Page 1 of 3 123 LastLast
Results 1 to 10 of 30

Thread: suggest a masters topic for me

  1. #1
    Member
    Type
    intp
    Join Date
    Dec 2013
    Posts
    842

    suggest a masters topic for me

    I think I want to start a masters degree in January/February and I'm fairly confused about how the topic thing works.

    I've been procrastinating for a while on this, so hence I'm making this thread in case anyone has some nice pointers or suggestions to get me moving. It would be very well appreciated.

    The masters degree would be a research based one. It would be directed towards data analysis/data science/that whole thing. That to me sounds like fun. I'm slightly wary that the area might be overhyped, but I'd also ultimately be happy with other options in that general direction.

    I'm basically looking to get into that kind of area, by getting a masters degree, because I never did any internships.

    I have a degree in "maths and computer science", so I have a solid undergraduate level of maths and also solid programming ability without much practice working on real things. I think it would be best to try to capitalise a little bit on maths ability, but also combining with programming stuff. That mix of the two really interests me as well (the variety, application, etc).

    But I'm very confused about what to do from this point. I think I'm supposed to email some professors, or the school in general and make some application and pretty much tell them what I'm interested in. But I don't really know. I don't know what exists and whether something is viable or whatever.

    So, what are some interesting masters projects I could consider?

  2. #2
    Utisz's Avatar
    Type
    INxP
    Join Date
    Dec 2013
    Location
    Ayer
    Posts
    2,769
    It's a hard question to try to answer to a stranger over the internet. There's a lot of different topics in the area of data analysis/data science. I think you should first think about if you're interested in something like machine learning/data mining, or something related to data management (databases/NoSQL/distributed processing frameworks like MapReduce, Spark, Giraph), or perhaps something related to data integration or data exchange (knowledge representation, query languages, schema expression).

    Machine learning relates to the theory of learning patterns from lots of data (think regression methods, neural networks, all that).
    Data mining is essentially applied machine learning: lots of experimental stuff (think spam filters, recommender systems, event detection, business intelligence, etc.).

    Data management refers to scalability in data-intensive scenarios: propose algorithms, implement them, optimise them, test them for interesting use-cases in terms of performance and scalability (think communication protocols, indexing techniques, join algorithms, query optimisation, etc.).

    Data integration/data exchange relates more to how to represent data, perhaps logically, perhaps syntactically, designing formal languages for data and queries, looking at computational complexity of problems associated with the languages (think logical languages, formal problems like satisfiability, entailment, formal semantics, model theory, NP-completeness, undecidability, etc.).
    Applied data integration relates to multi-model scenarios (how to answer queries over multiple relational databases and, say, a Web API for example) or to mapping languages (how to express/compute relationships across databases or datasets).


    Maybe nothing jumps out but these are some high-level options to start thinking about in terms of what you're interested in (in other words, I would recommend to forget "data science" and start thinking more low-level technical stuff).

    Also for a masters, it's important to have a concrete problem to tackle. Probably the topic might (should?) end up touching on multiple areas rather than going deep into one. I think masters topics are better when they're broad rather than deep: broad but focused on a concrete problem, be it theoretical or applied.

  3. #3
    Utisz's Avatar
    Type
    INxP
    Join Date
    Dec 2013
    Location
    Ayer
    Posts
    2,769
    (Should add it's quite rare for a masters student to have their own topic decided. Normally most masters students are offered fixed topics by profs and they choose one, or they apply to work on a larger research project where the topic is quite fixed. I dunno how things work where you are though.)

  4. #4
    Utisz's Avatar
    Type
    INxP
    Join Date
    Dec 2013
    Location
    Ayer
    Posts
    2,769
    Also if you're open to travelling, you should look through the DBWorld and AISWorld mailing lists for recently adverstised masters topics. Here's one advertised in Ottowa about data mining in MOOCs, for example.

  5. #5
    凸(ಠ_ರೃ )凸 stuck's Avatar
    Type
    xNxx
    Join Date
    Dec 2013
    Posts
    3,642
    "A Big Data Approach to Making Tough Life Decisions"

  6. #6
    igKnight Hephaestus's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Location
    Ceti Alpha V
    Posts
    9,142
    I've got one for you. It's a big topic, but might be better explored then I've bothered to look.

    Algorithms for Match-2/3 board generation and cheat detection in boardsets.

    There are two main types. One where you swap tiles/blocks to trigger sets, and one where you select sets to remove. I the select to remove sort, tiles/blocks usually do not repopulate. When you swap, they usually repopulate. In either case, the norm is for tiles/blocks to shift down if the space beneath them is empty.

    For the rest of this introduction I may use the terms tile(s) and block(s) interchangeably, though I will endeavor to restrict myself to blocks. When I refer to a board, I mean the game board which holds and displays the games state.

    The common case I've seen is to have a board separated into an 8x8 grid. At the beginning of play, the board is populated with some stripe of random distribution from the possible blockset, usually there are at least four types of block, often more. The problem I'm currently intrigued by, and am thus handing off to you because it would be fucking hard and I don't have any ideas for traction yet, is how do you determine if the game is cheating on placement of blocks.

