I will be entering my third year of university in my next academic year, once I've finished my placement year as a web developer, and I would like to hear some opinions on the two modules in the Title.
I'm interested in both, however I want to pick one that will be relevant to my career and that I can apply to systems I develop.
I'm doing an Internet Computing degree, it covers web development, networking, database work and programming. Though I have had myself set on becoming a web developer I'm not so sure about that any more so am trying not to limit myself to that area of development.
I know HCI would help me as a web developer, but do you think it's worth it? Do you think Neural Network knowledge could help me realistically in a system I write in the future?
Thanks.
EDIT:
I thought it would be useful to follow-up with what I decided to do and how it's worked out.
I picked Artificial Neural Networks over HCI, and I've really enjoyed it. Having a peek into cognitive science and machine learning has ignited my interest for the subject area, and I will be hoping to take on a postgraduate project a few years from now when I can afford it.
I have got a job which I am starting after my final exams (which are in a few days) and I was indeed asked if I had done a module in HCI or similar. It didn't seem to matter, as it isn't a front-end developer position!
I would recommend taking the module if you have it as an option, as well as any module consisting of biological computation, it will open up more doors should you want to go onto postgraduate research in the future.
The worthiness depends on three factors:
How familiar are you with the topic already?
How good is the course/class you want to take?
What are your interested in more?
Especially for HCI, there is a broad range of "common sense" information you would also easily obtain from reading a good book or a wider range of articles about it also published on the internet. On the other hand, there indeed exist many deeper insights mostly obtained by Psychology studies. If the course is done right, you can indeed learn a lot about the topic and the real considerations to use for developing an interface.
For Neural Networks, one has to say that this is a typical hype topic. It would be mainly interesting in what application domain the course wants to deal with neural networks. You can be quite sure that you won't program or use any neural networks for web development. On the other hand, if the course is done right, this could be a good opportunity for you to broaden your knowledge. Especially, deepening your understanding about the theory of computer science. This highly depends on how the course is laid out, though.
HCI is a topic which helps your career as a web developer, but only if you feel incompetent in that topic (then it is a must) or it is done very well. Neural Networks is a topic which has more potential of being really interesting hardcore computer science stuff, where you indeed learn a better understanding about something. If you are interested in NN, you should not pass the opportunity to get an education which is not narrowly concentrated on the domain of web development -- and, after all, perhaps find more interest in other stuff (it is always good to know other directions you would perhaps like to go into for the future).
Neural networks sound cool until you read the fine print:
In modern software implementations of
artificial neural networks the
approach inspired by biology has more
or less been abandoned for a more
practical approach based on statistics
and signal processing.
This is something that has mystified me for years. Here you have an amazingly complex and powerful control system (real-world biological neural networks), and an academic discipline that appears to be about modeling these systems in software but that has in reality abandoned that activity.
If you're doing web development, your time is probably better spent in the HCI course.
Go with what interests you the most. The HCI stuff will be much easier to pick up later as needed, you'll likely never get another chance to learn about neural networks!
For prospective employers (at least the good ones!) you need to show a passion and excitement about what you do. I'd sooner hire someone who can enthusiastically talk about neural networks than someone who has an extra credit in HCI.
Unless you want to do the research end of the world, ie, get a Masters/PhD, go HCI.
I studied Neural Computation at University when I studied AI. I now run my own company. The number of times since I studied that I have used my NN skills equals zero. I'm glad I did it, as it was quite fascinating, but I would have found HCI much more useful from the position I'm at now. I think that you'd pick up a lot more insight from an HCI course relevant to the software industry, but if you think you experience should be more on the esoteric/almost arty side of development, go for NN.
Which sounds like more fun? Or, equivalently, which will you work harder at? Pick that one.
Did two courses in NN and some other AI-courses - its fun to poke round with that stuff and I actually managed to implement the stuff in some of the things I've done like face-recognition, and it's useful in some other areas to if you wanna plot your lab data etc. I have never used the NN:s in my web development career though I am sure it could be used for something however what it all really boils down to is to find a client or employee willing pay for it when you can just take the straight path. So I would rather read book about it if I wasn't that hardcore about it.
