Hey there! Sign in to join this conversationNew here? Join for free
x Turn on thread page Beta
    • Thread Starter
    Offline

    3
    ReputationRep:
    I have an unconditional offer to study MSc Data Science and Machine Learning at UCL as well as an offer for MSc Artificial Intelligence in Edinburgh.

    Currently I work as a high school math teacher but have previously done a PhD in string theory. I have zero formal training in computer programming and wrote my first Python program only about a year ago. Since then I have been following some online courses to pursue my interest in ML.

    I have broken my question into the following parts:

    (1) As mentioned above, I have a strong Maths background (PhD in subject involving lots of tensors, vector calculus and I have reasonable statistics also) but until recently I had never done any programming. I have since tried to teach myself some basic Python and have now completed two of the Udacity nanodegrees on Deep Learning and Artificial Intelligence. Whilst I'm definitely more comfortable programming than I was, I still feel a long way off being able to do e.g. a Kaggle competition by myself etc. Realistically, will I struggle a lot?

    (2) Is Python knowledge sufficient for this subject or should I look into another language?

    (3) I am not interested in pursuing a PhD. The purpose of me doing this course is to get a job in this area. Which course is better - on the surface Edinburgh seems more theoretical and UCL more practical?

    (4) Edinburgh have asked me to give a yes/no and pay a £1500 deposits by next week whilst UCL have given a much more generous deadline of September. Should I read anything into this?

    (5) I became interested in machine learning (and deep learning in particular) because of medical applications of CNNs, NLP projects and potential applications in finance. I did an interview for a quant job with a bank last week and they quizzed me on ML but it emerged that they only really used regression and classification due to "lack of trust" in advanced ML. Is this true? What are the current employment opportunities like in UK/Europe/US?

    (6) Lastly, if I'm more interested in deep learning, it looks like UCL has more to offer on this. Is that the general consensus in the industry?

    Thanks you very much.
    • TSR Support Team
    • Very Important Poster
    • PS Reviewer
    Offline

    20
    ReputationRep:
    TSR Support Team
    Very Important Poster
    PS Reviewer
    (Original post by needahappyjob)
    ....
    This isn't my area of expertise at all but I can start you off at least.

    The entry requirements will help you to determine how much subject knowledge you need. If you can learn another language before you start that would be great. Remember that people will come with a variety of backgrounds and there will be support available. If you're doing this with the aim of getting a job I would imagine more practical work would be beneficial and if UCL is giving you more or what you are interested in then it looks like that offer is winning. In terms of the deposit and timelines, no there's nothing to read into it. Unis receive loads of applications and want to work out how many people will actually be coming as early as possible so they can plan.
    Offline

    4
    ReputationRep:
    I've sent you a PM with all the information I have. I'm in a very similar situation to you.

    I'll answer (2) here. Yes, Python or R is the way to go. I think the majority of data scientists use one language or the other - probably PhD graduates will tend to favour R. I think Python is probably easier to learn and is more useful in general for all sorts of applications.

    I didn't talk about (5) in my PM. I don't know much about finance and am obviously just on the start of my journey into AI... but from reading books such as "When Genius Failed" and "Risk Savvy", it seems rather dangerous to put too much trust into quantitative models of financial markets. In the Risk Savvy book, the author said that in something even as simple as using the Optimum Portfolio investment method, apparently it would take about 500 years to outperform the 1/N method (1/N means investing equally in each stock), due to the insufficient data to make accurate predictions. Correct me if I'm wrong, but aren't neural networks rather hard to interpret in terms of the relative importance that features play in outputting a certain prediction, so I can imagine why they may favour the relative simplicity of logistic regression or SVMs.
    Personally, I'm much more interested in medical applications of ML.
 
 
 
Poll
Do you like carrot cake?

The Student Room, Get Revising and Marked by Teachers are trading names of The Student Room Group Ltd.

Register Number: 04666380 (England and Wales), VAT No. 806 8067 22 Registered Office: International House, Queens Road, Brighton, BN1 3XE

Write a reply...
Reply
Hide
Reputation gems: You get these gems as you gain rep from other members for making good contributions and giving helpful advice.