Jump to content

compscian

Members
  • Posts

    78
  • Joined

  • Last visited

Everything posted by compscian

  1. @Rigenate Just curious, why not just change research groups? Are there really no one in your current school who does the type of work you like, or is there some other program in your current school which fits well with your interests? I know of many students who internally changed programs e.g. from Physics to EE, Math to CS, Math to Economics etc. In most of these cases, the research interests of student changed, they talked to an adviser who matches their interest (and about half the times, they will be outside their current department), and started working on a fresh project.
  2. I am OK with you PMing me. For the more general questions, I can take them here so that others benefit too. For more specific or personal questions, we can try PM on gradcafe. Some statistics departments are joint with IEOR ones -- they are particularly good choices (e.g. Georgia Tech, Princeton etc). However, note that OR or data science programs will not help you towards the ultimate goal of PhD in NLP, for which you have to be in CS, EE (esp. for speech, sequence prediction, or dialogue systems), or Linguistics (semantics, language understanding). You'll run into the same problem of department mismatch after your MS, and your chances might be even lower than now (bar for PhD >> bar for MS). If you are particularly interested in NLP, and also want to develop a parallel skill set in NLP, I'd actually suggest a computational linguistics MS program. For example UWashington has an excellent program. Otherwise, try and apply for "data-science" programs. Operations Research won't look good on your resume since it's very far from NLP.
  3. Hi @Tanul I'm an ME who transitioned into ML, and can probably share some insights. When you say ML program, which department are you talking about? If you want to join a CS/EE or Stats department, you will get degrees in that discipline. Hence they will naturally expect familiarity and pre-requisities (eg. if you join CS/EE/Stats you might have to take operating systems / communications / measure theory). Do you have a background in these areas; and are you interested in doing these courses for sake of degree? In absence of above, your best bets are "data-science" type programs at places like NYU. Alternatively, you can look into operations research or information management programs and take a good number of electives in Stats and CS. Online courses are not useful by themselves for a number of reasons -- see this post by a professor http://qr.ae/1JprKQ Kaggle competitions are useful on your resume if you actually accomplish something impressive. Participation means nothing since anyone can do that. To get good results in Kaggle, you need a good understanding of ML, and it's not easy. On the other hand, if you get impressive results, it will certainly add a lot of value, but it will be hard. My suggestions: Retake GRE, maintain a good online presence with your codes on github, apply to data science, business analytics, or info management type programs. Honestly, there isn't much you can do to improve your profile at the stage. One long shot is to join some professor at IIT or IISc as a research assistant and work for a year. However, convincing them to take you is not easy -- but strong interest in the form of MOOCs, Kaggle, and github repos might help. You can then use the professor's recommendation letter and apply for Fall 2017. It's a circuitous path, but is likely your best option if you want to get into CS or Statistics departments.
  4. @marmle I don't know about all the univs, but I can comment about the CS/EE and Statistics departments of UW, Penn, Columbia, and Rice. At UW, nearly all the admitted students with the ML emphasis in CSE, EE, and Stats departments had at least one published paper or in the pipeline. Similar was the case with Columbia. Penn and Rice are also great places -- I didn't visit these two, but was admitted. I was told that admission was competitive here as well, and professors looked for one really distinguishing factor. My understanding is that grades don't really matter much (my undergrad was in MechE with Applied Math minor) -- there would be 100s of applicants with really high grades anyway. Publications and research experience count a lot more. ML and related areas are insanely competitive at the moment, so I would recommend you apply to a few extra backup places. PS: Operations Research in places like Princeton and GATech are also possibilities for ML/Stats related work, and are a little less competitive. Best wishes!
