I once did some non-linear dimensionality reduction on data from multiple EEG electrodes. Used the non-linear dimensionality reduction toolbox which is available in MATLAB. Take my answer with a pinch of salt though, since I don't work on that anymore!

The downside to MATLAB is that it's not freely available. If you're going to spend the time learning a new language, I'd always recommend an open source option. Personally, I use R for most of my statistics, and am starting to get into doing a bit with python. In addition to the fact that you can always get those for free, there are continuously new packages and toolboxes being developed and a strong open source community.

The careful and correct (and wishy-washy and not particularly useful) answer to this question is: It depends on the application. Some languages are better for some applications, and other languages are better for others.

I’m not going to give that answer. Instead I’ll throw caution to the wind and answer recklessly: neuroscientists new to programming should start with Matlab.

Matlab has an enormous user base in the neuroscience community, with lots of readily available scripts. Matlab has a huge number of well-documented and well-supported toolboxes, making it straightforward to use sophisticated algorithms even if you haven’t quite mastered the math behind them 😊. And, most importantly, Matlab’s barriers to entry are really low: even pure novices can learn to write useful scripts in a matter of weeks or sooner.

I don’t think any of this is as true of the alternatives, even Python. Python has a lovely syntax, many available resources, and a certain cachet (all the cool kids seem to be using Python nowadays). But my experience is that it’s harder for novices to get started with Python than with Matlab. And navigating the Python ecosystem (the various packages, Anaconda, Enthought, and so forth) is tougher. With Matlab, there’s one IDE and a successful company employing a talented group of engineers who are being paid to answer your questions and help you out. Open source is great as an ideology, but when you’re stuck on some problem, capitalism has its benefits.

As for the other alternatives, a lot (too much) can be said. R is dominant in data science, but I don’t think it’s quite as full service as the alternatives. For neuroscience generally, I don’t think R has a special place. C is the mother of languages, but for the kind of applications workaday neuroscientists are likely to encounter, it’s not necessary and too demanding (as are other compiled languages).

One need not stick with a single language or programming environment. Over the course of my long and checkered career, I have used, as a day-to-day language, Fortran, C, Igor Pro, C++, and Matlab. I’ve also dabbled in Python and Labview, and looked at R enough to have an opinion about it. Once you get started, going from one to the other is really not that big a deal – despite the differences in syntax or the use of OOP or whatever. The details don’t matter for us. By “us” I mean workaday neuroscientists. We’re trying to be first-rate neuroscientists, not second-rate programmers. The arcana first-rate programmers sometimes worry over (pass by reference versus pass by value!) are not relevant to us amateurs – nor is absolute speed (usually) or scalability. None of us is building a commercial application. We just need something that works for us.

The most often stated objection to Matlab is that it’s expensive. That’s true, but for most of us it isn’t a big issue. If you work at an academic institution or a large company, the odds are very good that your institution or company has a site license you can use. And Matlab isn’t so expensive that it counts as a major expense even for small companies.

My two cents: just start with Matlab and then move on when you’re ready.

[Disclaimer: I don't work for Mathworks. Everything I've written is a personal opinion and does not represent the institution that employs me.]