Native support for differential programming has been added to Swift for the Swift for Tensorflow project. Julia has similar with Zygote.
What exactly is differentiable programming?
what does it enable? Wikipedia says
the programs can be differentiated throughout
but what does that mean?
how would one use it (e.g. a simple example)?
and how does it relate to automatic differentiation (the two seem conflated a lot of the time)?
I like to think about this question in terms of user-facing features (differentiable programming) vs implementation details (automatic differentiation).
From a user's perspective:
"Differentiable programming" is APIs for differentiation. An example is a def gradient(f) higher-order function for computing the gradient of f. These APIs may be first-class language features, or implemented in and provided by libraries.
"Automatic differentiation" is an implementation detail for automatically computing derivative functions. There are many techniques (e.g. source code transformation, operator overloading) and multiple modes (e.g. forward-mode, reverse-mode).
Explained in code:
def f(x):
return x * x * x
∇f = gradient(f)
print(∇f(4)) # 48.0
# Using the `gradient` API:
# ▶ differentiable programming.
# How `gradient` works to compute the gradient of `f`:
# ▶ automatic differentiation.
I never heard the term "differentiable programming" before reading your question, but having used the concepts noted in your references, both from the side of creating code to solve a derivative with Symbolic differentiation and with Automatic differentiation and having written interpreters and compilers, to me this just means that they have made the ability to calculate the numeric value of the derivative of a function easier. I don't know if they made it a First-class citizen, but the new way doesn't require the use of a function/method call; it is done with syntax and the compiler/interpreter hides the translation into calls.
If you look at the Zygote example it clearly shows the use of prime notation
julia> f(10), f'(10)
Most seasoned programmers would guess what I just noted because there was not a research paper explaining it. In other words it is just that obvious.
Another way to think about it is that if you have ever tried to calculate a derivative in a programming language you know how hard it can be at times and then ask yourself why don't they (the language designers and programmers) just add it into the language. In these cases they did.
What surprises me is how long it to took before derivatives became available via syntax instead of calls, but if you have ever worked with scientific code or coded neural networks at at that level then you will understand why this is a concept that is being touted as something of value.
Also I would not view this as another programming paradigm, but I am sure it will be added to the list.
How does it relate to automatic differentiation (the two seem conflated a lot of the time)?
In both cases that you referenced, they use automatic differentiation to calculate the derivative instead of using symbolic differentiation. I do not view differentiable programming and automatic differentiation as being two distinct sets, but instead that differentiable programming has a means of being implemented and the way they chose was to use automatic differentiation, they could have chose symbolic differentiation or some other means.
It seems you are trying to read more into what differential programming is than it really is. It is not a new way of programming, but just a nice feature added for doing derivatives.
Perhaps if they named it differentiable syntax it might have been more clear. The use of the word programming gives it more panache than I think it deserves.
EDIT
After skimming Swift Differentiable Programming Mega-Proposal and trying to compare that with the Julia example using Zygote, I would have to modify the answer into parts that talk about Zygote and then switch gears to talk about Swift. They each took a different path, but the commonality and bottom line is that the languages know something about differentiation which makes the job of coding them easier and hopefully produces less errors.
About the Wikipedia quote that
the programs can be differentiated throughout
At first reading it seems nonsense or at least lacks enough detail to understand it in context which is why I am sure you asked.
In having many years of digging into what others are trying to communicate, one learns that unless the source has been peer reviewed to take it with a grain of salt, and unless it is absolutely necessary to understand, then just ignore it. In this case if you ignore the sentence most of what your reference makes sense. However I take it that you want an answer, so let's try and figure out what it means.
The key word that has me perplexed is throughout, but since you note the statement came from Wikipedia and in Wikipedia they give three references for the statement, a search of the word throughout appears only in one
∂P: A Differentiable Programming System to Bridge Machine Learning and Scientific Computing
Thus, since our ∂P system does not require primitives to handle new
types, this means that almost all functions and types defined
throughout the language are automatically supported by Zygote, and
users can easily accelerate specific functions as they deem necessary.
So my take on this is that by going back to the source, e.g. the paper, you can better understand how that percolated up into Wikipedia, but it seems that the meaning was lost along the way.
In this case if you really want to know the meaning of that statement you should ask on the Wikipedia talk page and ask the author of the statement directly.
Also note that the paper referenced is not peer reviewed, so the statements in there may not have any meaning amongst peers at present. As I said, I would just ignore it and get on with writing wonderful code.