Fundamental Neural Networks doesn't take to much knowledge in math, and was what I used in my first course.
as a programmer to be you need the knowledge of neural network. if parallel processing is the way to go in hardware then future programmers must be knowledgable in neural network. don't forget that NN works better with noise or imprecise data but other systems may not. Note that most data we use for analysis are sample data which is a fraction of the whole and you could imagine if some in the sample are way off. so you need knowledge of NN if you want to last in computer programming field.
Related
I don't fully understand the complete development cycle and transition from general purpose boards to microcontroller-based serious industrial hardware.
Right now I use RPi or similar general purpose boards and follow this development process:
design hardware with SoC (RPi) in mind.
order/buy hardware
connect main board and peripherals
install OS (almost always Linux)
install libraries, applications, toolchain
create corresponding software with a previously installed toolchain
when the solution is working correctly, move hardware to an appropriate case.
deploy
It may include additional steps but the way I see it, everything is already designed, assembled and test before I even start my development. I only need to choose connect devices, connect wires and create a software. Software is mostly free.
The downside is that such solution lacks quality. I doubt hardware is able to withstand harsh industrial environment. It is also not small enough.
Now I am trying to dive into STM32/Quark/[any microcontroller] world. What I understood so far is:
buy a development board
create software
test
What confuses me is the part when you switch from dev. board to... What?
I mean dev. boards are not designed to be used in a final product, do they?
I guess a need a custom solution.
Do I need to design a custom electronic circuit, produce it by means of an external manufacturer and install my microcontroller and additional ICs there?
I see various presentation's of modern small-size CPUs and I what to know how to develop a device with them.
I want to get an understanding of a full development cycle of an IoT low-power device, but don't know to how to ask correctly.
This isn't really an answer, I don't have enough reputation to simply add a comment, unfortunately. The fact is, answering your question is not simple, there is lot to it. Even after four years of college in Electronic Engineering Technology it was hardly a scratch on what the "real world" is. One learns so much more in the workplace and it never stops.
Anyway, an couple comments.
Microcontrollers are not all equal thus they are not all equally suitable for every task. The development boards and the evaluation boards that are created for microcontrollers are also not all equal and may have focus on applicability to a certain market segment, i.e medical, automotive, consumer IoT, etc..
Long before you start buying a development or evaluation board you have to decide on what is the most appropriate microcontroller. And even, is a microcontroller actually the best choice? ASIC or FPGA? What kind of support chips are needed? How will they interface? Many chip manufactures provide reference designs that can be used a starting point but there are always multiple iterations to actually develop a product. And there is testing, so much testing that we have a "test engineers."
You list development steps is lacking greatly, first and foremost the actual specifications have to be determined for whatever product is being developed and from these specifications appropriate hardware is selected for evaluation. Cost is always a driving factor and so fitting the right device to the product and not going overkill is very important. A lot of time is spent evaluating possible products from their datasheets to determine what products seem to be the right fit. Then there are all the other factors such as the experience with the device/brand/IDE etc. All of that adds to cost of development plus much more.
You mention software(firmware) is free. No, software and firmware are never free. Someone has to develop it and that takes time and time is money. Someone has to debug it. Debugging takes time. Debugging hardware is expensive. Don't forget the cost of the IDE, commercial IDEs are not cheap and some are much more expensive than others and can greatly effect the cost to develop. Compare the cost of buying an IDE to develop for a Maxim Integrated MAXQ MCU to any of the multitude of AVR or ARM IDE choices. Last I checked there were only two companies making an IDE for the MAXQ MCUs. What resources are available to assist in your design you can use with minimal or no licensing fees? This is the tip of the iceberg. There is a lot to it, software/firmware is not "free."
So fast forward a year, you finished a design and it seems to pass all internal testing. What countries are you marketing in? Do you need UL, CE or other certifications? I hope you designed your board to take into account EMI mitigation. Testing that in-house isn't cheap, certification testing isn't either, and failing is even more costly.
These are a very, very, few things that seem to be often ignored by hobbyists and makers thinking they can up with the next best thing and make a killing in some emerging market.