    Here's the neat thing: for these games to work well, you can't be truly random. The problem with randomness is that randomness guarantees dead boards-four colour mapping and all that. So you have to do something to keep dead boards from appearing, but what is a fair way to do it?

    Given:

    1. Your board must have representation of each type of block in use.

    2. Your board must have matches.

    3. Players are going to suspect you are cheating if the boards are consistently unfavorable.

    I know how to determine if the distribution is off. That's trivial if tedious: count blocks over a set of boards, and check the means. But how could you show bias away or toward a particular type of block in placement. For example, how would you detect a Markov chain that selects away from allowing more than two tiles of a particular colour to be adjacent at generation?

    Just some stuff rolling in my noggin in arguing with someone below my mathematical league about just such a game. I think the boards are actually random an their whinging is about not getting favorable boards, which they notice because they're are specific things that would be favorable that they are consistently not seeing--but I intuitively know there's a fuckton more unfavorable but random boards then there are favorable random boards (and intuitively knowing such things is not at all satisfying). I can show that the distribution of colours favours random, and I've yet to see a dead board so I'm certain their fudging at least that much. But I've no clue how I would be able to detect placement bias with any certainty.

    It seems interesting to me, so it might be interesting to others.
    --Mention of these things is so taboo, they aren't even allowed a name for the prohibition. It is just not done.

  7. #7
    Member
    Type
    intp
    Join Date
    Dec 2013
    Posts
    842
    @Utisz Thankyou very much for that. Quite helpful. I'm still sort of considering what you have said, so it's a bit difficult to respond, but this is where I'm currently at, pretty much:

    I'm quite naive about many aspects of it all. So I'm conscious that I'm reevaluating my perceptions regularly. So I may say some things which are not quite accurate. Sorry about that.

    A main goal here is to choose something that both maximises employment options at the end, and is sufficiently enjoyable.

    Data integration seems to me as though it is/will be quite a big thing. And the problem itself is something I think is really valuable and fascinating. I know almost nothing about this, so it would be good to make it a feature.

    Machine Learning has a kind of "coolness" factor, I think. I would like to understand it better. It seems like something I would read into in my own time anyway. It also seems that in most employment situations, it would be sufficient to understand the variety of machine learning approaches, to be able to select the right one for the specific situation, understand the technical issues/drawbacks, and implement it. So... possibly... I would include something like that, but just as an essential, but token, feature of the whole thing.

    Data management: I hadn't really thought about that. I hadn't really considered it as part of the area. I'll look into this.

    ------

    I hadn't really considered that the university would choose for me, or direct me towards the masters project focus. But I think that's how it would work. I think it's sort of dictated by the current research interests of the school and the specific professors/supervisors at the university. Actually I think I also have *slightly* more time as well. Applications should be done by the end of October.

    At this stage, now, I'm trying to get an idea for what an appropriate scenario could be. I may have to do some googling to get some ideas. If it is something vaguely towards something I'm interested in, then that would be a motivator. Lately I've been thinking there must be a lot of environmental data that needs to undergo some kind of processing/etc. I'm reasonably convinced that we're heading towards an environmental apocalypse, so something focusing on might be good. Just one possibility.

    @Hephaestus I think that seems like an interesting problem, but for my tastes it seems a little bit too theoretical/abstract. I think I'm really looking for something a bit more tangible, outside the world of abstract maths, but still involving maths, where I can pull data from messy real world things and get messy within that. Thanks for the suggestion though.

  8. #8
    Married Mouth-breather JohnClay's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Posts
    2,002
    INTPx Award Winner
    It could be related to general purpose AI in a specific domain (anything really, even the stock market)


    This is based on the ending of the previous video:

    This is even cooler than a fractal like the Mandelbrot Set... especially after a minute or two.... well the Mandelbrot Set is impressive that the equation is so simple....
    How it works:
    https://www.youtube.com/watch?v=BsSmBPmPeYQ


    The video shows an agent collecting rewards in previously unseen mazes using only raw pixels as input. The agent was trained using the Asynchronous Advantage Actor-Critic (A3C) algorithm and was only rewarded for picking up apples and orange portals during training.
    Paper link - http://arxiv.org/pdf/1602.01783.pdf
    Using that training it could also maybe navigate real world mazes...
    Last edited by JohnClay; 08-31-2016 at 10:46 PM.

  9. #9
    Senior Member Starjots's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Posts
    1,783
    Quote Originally Posted by stuck View Post
    "A Big Data Approach to Making Tough Life Decisions"
    An Evolutionary Search Algorithm for Weaponizable Jokes in Large Data Sets



    sadly, this is where the big bucks are...

  10. #10
    Member
    Type
    intp
    Join Date
    Dec 2013
    Posts
    842
    If anyone wants to jump in and suggest a way for me to say "I generally want to do something about data mining, but I don't care what. It's all going to be interesting, right?" without sounding like an idiot, that would be appreciated. I'm just shooting these people an email to ask if we can meet up briefly so can elaborate a bit more. So I'm just going for a couple of sentences on this.

    Some specific stated research interests from three potential professors:

    "privacy preserving data mining"
    "data mining / statistical computing / bayesian analysis / analysis of categorical data / statistical data mining"
    "modelling stochastic dynamics" - seems a bit outside my intentions, but worth a try...

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •