
4I'd add threading...– CarraFeb 3, 2011 at 13:44
Honestly, compiler construction!
13+1 Compilers was the most difficult and the most rewarding. Jan 31, 2011 at 5:40
3It was up there with the most over all work, and good prep for grunt coding, but I don't think it was all that difficult. Maybe harder without the using YACC or whatever tools we used, I dunno. Jan 31, 2011 at 15:09
4Compilers are really only difficult because most of the theory dates to times of insanely severe hardware constraints and a lot of the formal instruction hasn't advanced too far beyond that. Take a look at Let's Build A Compiler to see how easy compiler-writing can be if you approach it from a different angle. Feb 1, 2011 at 17:04
1@Martin York, as a compiler writer I think that complexity of implementing compilers is overestimated severely. In general, a compiler is much simpler than an interpreter. I suspect it's a Dragon Book and its kind are to blame, they're proposing the most complicated ways of doing simple things and paying too much attention to the least important step, namely parsing.– SK-logicFeb 1, 2011 at 20:13
1@Martin York, there are techniques of keeping an AST as simple and maintainable as possible, no matter how complicated the source language is. Also there is a number of very simple but powerful techniques for keeping each stage of compilation trivial and isolated.– SK-logicFeb 1, 2011 at 21:23
Design & Analysis of Algorithms
I think that question depends on the teacher you had, and how that subject was organized in your career.
Analyzing algorithms can be as hard as someone wants. Take in count that there are unsolved problems, and not only that: problems that can't be solved.
The thing is that you can have a problem, and if you know it can't be solved, that's perfect. But what if you don't? You can spend a lot of time trying to demonstrate it's NP-Complete, or trying to find a polynomial time solution to solve it.
Demonstrating NP-Completness is not easy. Yes, lots of problems are known, but the thing is to find the reductions to demonstrate that it's NP-Complete. And what if you spend lots of hours/days/months trying to demonstrate it, and it can be solved in polynomial time? :)
There are also other subjects, like Compilers, Group theory and Primitive Recursive Functions that can be as hard as the subject plan or the teacher wants ;)
1s/Analisis/Analysis ... otherwise exactly what I think ... primitive recursive functions, uargh!! Jan 31, 2011 at 5:52
Agreed, I muddled through my bachelors degree never being confident that I ever successfully 'proved' anything (although my Algorithms class was way too simple, due to the professor) Jan 31, 2011 at 14:26
I will show you in these days, how hard algorithms can be :) Feb 1, 2011 at 6:07
Pattern Recognition i.e. Artificial Intelligence. This refers to smart computing along with other pattern recognition tools like, Optical Character Recognition, Voice to text, facial identification, etc.
Many of the "cool" things you can do or wish you could do with computers rely on these algorithms, and we have been attempting to perfect them for decades without a whole lot of success.
It's hard because it's not something that's deterministic. Developing a good AI pattern recognition requires experimentation for every application you want to use it for, to ensure you pick the right algorithm, the right features, etc... Feb 1, 2011 at 3:06
1I am just beginning to climb this particular mountain (pattern recognition). It's hard. LOTS of math. Great, huge, intimidating piles of math, staring back at me, daring me to enter. Feb 1, 2011 at 23:31
well ... pattern recog can also be seen as applied statistics, it's not just a problem within the range of CS Mar 25, 2011 at 15:18
My pick is computability theory
(Hmm... maybe it's not that important, but it sure was difficult)
2I agree, and I would personally generalize it as en.wikipedia.org/wiki/Theory_of_computation.– Matt HJan 31, 2011 at 17:16
I'll agree that the Theory of computation was hard, but it was also one of my favorite subjects. Granted, I was double-majoring in Mathematics... Jan 31, 2011 at 17:41
+1 I double-majored too. I could handle an intro to this stuff, but the graduate version ... glad that I dropped it!– JobFeb 1, 2011 at 3:01
it was hard, not we know so much about it that it doesn't matter much. Feb 6, 2011 at 9:42
There are only two hard problems in Computer Science: cache invalidation and naming things. - Phil Karlton
category theory (discrete mathematics), but worth it
What specific benefits did you get from learning category theory?– zvrbaFeb 1, 2011 at 10:39
@zvrba: a deeper understanding of abstraction techniques and problem mapping Feb 1, 2011 at 16:03
Where did you learn it from?– zvrbaFeb 1, 2011 at 16:15
@zvrba: I don't see the book on my bookshelf, it's probably still in storage (remodeling) but i think it was this book amazon.com/Category-Computer-Scientists-Foundations-Computing/… Feb 1, 2011 at 18:12
If you do it just slightly wrong, it could cost a company millions.