I suggest you do a search on Amazon for "engineering development process", "lean manufacturing", "design for manufacturability", "design for internet of things", "engineering economics" and plan on spending some money to buy some books and time to read up on what the design process is from the various points of view that have to be considered.
Now maybe you mean to develop and deploy for your own use and cost, manufacturability, marketability and the rest are not so important to you. I still suggest you do some Amazon research and pick up some well recommended reading/learning material there that is pertinent to you actual goals. You may want to avoid textbooks, as they generally are more useful when accompanied with class lectures - plus they tend cost much more than the books written for the non-student.
Don't exclude the option of hiring the design and development of an idea out to a firm that specializes in it. It is expensive but is it more expensive than one-up in-house design and development? Probably not. How fast do you actually need your device? Will you lost if someone beats you to market? So many things to consider I could spend hours on this just pointing out things that may, or may not, even be relevant to you depending on what you actual goal is.
TL;DR There is a great deal to the design and development of a product be it marketed to consumers (such as IoT) or to industry. Specifications come first. The exact develop process is going to be influenced by the specifications. Your question cannot be easily answered and certainly not without knowing much more about your end goal. Amazon is a good source of books for really general questions like this.
Thanks for reading.
I want to implement a Baysian Network using the Matlab's BNT toolbox.The thing is, I can't find "easy" examples, since it's the first time I have to deal with BN.
Can you propose some possible applications, (with not many nodes) please ^^ ?
Have a look at Tom Mitchell's "Machine Learning" book, which covers the subject starting with small, simple examples. I suspect there are many course slides you could access online which also give simple examples.
I think it helps to start with higher level tools to get a feel for how to construct networks before constructing them in code. Having a UI also allows you to play with the network and get a feel for the way the networks behave (propagation, explaining away, etc).
For example have a look at the free Genie (http://genie.sis.pitt.edu) and its samples, and/or the 50 node limited Hugin-Lite (http://www.hugin.com/productsservices/demo/hugin-lite) with it's sample networks. You can then check your BNT implementations to make sure they verify against the software packages.
Edit: I forgot to mention Netica which is another BN/Influence diagram software package which I think has the biggest selection of examples http://www.norsys.com/netlibrary/index.htm.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 2 years ago.
Improve this question
To learn FPGA programming, I plan to code up a simple Neural Network in FPGA (since it's massively parallel; it's one of the few things where an FPGA implementation might have a chance of being faster than a CPU implementation).
Though I'm familiar with C programming (10+ years). I'm not so sure with FPGA development stuff. Can you provide a guided list of what I should do / learn / buy?
Thanks!
Necroposting, but for others like me that come across this question there is an in-depth, though old, treatment of implementing neural networks using FPGAs
It's been three years since I posted this, but it is still being viewed so I thought I'd add another two papers from last year I recently found.
The first talks about FPGA Acceleration of Convolutional Neural Networks. Nallatech performed the work. It's more marketing that an academic paper, but still an interesting read, and might be a jumping off point for someone interesting in experimenting. I am not connected to Nallatech in any way.
The second paper came out of the University of Birmingham, UK, written by Yufeng Hao. It presents A General Neural Network Hardware Architecture on FPGA.
Most attempts at building a 'literal' neural network on an FPGA hit the routing limits very quickly, you might get a few hundred cells before P&R pulls takes longer to finish than your problem is worth waiting for. Most of the research into NN & FPGA takes this approach, concentrating on a minimal 'node' implementation and suggesting scaling is now trivial.
The way to make a reasonably sized neural network actually work is to use the FPGA to build a dedicated neural-network number crunching machine. Get your initial node values in a memory chip, have a second memory chip for your next timestamp results, and a third area to store your connectivity weights. Pump the node values and connection data through using techniques to keep the memory buses saturated (order node loads by CAS line, read-ahead using pipelines). It will take a large number of passes over the previous dataset as you pair off weights with previous values, run them through DSP MAC units to evaluate the new node weights, then push out to the result memory area once all connections evaluated. Once you have a whole timestep finished, reverse the direction of flow so the next timestep writes back to the original storage area.