You can guess its definition by application of differentiability.
It's been used for optimization i.e. to calculate minimum value or maximum value
Many of these problems can be solved by finding the appropriate function and then using techniques to find the maximum or the minimum value required.
While thinking about software-engineering in general I came across the question why we don't see any improvements in the way we write/document code.
Think about it: There has not been a revolutionary improvement since we've moved from punch cards to text editing. The last improvement I've seen is syntax highlighting and context sensitive help (e.g. Intellisense or ctags). Not something I would call revolutionary.
That makes me wonder: Why is it so?
I'll start with something I miss badly:
Lots of my code deals with geometry.
For documentation describing geometric relationships always ends up in a big heap of hard to read mathematical stuff (due to the lack of proper equation type-setting in ASCII). However, if I could embed a little drawing or scribble into the code everything would be much easier, neater and better to be understood.
What can you think up that would make your coding/text editing/documention tasks easier?
I'm surprised that nobody has yet mentioned No Silver Bullet. In 1986 (!), Frederick Brooks predicted that:
There is no single development, in either technology or management technique, which by itself promises even one order-of-magnitude [tenfold] improvement within a decade in productivity, in reliability, in simplicity. [...] We cannot expect ever to see two-fold gains every two years."
And in 23 years, he's been proven right. We've come up with a number of things such as syntax highlighting and Intellisense which have improved productivity significantly, but certainly not by an order of magnitude. As time marches on, we'll continue to make several incremental improvements, but the fact is there is no silver bullet: there's not going to be some magical revelation in the way we write code that will improve productivity by an order of magnitude.
I'm suprised that no one seems to have mentioned Donald Knuth's seminal Literate Programming - write your code as if it were a book or a scientific paper.
There has not been a revolutionary improvement since we've moved from punch cards to text editing
Never used a line editor, have you?
But seriously, text (especially in the representations chosen for modern languages) is
easily processed
fairly easy to specify
information dense
precise
Anything that comes along to replace it has to be a net win across all four of those properties. Not easy.
I disagree. We do have changes, small, but changes.
How common is the "for each" construct? Compare it to 20 years ago. How about the Domain Specific Languages movement? What about the idea that we should code in layers? How about Behavior Driven Development? Coding by complying to a specification... which writes a nice document as output when all runs fine. How about the standarization of regular expressions? PCRE. What about Alan Kay's group's DSL related work on "Moore's Law for Software", which explored a more advanced implementation of Cairo and generated TCP/IP code using diagrams from RFCs?
Documentation is a two way dialog. Both as code being more understandable and people learning this special language. You wouldn't say that German needs documentation if you know German. I know natural languages are very far away from computer languages, but there's a movement to make code more expressive. It's not about the new tools, it's about how we are coding.
One thing I've done recently in some of the more math-heavy sections of my application is to include the LaTeX markup for the particular equation as a comment/docstring. Right now, I just copy-paste into an online equation editor, but it would be very helpful to see the formula itself (with things like Greek letters and sub/superscripts) rather than a bunch of ASCII code.
Source Code In Database. In a nutshell, source code is parsed and put into a database. You'd then need an integrated IDE to view and edit the code, but at this point, syntax is decoupled from format. YOUR IDE could show you a program in a way that's completely different from someone else's, tuned to the task you're working on. I'd list some specific examples, but that article covers pretty much everything.
I'm surprised nobody mentioned it - javadoc is basically HTML, so there's nothing preventing you from embedding images (or anything else) in code. Simple, effective and ubiquitous, it's one of the things Java did right.
DrScheme let's you do these things. Here's the things you can insert from the PLT website:
http://docs.plt-scheme.org/drscheme/Menus.html#(part._.Insert)
3.1.6 Insert
Insert Comment Box : Inserts a box that is ignored by DrScheme; use it to write comments for people who read your program.
Insert Image... : Opens a find-file dialog for selecting an image file in GIF, BMP, XBM, XPM, PNG, or JPG format. The image is treated as a value.
Insert Fraction... : Opens a dialog for a mixed-notation fraction, and inserts the given fraction into the current editor.
Insert Large Letters... : Opens a dialog for a line of text, and inserts a large version of the text (using semicolons and spaces).
Insert λ : Inserts the symbol λ (as a Unicode character) into the program. The λ symbol is normally bound the same as lambda.