  5. I'm more familiar with the EE (or EECS if I can say) side of things as opposed to pure CS areas. My reading is pretty similar to yours. In decreasing order of competition. Machine learning and related areas like computer vision, NLP, speech, and general data science are the most competitive. From my experience, these also provide the most lucrative post PhD outcomes -- especially in the industry. The signals & systems side of EE -- optimization, info theory, comms, control, DSP, compressed sensing. This attracts the mathematically minded folks who want to work on theoretical questions and stay in academia. Some of the more practical ones might jump ship to (1) above; and separation within these areas in (2) is wafer thin. Industry lags behind theory by at least a decade or two, and hence jobs meaningful of your research is hard. As you said, it's make it or break it -- professor or bust (or jump ship to data science band wagon). IoT, embedded systems -- possibly the next big thing. Computer Arch, High performance computing, HCI After this, I think (5), (6), (7), (8) in your list is quite accurate.
  6. @TakeruK and @rising_star Thought I should update you. I talked to the professor, and he is supportive of me moving. However, the department would be unable to process my offer for this Fall due to administrative issues. After considering all possibilities, I am sticking with my original plan. I will be working with this professor remotely -- he has made the same arrangement for a few of his current students. I'll have a co-adviser in this university (there are 2 others who do related work). He has also agreed to the possibility that if I really want to work with him, and if the project in the first year works out well, I can transfer to the other program.
  7. @TakeruK and @rising_star Thanks for your comments. I am planning to wait till we talk (Saturday) before contacting the university (on Monday). I hope it won't be too late. If this conference were not around, I am sure we would have talked sooner. What really bugs me is that this professor was very interested in recruiting me, and I'd attribute his presence as the factor that swayed my decision towards UW. I'm just praying he doesn't leave me stranded. We had a good rapport, and in fact even started laying down the foundations for my project through reading assignments and implementations of published results. But I can sympathize with his decision - he got his PhD from this univ, went out and made a name for himself, and is returning back to his alma mater. At times like this, you can't help but feel that world works in mysterious ways to screw you over backwards
  8. @fuzzylogician @TakeruK Thank you so much for the response. The professor is currently attending a premiere conference and wants to talk this Saturday. I'll bring up these issues when we talk. There are a few issues I am very anxious about. Firstly, if I contact the new school, and even if the current one doesn't know about it, I will personally know that the new place is a better fit for me. This will definitely affect my performance if I am not allowed to accept again and have to go with the old school. The new school is actually more reputed overall (but same in my field). I went with the lower ranked school specifically for a better research fit. Now, this school isn't the best fit for me, and I have also let go of a stronger department (esp with the move of this prof). Also, I'm an international student, and have completed a good chunk of the visa process. If I switch places now, it might lead to a bureaucratic mess or very long processing time. The stress is going to kill me.. But this is less of an issue compared to the above. I feel devastated and hesitant to discuss this with anyone. My parents have a number of health issues already, and I'm really scared of what might happen to them if I break this suddenly. This just puts a lot more pressure on me to somehow make it work I understand that you can't make accurate predictions. But, have you seen situations like these before? Will universities make offers open again under circumstances like these? I'm really anxious and have already asked for a week of leave. This might just be the worst week of my life yet!
  9. Guys, I am freaking out. I just heard from my potential adviser that he is moving to another university. The irony is that I was actually admitted to the university he is moving to, but I decided to accept UW's offer because I wanted to work with him. My understanding is that he got the offer and accepted it after April 15th. Can I write to the new university asking them if they can give me the offer again, and explain that I will accept and come because of my adviser moving there. Would they hold it against me for rejecting their offer earlier? Also, if I do this, and I don't get a positive response, I'll feel miserable going to the other university. They will obviously know that I tried to move, and am not very happy here. I am very interested in working with this professor - he was very enthusiastic while recruiting me. I have also been doing some work with him (reading assignments and some exercises) from May, and he has been giving me constant feedback every two weeks or so. I'm really torn as to how to proceed. As a city and department in general, I like both the places. The only differentiating factor was my adviser. I'll feel gutted if I don't get back the offer
  10. I'll add a different perspective: it doesn't really matter what your MS is in. Any good PhD program worth their salt will value relevant research experience more than coursework. If your ultimate goal is a PhD, I'd suggest you pick a program that is very light on requirements and will give credits for research work, even if thesis option is unavailable. Personally, I did an undergrad in MechE/AMath and wanted to do a PhD in ML/data science. I had little coursework apart from the standard math sequence: CS (zero outside ML/AI); EE (only optimization and signal processing; zero in circuits); Statistics (no dept in my univ). I spent a year with an EECS professor doing research and published a paper, and didn't have trouble getting admitted to all three departments above. Coursework doesn't matter.