I want to point out a potential issue with implementing a Neural Network in FPGA. FPGAs have limited amount of routing resources. Unlike logic resources (flops, look-up tables, memories), routing resources are difficult to quantify. Maybe a simple Neural Network will work, but a "massively parallel" one with mesh interconnects might not.
I'd suggest starting with a simple core from OpenCores.org just to get familiar with FPGA flow, and then move on to prototyping a Neural Network. Downloading free Xilinx WebPack, which includes ISIM simulator, is a good start. Later on you can purchase a cheap dev. board with a small FPGA (e.g. Xilinx Spartan 3) to run your designs on.
A neural network may not be the best starting point for learning how to program an FPGA. I would initially try something simpler like a counter driving LEDs or a numeric display and build up from there. Sites that may be of use include:
http://www.fpga4fun.com/ - Excellent examples of simple projects and some boards.
http://opencores.org/ - Very useful reference code for many interfaces, etc...
You may also like to consider using a soft processor in the FPGA to help your transition from C to VHDL or Verilog. That would allow you to move small code modules from one to the other to see the differences in hardware. The choice of language is somewhat arbitrary - I code in VHDL (syntactically similar to ADA) most of the time, but some of my colleagues prefer Verilog (syntactically similar to C). We debate it once in a while but really it's personal choice.
As for the buyers / learners guide, you need:
Patience :) - The design cycle for FPGAs is significantly longer than for software due to the number of extra 'free parameters' in the build, so don't be surprised if it takes a while to get designs working exactly the way you want.
A development board - For learning, I would buy one from one of the three bigger FPGA vendors: Xilinx, Altera or Lattice. My preference is Xilinx at the moment but all three are good. For learning, don't buy one based on the higher-end parts - you don't need to when starting using FPGAs. For Xilinx, get one based on the Spartan series such as the SP601 (I have one myself). For Altera, buy a Cyclone one. The development boards will be significantly cheaper than those for the higher-end parts.
A programming cable - Most companies produce a USB programming cable with a special connector to program the devices on the board (often using JTAG). Some boards have the programming interface built in (such as the SP601 from Xilinx) so you don't need to spend extra money on it.
Build tools - There are many varieties of these but most of the big FPGA vendors provide a solution of their own. Bear in mind that the tools are only free for the smaller lower-performance FPGAs, for example the Xilinx ISE Webpack.
The software comprises stages with which you may not be familiar having come from the software world. The specifics of the tool flow are always changing, but any tool you use should be able to get from your code to your specific device. The last part of this design flow is normally provided by the FPGA vendor because it's hardware-specific and proprietary.
To give you a brief example, the software you need should take your VHDL and Verilog code and (this is the Xilinx version):
'Synthesise' it into constructs that match the building blocks available inside your particular FPGA.
'Translate & map' the design into the part.
'Place & route' the logic in the specific device so it meets your timing requirements (e.g. the clock speed you want the design to run at).
Regardless of what Charles Stewart says, Verilog is a fine place to start. It reminds me of C, just as VHDL reminds me of ADA. No one uses Occam in industry and it isn't common in universities.
For a Verilog book, I recommend these especially Verilog HDL. Verilog does parallel work trivially, unlike C.
To buy, get a relatively cheap Cyclone III eval board from [Altera] or Altera's 3 (e.g. this Cyclone III one with NIOS for $449 or this for $199) or Xilinx.
I'll give you yet a third recommendation: Use VHDL. Yes, on the surface it looks like ADA. While Verilog bears a passing resemblance to C. However, with Verilog you only get the types that come with it out of the box. With VHDL you can define your own new types which lets you program at a higher level (still RTL, of course). I'm pretty sure the Xilinx and Altera free tools support both VHDL and Verilog. "A Designers Guide to VHDL" by Ashenden is a good VHDL book.
VHDL has a standard fixed-point math package which can make NN implementation easier.
It's old, because I haven't thought much about FPGAs in nearly 20 years, and it uses a concurrent programming language that is rather obscure, but Page & Luk, 1991, Compiling Occam into FPGAs covers some crucial topics in a nice way, enough, I think, for your purposes. Two links for trying stuff out:
KRoC is an actively maintained, linux-based Occam compiler, which I know has an active user base.