Insert Java Comment Box : Inserts a box that is ignored by DrScheme. Unlike the Insert Comment Box menu item, this is designed for the ProfessorJ language levels. See ProfessorJ.
Insert Java Interactions Box : Inserts a box that will allows Java expressions and statements within Scheme programs. The result of the box is a Scheme value corresponding to the result(s) of the Java expressions. At this time, Scheme values cannot enter the box. The box will accept one Java statement or expression per line.
Insert XML Box : Inserts an XML; see XML Boxes and Scheme Boxes for more information.
Insert Scheme Box : Inserts a box to contain Scheme code, typically used inside an XML box; see XML Boxes and Scheme Boxes.
Insert Scheme Splice Box : Inserts a box to contain Scheme code, typically used inside an XML box; see also XML Boxes and Scheme Boxes.
Insert Pict Box : Creates a box for generating a Slideshow picture. Inside the pict box, insert and arrange Scheme boxes that produce picture values.
You also insert your unit tests with the code that you're testing. Pretty neat stuff.
I think integrated IDEs with semantic highlighting and **semantically-constrained suggestions* (a la IDEA or Eclipse) are a huge advancement.
But that happened 8-10 years ago.
Template-based programming feels useful never seems to catch on. Recently I was impressed with a demo of the Meta-programming system, which leverages the interactive nature of the IDE to simplify the task of writing templates and what are (essentially) type-aware macros.
Meta-programming might help you define geometry-based macros that would substitute for a number of lines of code. I could imagine something that let you embed a more-readable 'math language' inside Java, and then parses its contents into something machine-readable.
I'd say version control was a pretty huge leap in how we work. The ability to keep a full record of every change anyone has made to the codebase, and to revert changes where necessary, has made a big difference.
I certainly respect Fred Brooks' argument No Silver Bullet, but I think the way we write code is nowhere near optimal, so there is lots more room for improvement. I tried to explain this in my book.
We're all familiar with "code golf", where you compete relentlessly to minimize something. That is a good way to approach the minimum possible value of that something.
What's great about this is that you are allowed, even encouraged, to break from traditions, prior conceptions, accepted wisdom, in the quest for winning. In short, you learn new things.
If the measure to be minimized is wall-clock execution time, you can do aggressive optimization.
If the measure is source code size (lines or characters) you get "code golf".
The measure I like best is "edit count". That is, given a code base, suppose a new requirement comes along. That requirement is implemented, completely, by editing the code base. Then a "diff" is done from old to new code base. The number of differences found is the edit count. Averaged over the set of likely new functional requirements, that is the measure to minimize.
If this is done aggressively, being free to contradict all conventional wisdom, the code base approaches a state I would call a domain-specific language (DSL). In this language, concepts expressed in code are in nearly 1-1 correspondence with problem-oriented concepts. In this state, it is not easy for the source code to be self-inconsistent (i.e. have bugs) because the fewer edits that have to be made to the source code, the fewer chances there are to make a mistake. It's also the case that such code tends to be short. But unlike "code golf" it tends to be very clear, because it maps the problem concepts so clearly.
So, tools and techniques that help in minimizing edit count can, in my opinion, be considered "silver bullets". DSL is one such. Code generation is another. My favorite optimization technique is another. For coding dynamically changing UIs there is differential execution. There are bound to be more, waiting to be discovered. Of course, everything depends on the training and experience of the "marksman" (the coder).
I think there are lots of new ideas to be discovered. The trick is to tell the difference between the ones that move us forward, versus the ones that hold us back.
I think this is where Doxygen and other documentation systems help. If we can embed small, discrete comments that link to other information such as:
/* help: fooimg.png */
And then have an external documentation system do that, then great.
Even better would be allowing our text-editor to treat those things as hyperlinks to external documentation.
I would reference a drawing as a reference in the code documentation. I see no reason why you can't have footnotes in code.
The ability to make a section of code read-only is something I've wanted
It sounds like you might be interested in Jonathan Edward's research. See, for example:
"The Summer of Code"
"What's next?"
"The future of programming"
Diffing and searching pictures is hard. Diff and search are very important to programmers. Using pictures instead of text is only a marginal improvement in many situations, it has some drawbacks, and it requires general acceptance before it's really worth doing (since you don't make things more understandable if your reader doesn't grok what you've done).
Plus, programmers have a million little tricks that make their lives easier, based on text representations of code, that they'd lose if you gave them code to read that was expressed in anything other than text. Sure, they might replace or re-implement those tricks over time, but in the short term they're gone.