Although increasingly popular, Crypto isn't unique to software. Mar 25, 2011 at 18:01
Crypto isn't that hard. The problem is that security can't be tested easily, so you only notice your mistakes when somebody hacks you. But lack of testability applies to most forms of IT security, not just crypto. Feb 24, 2013 at 9:07
Operating Systems, especially the part that has anything to do with threading.
And the reason isn't because it was that hard to make 5 philosophers eat pizza with a fork. The reason is because writing multithreaded code is in and of itself difficult and not necessarily easy for the human (at least male - according to my wife) mind to compute.
9Let your wife write the multithreaded code then :)– user1249Jan 31, 2011 at 23:05
3Remember, when it comes to shared-memory multithreading, the computer is a sneaky swine that is out to get you. Doubly so when dealing with a multicore processor; one core can be distracting you in front of your eyes where you're watching, and the other can then go behind you and stab you in the back. Feb 8, 2011 at 9:38
I too vote for Compiler Design. Especially where the DFA and NFA part comes in. I am also not so clear about NP problems and stuff.
Yeah, I'd have had a harder time with Compilers if I hadn't taken Theory of Computation first. Jan 31, 2011 at 15:07
DFAs and NFAs are chicken feed. Wait until you have to do LALR(1) parsing. Jan 31, 2011 at 18:08
Well technically this is a branch of mathematics, but is highly relevant in CS.
Nearly everything in CS is based on queues (visible (obvious) and invisible (not so obvious or implied)).
In the early days of CS the queues were obvious.
A queue of programs (each program a deck of cards).
Nowadays the queues are not so obvious. The internet for example: a packet switched network, but the packets form queues and routing the packets is a form of queue minimization.
Hey! (? Are you a (Lisp programmer)– Mark CFeb 1, 2011 at 17:01
Not (As much as (One could (see), But (It has (been known)). To Happen)). Feb 1, 2011 at 18:40
It's not too hard on the toy problems you're given in the course, but once you start considering real problems it turns into serious drudgery.
Interpreting client requirements when the client doesn't really know what they want. This is not taught in college, and is one of the the most essential skills to have.
1I'm not sure I agree with this one as being a Computer Science concept. I also don't see how it can be solved using the scientific method.– jmort253Feb 1, 2011 at 6:09
@jmort253 - This is true, but computer science tries (unsuccessfully in my opinion) to investigate this field with formal methods of design and validation. Feb 2, 2011 at 8:46
I agree is not a "computer science" concept - but when I started my career I was unaware/oblivious to the fact that clients don't know what they want. I thought ALL software projects came with some kind of a formal requirements doc. Maybe a lecture topic for a software engineering course (maybe my college didn't cover it)? Feb 3, 2011 at 13:52
Personally, mine was Formal Logic. It was tough to start with, but once you get the rules down and manage to play with it enough, your brain goes
Logic++;, which in development is a very good thing.
As a side-note, I am answering the question directly - this was definitely not the hardest subject when I did my degree, but it was probably the hardest "real-life applicable" subject.
Formal Logic is something that I had a love/hate relationship with. I liked thinking through the concepts, but I could never understand how it was helping me until later when I encountered real-world problems that required logical thinking.– jmort253Feb 1, 2011 at 6:11
@jmort253 - It was the same for me really. I even struggled to the point of thinking I'd fail it, studied so long and hard until it finally clicked in my head. After that, the benefits have been amazing. Feb 3, 2011 at 12:10
Compiler Constructions. Hard but must to understand the concepts behind
5You should give an up-vote for the same answer which was provided before you, rather than giving the same answer again. Feb 1, 2011 at 16:39
Kernel Design anyone ? Well I don't really know how it's done and what is the targeted features for an OS, but for me thinking about designing a kernel must be a daunting task.