Roger Peel has a logic synthesis page that has some documentation of his linux-based workflow from Occam code synthesis through to FPGA I/O.
Occam->FPGA isn't where the action is, but it may be a much better place to start than, say, Verilog.
I would recommend looking into xilinx high-level synthesis, especially if you are coming from a C background. It abstracts away the technical details in using a hdl so the designer can focus on the algorithmic implementation.
The are restriction in the type of C code you can write. For example, you can't use dynamically sized data structures, as that would infer dynamically sized hardware.
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 7 years ago.
Improve this question
In my copious free time, I collaborate with a number of scientists (mostly biologists) who develop software, databases, and other tools related to the work they do.
Generally these projects are built on a one-off basis, used in-house, and eventually someone decides "oh, this could be useful to other people," so they release a binary or slap a PHP interface onto it and shove it onto the web. However, they typically can't be bothered to make their source code or dumps of their databases available for other developers, so in practice, these projects usually die when the project for which the code was written comes to an end or loses funding. A few months (or years) later, some other lab has a need for the same kind of tool, they have to repeat the work that the first lab did, that project eventually dies, lather, rinse, repeat.
Does anyone have any suggestions for how to persuade people whose primary job isn't programming that it's of benefit to their community for them to be more open with the tools they've built?
Similarly, any advice on how to communicate the idea that version control, bug tracking, refactoring, automated tests, continuous integration and other common practices we professional developers take for granted are good ideas worth spending time on?
Unfortunately, a lot of scientists seem to hold the opinion that programming is a dull, make-work necessary evil and that their research is much more important, not realising that these days, software development is part of scientific research, and if the community as a whole were to raise the bar for development standards, everyone would benefit.
Have you ever been in a situation like this? What worked for you?
Software Carpentry sounds like a match for your request:
Overview
Many scientists and engineers spend
much of their lives programming, but
only a handful have ever been taught
how to do this well. As a result, they
spend their time wrestling with
software, instead of doing research,
but have no idea how reliable or
efficient their programs are.
This course is an intensive
introduction to basic software
development practices for scientists
and engineers that can reduce the time
they spend programming by 20-25%. All
of the material is open source: it may
be used freely by anyone for
educational or commercial purposes,
and research groups in academia and
industry are actively encouraged to
adapt it to their needs.
Let me preface this by saying that I'm a bioinformatician, so I see the things you're talking about all the time. There's some truth to the fact that many of these people are biologists-turned-coders who just don't have the exposure to best practices.
That said, the core problem isn't that these people don't know about good practices, or don't care. The problem is that there is no incentive for them to spend more time learning software engineering, or to clean up their code and release it.
In an academic research setting, your reputation (and thus your future job prospects) depends almost entirely on the number and quality of publications that you've contributed to. Publications on methods or new algorithms are not given as much respect as those that report new biological findings. So after I do a quick analysis of a dataset, there's very little incentive for me to spend lots of time cleaning up my code and releasing it, when I could be moving on to the next dataset and making more biological discoveries.
I'll also note that the availability of funding for computational development is orders of magnitude less than that available for doing the biology. In a climate where only 10% of submitted grants are getting funded, scientists don't have the luxury of taking time to clean and release their code, when doing so doesn't help them keep their lab funded.
So, there's the problem in a nutshell. As a bioinformatician, I think it's perverse and often frustrating.
That said, there is hope for the future. With second-and-third generation sequencing, in particular, biology is moving into the realm of high-throughput discovery, where data mining and solid computational pipelines become integral to the success of the science. As that happens, you'll see more and more funding for computational projects, and more and more real software engineering happening.
It's not exactly simple, but demonstration by example would probably drive the point home most effectively - find a task the researcher needs done, find someone who did take the time to make a tool w/source available, and point out how much time the researcher could save as a result due to having that tool available - then point out that they could give back to the community in the same fashion.
In effect, what you are asking them to do is become professional developers (with their copious free time), in addition to their chosen profession. Their reluctance is understandable.