You don't see lawyers switching from English to little back-of-a-napkin diagrams in contracts, either (the Creative Commons licenses try, but cannot make the picture be the formal representation of the contract). Probably for similar reasons.
If someone comes up with a programming language and IDE that, on balance, beats text-based ones; and successfully markets it; then you'll see the start of a revolutionary shift from text to a new format. If nobody comes up with any such thing, then we're not missing out. If someone comes up with something that is more productive but it doesn't gain traction because of independent advantages of other technologies, then that loss is the price we pay for free-market capitalism. Perhaps the ideas will be recycled eventually...
That said, integration between code and documentation could clearly be improved, and there are many efforts underway to do so, using various techniques with varying success. Again, the problem is that any particular cunning plan can in practice only really be implemented in one or a few languages and development environments at a time, and so has difficulty proving that it really is better. Embedding documentation in code is possibly the only universal advance since the invention of the API...
I think there's still a lot that can be done with text, though. For example, debugger technology makes a big difference to programmer productivity in certain common circumstances (namely: when a test fails or something else unexpected happens, but it's not obvious what the faulty assumption is in the code you're looking at). There may be lower-hanging fruit in terms of making programming better, than the actual business of expressing the program.
The last improvement I've seen is
syntax highlighting and context
sensitive help
Then you haven't looked much. Modern IDEs can do far, FAR more than that, namely show you the semantic structure of code (e.g. inheritance hierarchies) and even manipulate it (automatic refactoring) or enrich it with external data (such as who last changed a particular line of code).
I've used emacs, I like text macros. But, what I really want is parse macros. I'd like my editor to expose the machinery behind refactoring in such a way that I can write my transformations on the parse tree of the language itself.
For example, Python added += at one point when my code was littered with x = x + 1 lines. If I could have written a search and replace command that worked on the parse tree, I could have quickly cleaned up large amounts of my source code.
So, I want standard search and replace, but I want it at the level where the structure of my code has meaning, at the abstract syntax tree.
If you've ever used ReSharper, each of its refactorings and recommendations are written in the manner I describe, they find a pattern in the parse tree and suggest a replacement, or for a refactoring, apply a known replacement. I want access to that machinery for my own tasks!
Have you used Doxygen or similar for documenting your code? You can add links to images, and other file types (often stored in same directory as source code) that will get sucked into the generated documentation. I realize that this is one step removed from seeing the detail directly in you favorite editor but it definitely improves how we document our code.
Programming languages are a specialized form of mathematical notation, since you can express a programming language mathematically. Notation changes slowly, and so we don't get fast progress in our languages. Mostly, we advance when we come up with a new thing to fit into the notation, like using i to refer to the square root of negative one.
There are documentation schemes that allow you to embed things other than text. There was at least one programming scheme, Donald Knuth's Web, that allowed you to have a presentation and an execution version of a program (unfortunately, the base source code, the stuff you'd actually hack, was rather messy).
You could easily have a text editor that could treat comments as HTML, provided of course it could recognize comments as it saw them.
I've been thinking a lot about how to make coding faster and more efficient for the past years, always trying to keep it realistic and doing minimalistic implementations. Those are not revolutionary ideas, but since the original poster talked about the punching card to code typing transition, I thought of talking about other ways of communicating to the computer what we want to program.
My ideas are visual or vocal programming. The motivation behind is that there are only a number of ways a loop can be efficiently programmed, and an aware IDE could make some smart code substitution decisions depending on inputs other than typed lines of code.
Visual programming vs Coding: encapsulate (literally) code into "boxes" which have inputs and outputs, and connect them together across a horizontal timeline. This is a high-level concept that would be intrinsically interesting for multithreading development since you can have multiple lines or threads happening at the same time. Every process can be divided into a "box", no matter how you see it. Sending an e-mail in its most basic form is a box which takes an email as input and outputs a success/fail signal. Since the boxes and the lines are distributed across a timeline, the notion of time and event chronology isn't lost and feedback lines are possible.
Vocal programming vs Coding: The effectiveness of this technique would revolve around the effectiveness of the vocal syntax decided to create code and move the cursor. For example, you can say to the microphone "for variable zero to 10" and the system will automatically generate the following code placing the cursor inside:
for (x=0;x<10;x++){
// Cursor would be there after after the call
}
In terms of usability, you would need to be in a relatively silent room to minimize other sounds that might harm the voice recognition so this technology could be used in specialized environments mostly.