  11. In that case, I think you have covered nearly all the departments/universities Also, MIT and Princeton don't have a "statistics" department yet, but they have an Operations Research program which is basically statistics with an increased emphasis on optimization. They might interest you too. Best wishes!
  12. If you are interested in "machine learning", then you don't need to limit yourself to statistics or biostatistics departments. You can apply to CS or EE programs too. For example, UT-Austin (not on your list) has a strong machine learning program in both CS (applied flavor) and EE (theory flavor). Also, I would actually apply to TTIC instead of UChicago if you are more into ML as opposed to traditional statistical estimation. UMass also has a very strong ML program (stats and math in same dept + CS is strong).
  13. From your profile, you will have better chances at statistics or computational linguistics programs as opposed to CS, which requires competent knowledge across theory, systems, and ML; and not just ML. You can of course sit in a stats program and have an adviser in CS (your best option IMO). For a good list of schools, I'll refer you to Jordan Boyd-Graber's pretty comprehensive list: http://qr.ae/8YoNwX My personal picks: UW, WUSTL, Stanford, Columbia, CMU are the strongest for ML/NLP overall. Regina Barzilay at MIT does some of the most interesting work. JHU, TTIC and UMass (Andrew McCallum) are easier to get into, but also have great ML/NLP research groups.
  14. This may no longer be relevant for this year, but I'm going to post it anyway for future students. There is a fundamental difference between factor and criteria: Factor: a circumstance, fact, or influence that contributes to a result. (read: attribute) Criteria: a principle or standard by which something may be judged or decided. In other words, factors help you decide between options: to say one option is better than the other. Criteria is the evaluation metric itself. For example, if you want to drive from point A to point B, there are a number of factors that will influence your ride. They can include weather, traffic, condition of roads etc. Criteria is what you want to optimize: for some people it will be travel time, for others it can be smoothness of ride, fuel consumption etc. Once you fix the criteria, you can then use various factors to reason which option/route is better. Salary, employment opportunities, prestige of program etc are all factors. What you must optimize on are return of investment, and some measure of happiness for which fit with program and city is a good proxy.
  15. I have been very consistent. The point of my first post was to enumerate various considerations and how to reason about them. If you notice, I never recommended one option over the other, but just gave a list of things to consider. Secondly, the norm being refereed to by us all here is essentially field-specific norms i.e. generalizations of various patterns that are specific to the field like salary, funding opportunities, employment goals, employment hubs (cities) etc. You can call this whatever you want, but I am essentially referring to this. Finally, criteria remain the same across all fields because goals for professional masters programs remain the same across all fields - to land the (dream) job of interest and ear a lot! Essentially, there are only two main criteria - return on investment and fit with the program/city. Both of them depend on all the above factors and more (eg climate). At the core, only two questions are important: (a) will you be happy at the program/city and can it help you land the job you desire; (b) is the cost of attending the program worth the benefits. Factors in the second para can help answer these questions, and the amount of importance given to different factors are subjective/personal.
  16. I think we are saying the same thing with different vocabulary. Norm = specifics of the field. The underlying criteria is essentially return on investment in some form or the other. How much return you can expect from a degree obviously varies across fields and universities - you can call this "norm", specifics, whatever. Another important criteria is fit with program and city in general. How you define and measure "fit" is field specific (or norm). In any case, I believe I have made my point, and have no intention to carry on with this squabble.
  17. @rising_star I agree that I don't know anything about SLP. I just gave my perspective on what factors people consider as important and how to reason about them. Though the specifics may vary, the criteria on which decisions should be made remain constant across fields. Of course, some of these factors are more (or less) relevant for different fields. Just my 2 cents.