Does anyone have any suggestions for how to persuade people whose primary job isn't programming that it's of benefit to their community for them to be more open with the tools they've built?
Give up. Seriously, this is like teaching a pig to sing. (I can say this because I used to be a physicist so I know what they're like.)
The real issue is that your colleagues are rewarded for scientific output measured in publications, not software. It's hard enough in computer science to get recognized for building software; in the other sciences, it's nearly impossible.
You can't sell good development practice to your biology friends on the grounds that "it's good for you." They're going to ask "should I invest effort in learning about good software practice, or should I invest the same effort to publish another biology paper?" No contest.
Maybe framing it in terms of academic/intellectual responsibility would help, to a degree - sharing your source is, in many ways, like properly citing your sources or detailing your research methodology. There are similar arguments to be made for some of the "professional software developer" behaviors you'd like to encourage, though I think releasing the code is probably an easier sell on these grounds than other things which could require significantly more work.
Actually, asking any busy project team to include in their schedule time for making their software suitable for adoption by another team is extremely hard in my experience.
Doing extra work for the public good is a big ask.
I've seen a common pattern of "harvesting" after the project is complete, reflecting that immediate coding for reuse tends to get lost in the urgency of the day.
The only avenue I can think of is if the reuse is within an organisation with a budget for a "hunter gatherer", someone whose reason for being there is IT.
You may be on more of a win for things such as unit tests because they have immediate payback for the development.
For one thing, could we please stop teaching biologists Perl? Teaching non-professional programmers a write-only language is practically guaranteed to lead to unmaintainable, throw-away code. Python fills the same niche, is just as easy to learn (it's even used to teach kids programming!), and is much more readable.
Draw parallels with statistics. Stats is a crucial part of scientific research, and one where the only sensible advice is: either learn to do it properly, or get an expert to do it for you. Incorrectly-done stats can completely undermine a paper, just as badly-written code can completely undermine a public database or web resource.
PS: This blog is very good, but getting them to read it will be an uphill struggle: Programming for Scientists
Chris,
I agree with you to a degree, but in my experience what ends up happening is that in their eagerness to publish you end up with too many "me too" codes and methods, which don't really add to the quality of science. If there was a little more thought about open sourcing code and encouraging others to contribute (without necessarily getting publications out of it) then everyone would benefit.
Definitely agree that a separation between the scientific programmers and the software engineers is a good thing, especially for production applications. But even for scientific programming, the quality of my code would have been so much better if I had followed good practices at the time.
In my experience the best way of getting people to program cleanly is to show a good example when you're working with them.
eg: "I never spend hopeless days debugging my code because the first things I code are automated unit tests that will pinpoint problems when they are small and easily detectable"
or: "I'm very bad at keeping track of versions of things, but sometimes my new code does break what did work before. So I use svn/git/dropbox to keep track of things for me"
In my experience that kind of statement can raise the interest of "biologists that learned how to script".
And if you need to collaborate on a bigger project, make it clear that you have more experience and that everything will go more smoothly if things are done your way.
Regarding publication of code, current practice is indeed frustrating. I would like to see a new journal like Source Code for Biology and Medecine, where code is peer-reviewed and can be published, but that has no (or very low) publication costs. Putting code on sourceforge or others is indeed not "scientifically worth it" because it doesn't make a line on your publication list, and most code is not revolutionary enough to warrant paying $1,000 for publication in Source Code for Biology and Medecine or PLoS One...
You could have them use a content management system, like Joomla. That way they only push content and not code.
I wouldn't so much persuade as I would streamline the process. Document it clearly, make video tutorials and bundle some kind of tool chain that makes it ridiculously easy to get source repositories set up without requiring them to become experts in something that isn't their main field.
Take a really good programmer who already knows best practices, ask your scientists to teach him what they need and what they do, eventually the programmer will have minimum domain knowledge (I suspect it takes between 1 and 3 years depending on the domain) to do what scientists asks for.
Developers always learn another domain of competency, because most of their programs are not for developers, so they need to know what the "client" do.