This is the result of my extensive programming experience using a wide range of hardware and programming languages. Let me know what you guys think, I'd love having a constructive discussion about that.
A few weeks back the "Intentional Software" created quite a buzz about their new language. I've yet to watch the presentation, but here is a quote from a review by Martin Fowler:
They started worryingly, with the
usual unrevealing Powerpoints, but
then they switched to showing the
workbench and the curtain finally
opened. To gauge the reaction, take a
look at Twitter.
#pandemonial Quite impressed! This is sweet! Multiple domains, multiple
langs, no question is going unanswered
#csells OK, watching a live electrical circuit rendered and
working in a C# file is pretty damn
cool.
#jolson Two words to say about the Electronics demo for Intentional
Software: HOLY CRAPOLA. That's it, my
brain has finally exploded.
#gblock This is not about snazzy demos, this is about completely
changing the world we know it.
#twleung ok, the intellisense for the actuarial formulas is just awesome
#lobrien This is like seeing a 100-mpg carburetor : OMG someone is
going to buy this and put it in a
vault!
Two quotes come instantly to mind:
"If it ain't broke, don't fix it."
"Use the best tool for the job."
Of course, although the core code is still written as text, alll the tools and libraries have changed massively since the days of punched cards.
This has been touched on by others, and it wouldn't revolutionise programming, but anyway...
I think it would be nice if code editors moved slightly beyond plain text editors. Even with syntax highlighting and code completion (which I think are incredibly good things), the editors of today (at least, the ones I use) still display exactly the same ASCII text (or whatever encoding is used) that is in the source files. I would be interested to see how well it would work if editors displayed, for example (some examples are more adventurous than others):
Comments in a text box with a light-blue background and no // or /* ... */ visible
Javadoc comments could have semi-rich text editing support (for those who do HTML Javadoc comments) (seriously, I would appreciate it if code editors rendered Javadoc comments as HTML because they're not the easiest to skim over when their HTML as plain text)
Functions in text boxes that could be collapsed to show only the signature (the collapsing can be done by current editors) and can be dragged around as boxes
Lines between function boxes to indicate how functions are connected
Zooming out so that rather than seeing a single source file (class in many languages) you can see multiple files and the way connect to each other (this would essentially be building UML-like diagramming directly into the code editor)
I think this (in my mind at least) would work without requiring additional markup in the source files so users of plain text editors wouldn't be disadvantaged by having all this extra markup cluttering the files.
Part of the problem might stem from the fact that when you don't code we don't call it programming: Assembling modular components using a GUI for instance.
You might be interested in these alternative programming "languages".
[Ladder][1], designed to mimic the way relay-logic-schemes work. Horrible IMO, but easy to understand for the old guys who did logic with sticks and stones. [http://www.amci.com/tutorials/images/ladder-diagram.gif][2]
[SFC, Sequential function chart][3], designed to simplify parallell programming. Code is written into boxes and these boxes can be placed paralell to each other and will thus execute simultaneously. By connecting the end of several boxes you can syncronize events. Very common for automation applications.
[Mathematica][5]!!!, Might not be the best programming language but the syntax highlighting(if you can call it that) is awesome! For example you can input a matrix by seeing the matrix nicly aligned instead of a huge double[][]. Graphs can be inserted in the code and the formatting of mathematical expressions looks like it does when you write on a paper. No more paranthesis-madness or long Math.PI expressions that really only need one character. And best of all, the files are just plain text even if it is rendered nicely in the editor!
Debuggers is also an area where lots of improvement has been done. Debuggers with replay are starting to come and also visual debuggers where data can be modified in real time. Edit and continiue is also a feature i wouldn't want to live withot.
WTF "new users can only post a maximum of one hyperlink", you will have to google the stuff i originally added to this post >:(
A brain-to-computer translator. Typing is the real
bottleneck. It really just needs to derive the algorithms I
think up and convert that to machine code.
I would say a lot of the newer languages are pretty great at
quickly creating algorithms. The improvements aren't so much
revolutionary now as they are evolutionary.
Dare I say it might actually be a new development language (perhaps even a new paradigm) to take us through such revolution;
I think you might want to take a look at Leo. This is one guys attempt at answering what you're asking about. I still can't wrap my VIM head around it personally, but others take to it quickly. It's not just a programming IDE, but more of an information organizer. It's written in Python, but I don't see why you can't code in other languages with it. The power of Leo is not so much the language, but the ability to express your thoughts and organize them whether it be in code, diagrams, images, or diagrams. Look over the tutorial and examples to get a feel for it. You might like it.