  18. Very field specific question, and I wouldn't give a blanket answer like @rising_star. In my field (CS), most students would prefer an unfunded MS from Stanford or CMU over a fully funded offer from a school outside the top 20-25. This is because (a) there are ample TA opportunities in CS, so at least 1 sem is likely funded; (b) summer internships are very common, and can nearly cover 1 sem. Effectively people pay just half the degree cost. Considering the job opportunities and increased salary these degrees command (especially in startup scene), the investment is well worth it. You can try to reason along the same lines. How much is it likely to cost you, and is the additional cost worth it? Some parameters to look are job placements, salaries offered to graduates, opportunities for PhD after masters, and more subjective ones like your compatibility with the program and city in general. Are there any opportunities for funding in later semesters (like TA) and do students effectively use such opportunities etc.
  19. Can't you attend the conference using a B1/B2 visa (business & tourism). This is the visa I used for attending conferences while studying in India, and they are valid for 10 years from date of issuance. If you present the conference invitation letter (I'm assuming you are presenting), you shouldn't have any problems getting a B1/B2 visa. After the conference, apply for F1 again.
  20. Hi @inbrsuan I think you are just mixing up too many factors. First recognize what you want to do - AI, statistics, neuroscience etc etc. All of them are distinct fields, and though they share overlap, they are different in aims and scope. Once you have decided and are sure, you should just pick the appropriate program. For AI, it's EECS, plain and simple. It's better to drop a few years, get the relevant experience, and be in a department that supports your interest; as opposed to just joining something and feeling miserable about research fit mismatch. One way to make sure about research fit is to see what current students in the program are doing, whom they are working with, and whether the adviser is open to taking more students from the program. It's highly likely that you will have a career similar to the current alumni - are you satisfied. These are relevant and important questions, and I wouldn't cut corners like what you seem to be describing. I am not familiar with them. if you are talking about "ischool" programs (ie PhD in information systems and such), they are quite related to ML and AI, but again the aim and scope are different. Also, I think ischool programs are not as strong as EECS, since most students who are admitted to both will prefer EECS. You should make sure that you can work with the adviser you want if you are admitted to such programs. For job in industry, EECS vs ischool isn't a very big difference I think. However, academia will definitely prefer EECS. Again, EECS and Statistics have different aims and scope, different definitions for what constitutes a "successful project", and hence correspondingly slightly different training. Expected deliverable for theoretical statistics will be a strong theorem and proof; for applied statistics clear description of methodology and tabulated results. For EECS, it will be development, analysis, and numerical experiments for proposed algorithms. The nature of problems that interest ML/AI people and Statistics people are also very different. For example, much of recent ML/AI work (deep learning) was motivated by applications in computer vision, which has no presence in statistics. Unless you really care about vision, language, speech, or robotics - you can't hope to advance the state of the art in neural networks. However, statisticians don't work on any of these problems. IMHO, this is why statistics has lost ground to ML, even though it is basically the same thing, but done and conceived very differently.
  21. When I say EECS programs, I mean either EE programs with CS focus (Berkeley, MIT, Washington, Michigan etc.) or CS programs with AI or AI+theory dominant research. There are of course EE programs with focus on devices or CS programs with systems-dominant research (eg Wisconsin Madison) - you shouldn't apply to them. Stanford ICME has an MS option, but their stated interest is in "data science" which is different in both goal and spirit from AI. Caltech CMS or even CNS is probably the closest to what you are looking for. CDO is very very different, it is a traditional computational math program with ME, AE etc dominant research. You should probably be looking at some MS program in https://idss.mit.edu/ I realized the shift in my interest in senior year. So had 1 semester left, during which I completely overloaded myself with ML courses (normal load for final semester is 3 courses, I took 6 - nearly all in ML). I have since been working in an industrial research lab (think Microsoft, Google, IBM etc) related to ML and optimization. It was easier for me to transition to EE than CS-proper. So I applied to only those programs which are either merged with CS (ie EECS) or have close connections with CS and will let me work with a CS adviser. I am most likely to go to Washington which has an awesome ML group and shared resources among CS, EE, and Statistics.