To be devil's advocate, is teaching scientists to be good software engineers the right thing to do? Software in research is usually very purpose specific - sometimes to the point where a piece of code needs to run successfully only once on a single data set. The results then feed into a publication and the goal is met. And there's a high risk that your technique or algorithm will be superseded by a better one in short order. So, there's a real risk that effort spent producing sparkling code will be wasted.
When you're frustrated by wading through a swamp of ill-formed perl code, just think that the code you're looking at is one of the rare survivors. Mountains of such code has been written, used a few times, then discarded never to see the light of day again.
I guess I'm just saying there's a big place in research for smelly heinous one-off prototype code. There are good reasons why such code exists. It may not be pretty, but if it gets the job done, who cares? We can always hire a software engineer to write the production-ready version later, IF it turns out to be justified, and let our scientists move on.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
Questions asking us to recommend or find a tool, library or favorite off-site resource are off-topic for Stack Overflow as they tend to attract opinionated answers and spam. Instead, describe the problem and what has been done so far to solve it.
Closed 8 years ago.
Improve this question
I'm really interested in Artificial Neural Networks, but I'm looking for a place to start.
What resources are out there and what is a good starting project?
First of all, give up any notions that artificial neural networks have anything to do with the brain but for a passing similarity to networks of biological neurons. Learning biology won't help you effectively apply neural networks; learning linear algebra, calculus, and probability theory will. You should at the very least make yourself familiar with the idea of basic differentiation of functions, the chain rule, partial derivatives (the gradient, the Jacobian and the Hessian), and understanding matrix multiplication and diagonalization.
Really what you are doing when you train a network is optimizing a large, multidimensional function (minimizing your error measure with respect to each of the weights in the network), and so an investigation of techniques for nonlinear numerical optimization may prove instructive. This is a widely studied problem with a large base of literature outside of neural networks, and there are plenty of lecture notes in numerical optimization available on the web. To start, most people use simple gradient descent, but this can be much slower and less effective than more nuanced methods like
Once you've got the basic ideas down you can start to experiment with different "squashing" functions in your hidden layer, adding various kinds of regularization, and various tweaks to make learning go faster. See this paper for a comprehensive list of "best practices".
One of the best books on the subject is Chris Bishop's Neural Networks for Pattern Recognition. It's fairly old by this stage but is still an excellent resource, and you can often find used copies online for about $30. The neural network chapter in his newer book, Pattern Recognition and Machine Learning, is also quite comprehensive. For a particularly good implementation-centric tutorial, see this one on CodeProject.com which implements a clever sort of network called a convolutional network, which constrains connectivity in such a way as to make it very good at learning to classify visual patterns.
Support vector machines and other kernel methods have become quite popular because you can apply them without knowing what the hell you're doing and often get acceptable results. Neural networks, on the other hand, are huge optimization problems which require careful tuning, although they're still preferable for lots of problems, particularly large scale problems in domains like computer vision.
I'd highly recommend this excellent series by Anoop Madhusudanan on Code Project.
He takes you through the fundamentals to understanding how they work in an easy to understand way and shows you how to use his brainnet library to create your own.
Here are some example of Neural Net programming.
http://www.codeproject.com/KB/recipes/neural_dot_net.aspx
you can start reading here:
http://web.archive.org/web/20071025010456/http://www.geocities.com/CapeCanaveral/Lab/3765/neural.html
I for my part have visited a course about it and worked through some literature.
Neural Networks are kind of declasse these days. Support vector machines and kernel methods are better for more classes of problems then backpropagation. Neural networks and genetic algorithms capture the imagination of people who don't know much about modern machine learning but they are not state of the art.
If you want to learn more about AI and machine learning, I recommend reading Peter Norvig's Artificial Intelligence: A Modern Approach. It's a broad survey of AI and lots of modern technology. It goes over the history and older techniques too, and will give you a more complete grounding in the basics of AI and machine Learning.
Neural networks are pretty easy, though. Especially if you use a genetic algorithm to determine the weights, rather then proper backpropagation.
I second dwf's recommendation of Neural Networks for Pattern Recognition by Chris Bishop. Although, it's perhaps not a starter text. Norvig or an online tutorial (with code in Matlab!) would probably be a gentler introduction.