I also think about computer security; I don't really know what makes a system unsafe except of course, obvious buffer overflows, XSS and SQL injections.
I'm not sure, but there seems that some algorithms are also unsafe; look at the MetaSploit project, it lists all type and kinds of security breaches: you can see there are a lot of ways a program can be flawed.
There are many awkward topics in the field, but my picks for sheer persistent difficulty are those involving Global System Properties. Examples of this general topic include:
- Safe and deadlock-free multi-threading
These are hard because you're after something that only exists when everything is correct; you need a global system property and yet virtually all the tools available (and all the ones that scale to real problems in my experience) only really do local reasoning. It's the process of going from reasoning about the pieces of the program to the whole shebang that's hard, particularly because it's entirely possible to have pieces that are all correct in themselves but where there still are subtle bugs because the components are incorrectly arranged; the bugs can be undesirable emergent characteristics…
Management Information Services
During my college period i used to have one management subject each semester which totally made me mad.
Tough! well subjects like Compiler Design, OS Design etc are tough but they are really Interesting and challenging. I really messed in subjects like Management Information System / Services etc as they are full of boredom and you have to go through lots of theory.
2Full of boredom because they're talking about the conceptual intricacies of each system, whilst half the people never wrote any system themselves (but they surely did use a variety of). Also, the seminals use so many loaded words yet fail to provide a real life example in plain English. Like decision support systems... couldn't you just drop a few screenshots of Google Analytics reports, FML, just to get the students on the same page before you run off having an intellectual orgasm in front of the audience. Jan 31, 2011 at 14:14
If you are working in C/C++ pointers are the most important concept to know. But somehow I never understood it fully in college.
12really? I mean, each person is different, but I think there are lots (I mean, lots) of topics harder than just pointers. For example, Computer's Architecture, Assambler that in some way are related to pointers ;) Jan 31, 2011 at 6:03
True, but you'll find understanding memory referencing through assemblers way easier, because you actually work with raw pointers, whilst in C/C++ your working with references to pointers, which just confuses the hell out of people because the abstraction is never outspokenly talked about. Jan 31, 2011 at 14:17
2Ah assambler, the best programmer's tea Feb 3, 2011 at 12:52
The guy asked the topics which are difficult but important, hence pointers.– Manoj RFeb 3, 2011 at 14:18
@Matt: You just made my day :D @Manoj R: Pointers are trivial if you just think of them as array access. Or is array access difficult?– back2dosFeb 8, 2011 at 10:06
Design and Analysis of Algorithms. It isn't so much that it's hard to understand and analyze known algorithms, it's that designing and analyzing new algorithms for hard problems is difficult, and requires a broad understanding of many areas and practice in applying many different techniques.
Constraint Programming. which deals with combinatorial problems, NP-complete problems.
Optimization of Algorithm is challenging Topic.
Which is the most difficult CS subject/theory that you studied but important to the field?
It was difficult because the theories are very loosely related to each other but they're used in CS. Too much memorization I guess...
Proof by Induction, Big O, recursion, divide and conqure, Graph Theory, blah blah.. argh!
Compiler for me was easy, because we had to take Theory of Automata. ^^
Z notation/formal methods used to hurt my brain at college. Mainly because I hated it. Hard is a lot easier when you enjoy what you're doing and much harder when you don't.
I like your answers (and I didn't forget upvoting them), like compiler, kernel, etc., but most of programmers never met these problems. There is a bit easier, but more common issue: concurrency - threads, locking. It's very easy to write a program which produces magical errors, if we make even a small bug in the concurrency architecture.
So, I say, it's not the hardest issue in computing, but because it's commonly used, it is a dangerous one.
Object Oriented Programming
It's probably because I cut my teeth on FORTRAN and APL, but the shift from strictly procedural languages to objects has been something I've struggled with for years. It doesn't help that so-called 'experts' write conflicting articles and tutorials on what it means to be object oriented and the best/proper ways of constructing object oriented programs.