  22. @inbrsuan 1. I don't think you require an undergrad degree in EECS to get into an EECS PhD program (I myself didn't have one). One option might be to get an MS in EECS, and then try for a PhD. This way, you'll have more exposure about the field of EECS and where your interests actually lie. 2. Yes, I am familiar with many computational science, computational math, or applied math programs. I got into the ICES program at UTA and a similar program at UMD last year. After I spoke to the professors, it was clear that the presence of EECS in these programs is minimal. They may list a few people on their website, but in reality the program is mostly MechEs, AEs, and some physicists. The only exceptions I am aware of are Caltech CMS and possibly Stanford ICME, both of which are big on the interface of math, CS, EE, and statistics. Naturally, these are the most competitive programs, and I was rejected. 3. Depending on the area in EECS, PhD students know as much data science as statistics students. Obvious areas are machine learning and signal processing. Students working in control theory, information theory, and communications also take quite a few statistics courses and use them effectively for their work.
  23. For the topics you mention above, EECS (either CS or EE) is the appropriate venue. I think it's worth thinking about the one-line research summary of different fields (in context of areas you mentioned) to determine where you'll fit. EECS looks at neural networks, learning, signal processing, control etc. from a computational viewpoint to endow machines with human-like capabilities. In other words, it's the study of information and decision systems. Computational neuroscience uses primarily biological and some computational tools to understand how the brain works. The difference between the two is that any theory of human brain coming out of neuroscience will be validated on how closely it mimics the human brain as opposed to how computationally sound it is. However, in EECS, as long as the algorithm provides the capabilities as intended, the biological plausibility takes a backseat. Applied math studies mathematical principles to understand how the world works, and hence is primarily a service discipline to physics, chemistry, biology, engineering etc. They generally don't have close connections to the information sciences, which seems to be where your interest lies. Bioinformatics uses computational tools to adress some questions in biology. I wouldn't choose this discipline since it's overspecialization. You can always sit in EECS, and if you find your interests in biology taking over, do a PhD in bioinformatics. On the other hand, sitting in bioinformatics and doing ML for robotics is not possible.
  24. @lovedeep Thanks for the response. Not sure if this is entirely appropriate, but would you happen to know how the industry views EE PhDs who worked on ML? I am very likely to work with only ML-proper professors, and would publish in NIPS, ICML and the like. However, the degree will formally be in EE. I am hoping that as long as it's a PhD in ML and I have papers in ML-proper venues, people wouldn't bother with the home department.
  25. Hi @inbrsuan and @localfdr I was in a similar situation last year where I applied to computational/applied math PhD programs intending to work on AI, machine learning, and neuroscience. I should tell you at the outset that this is an extremely bad idea! I essentially wasted a year, reapplied, and got into programs that would allow me to pursue the above interests. Firstly, understand that applied/computational math and AI are very far from each other. This depends to an extent on the university, but the major focus of applied math programs is to develop mathematical tools for analyzing and simulating problems that occur predominantly in natural sciences (Physics, Chem, Bio) and to a lesser extent, engineering. They totally disregard statistics, especially if there is a department called "statistics" in the university. To get a flavor of what most applied math programs offer, have a look at the program of Northwestern university (a top 5 program). It has nothing to do with ML or AI. Similarly, biostatistics will offer nothing in the vain of AI or ML. It's also hard to market a biostatistics degree for anything other than biostatistics. IMHO, it's overspecializing to such an extent that you will be ineligible for a large chunk of the job market. If it's a real serious AI job, they would naturally hire someone who has specialized in the same, as opposed to someone from biostatistics. The "flavor" offered in statistics is also very different from AI. People in AI study and use statistics to create intelligent agents, whereas people in statistics study the properties of estimators themselves. There is certainly overlap in theory and methods, but the end objective, career trajectories, and "flavor" are very different! If your interest is in Artificial Intelligence, you should apply to basically CS and EE programs. Only these will allow you to work on intelligent agents and techniques for programming them. A very few select programs like CNS at Caltech are also a possibility. I will stay clear of statistics, biostatistics, and applied math. If your interest is data science, statistics will also work in addition to CS and EE. However, data science and AI are very different, both in terms of final objective and flavor.
×
×
  • Create New...

Important Information

This website uses cookies to ensure you get the best experience on our website. See our Privacy Policy and Terms of Use