A good starter project would be OCR (Optical Character Recognition). You can scan in pages of text and feed each character through the network in order to perform classification. (You would have to train the network first of course!).
Raul Rojas' book is a a very good start (it's also free). Also, Haykin's book 3rd edition, although of large volume, is very well explained.
I can recommend where not to start. I bought An Introduction to Neural Networks by Kevin Gurney which has good reviews on Amazon and claims to be a "highly accessible introduction to one of the most important topics in cognitive and computer science". Personally, I would not recommend this book as a start. I can comprehend only about 10% of it, but maybe it's just me (English is not my native language). I'm going to look into other options from this thread.
http://www.ai-junkie.com/ann/evolved/nnt1.html is a clear introduction to multi-layers perceptron, although it does not describe the backpropagation algorithm
you can also have a look at generation5.org which provides a lot of articles about AI in general and has some great texts about neural network
If you don't mind spending money, The Handbook of Brain Theory and Neural Networks is very good. It contains 287 articles covering research in many disciplines. It starts with an introduction and theory and then highlights paths through the articles to best cover your interests.
As for a first project, Kohonen maps are interesting for categorization: find hidden relationships in your music collection, build a smart robot, or solve the Netflix prize.
I think a good starting point would always be Wikipedia. There you'll find some usefull links to documentations and projects which use neural nets, too.
Two books that where used during my study:
Introductional course: An introduction to Neural Computing by Igor Aleksander and Helen Morton.
Advanced course: Neurocomputing by Robert Hecht-Nielsen
I found Fausett's Fundamentals of Neural Networks a straightforward and easy-to-get-into introductory textbook.
I found the textbook "Computational Intelligence" to be incredibly helpful.
Programming Collective Intelligence discusses this in the context of Search and Ranking algorithms. Also, in the code available here (in ch.4), the concepts discussed in the book are illustrated in a Python example.
I agree with the other people who said that studying biology is not a good starting point... because theres a lot of irrelevant info in biology. You do not need to understand how a neuron works to recreate its functionality - you only need to simulate its actions. I recomend "How To Create A Mind" by Ray Kurzweil - it goes into the aspect of biology that is relevant for computational models, (creating a simualted neuron by combining several inputs and firing once a threshhold is reached) but ignores the irrelvant stuff like how the neuron actually adds thouse inputs togeather. (You will just use + and an inequality to compare to a threshold, for example)
I should also point out that the book isn't really about 'creating a mind' - it only focuses on heirarchical pattern recognition / the neocortex. The general theme has been talked about since the 1980s I beleive, so there are plenty of older books that probably contain slightly dated forms of the same information. I have read older documents stating that the vision system, for example, is a multi layered pattern recognizer. He contends that this applies to the entire neocortex. Also, take his 'predictions' with a grain of salt - his hardware estimates are probably pretty accurate, but i think he underestimates how complicated simple tasks can be (ex: driving a car). Granted, he has seen a lot of progress (and been part of some of it) but i still think he is over optimistic. There is a big difference between an AI car being able to drive a mile successfully 90% of the time, when compared to the 99.9+% that a human can do. I don't expect any AI to be truly out driving me for atleast 20 years... (I don't count BMWs track cars that need to be 'trained' on the actual course, as they aren't really playing the same game)
If you already have a basic idea of what AI is and how it can be modeled, you may be better off skipping to something more technical.
If you want to do quickly learn about applications of some neural network concepts on a real simulator, there is a great online book (now wiki) called 'Computational Cognitive Neuroscience' at http://grey.colorado.edu/CompCogNeuro/index.php/CCNBook/Main
The book is used at schools as a textbook, and takes you through lots of different brain areas, from individual neurons all the way to higher-order executive functioning.
In addition, each section is augmented with homework 'projects' that are already down for you. Just download, follow the steps, and simulate everything that the chapter talked about. The software they use, Emergent, is a little finnicky but incredibly robust: its the product of more than 10 years of work I believe.
I went through it in an undergrad class this past semester, and it was great. Walks you through everything step by step