Automated semantic source code transformations, where a program can be reliably examined and manipulated by using an abstract interface/frontend to it that is aware of the underlying semantics.
So that source code can be queried and dealt with pretty much like a SQL database.
Allowing you to do static analysis of source code and refactor even complex source code by doing something along the lines of:
FIND CALLERS OF FUNCTION "foo" WHERE SIGNATURE("int","int","char*") AND RETURN_TYPE("bool");
...
RENAME MACRO "max" TO "maximum" IN FILE "macros.hxx";
RENAME NAMESPACE "prj" TO "project";
RENAME SYMBOL "OLDFOO" IN NAMESPACE "project";
RENAME FUNCTION "log" TO "show_log";
RENAME CLASS "FOO" TO "OLDFOO";
RENAME METHOD "FOO::inc" TO "FOO::increment";
...
CHANGE SIGNATURE IN FUNCTION "foo" WHERE SIGNATURE("int","int") TO SIGNATURE("double","double");
CHANGE SIGNATURE IN METHOD "myClass::handle" WHERE SIGNATURE("char") TO SIGNATURE("unsigned char")
MOVE FUNCTION "foo" in FILE "stuff.cc" TO "foo_funcs.cc";
What tools are available to aid in decoding unknown binary data formats?
I know Hex Workshop and 010 Editor both support structures. These are okay to a limited extent for a known fixed format but get difficult to use with anything more complicated, especially for unknown formats. I guess I'm looking at a module for a scripting language or a scriptable GUI tool.
For example, I'd like to be able to find a structure within a block of data from limited known information, perhaps a magic number. Once I've found a structure, then follow known length and offset words to find other structures. Then repeat this recursively and iteratively where it makes sense.
In my dreams, perhaps even automatically identify possible offsets and lengths based on what I've already told the system!
Here are some tips that come to mind:
From my experience, interactive scripting languages (I use Python) can be a great help. You can write a simple framework to deal with binary streams and some simple algorithms. Then you can write scripts that will take your binary and check various things. For example:
Do some statistical analysis on various parts. Random data, for example, will tell you that this part is probably compressed/encrypted. Zeros may mean padding between parts. Scattered zeros may mean integer values or Unicode strings and so on. Try to spot various offsets. Try to convert parts of the binary into 2 or 4 byte integers or into floats, print them and see if they make sence. Write some functions that will search for repeating or very similar parts in the data, this way you can easily spot headers.
Try to find as many strings as possible, try different encodings (c strings, pascal strings, utf8/16, etc.). There are some good tools for that (I think that Hex Workshop has such a tool). Strings can tell you a lot.
Good luck!
For Mac OS X, there's a great tool that's even better than my iBored: Synalyze It!
(http://www.synalysis.net/)
Compared to iBored, it is better suited for non-blocked files, while also giving full control over structures, including scriptability (with Lua). And it visualizes structures better, too.
Tupni; to my knowledge not directly available out of Microsoft Research, but there is a paper about this tool which can be of interest to someone wanting to write a similar program (perhaps open source):
Tupni: Automatic Reverse Engineering of Input Formats (# ACM digital library)
Abstract
Recent work has established the importance of automatic reverse
engineering of protocol or file format specifications. However, the
formats reverse engineered by previous tools have missed important
information that is critical for security applications. In this
paper, we present Tupni, a tool that can reverse engineer an input
format with a rich set of information, including record sequences,
record types, and input constraints. Tupni can generalize the format
specification over multiple inputs. We have implemented a
prototype of Tupni and evaluated it on 10 different formats: five
file formats (WMF, BMP, JPG, PNG and TIF) and five network
protocols (DNS, RPC, TFTP, HTTP and FTP). Tupni identified all
record sequences in the test inputs. We also show that, by aggregating
over multiple WMF files, Tupni can derive a more complete
format specification for WMF. Furthermore, we demonstrate the
utility of Tupni by using the rich information it provides for zeroday
vulnerability signature generation, which was not possible with
previous reverse engineering tools.
My own tool "iBored", which I released just recently, can do parts of this. I wrote the tool to visualize and debug file system formats (UDF, HFS, ISO9660, FAT etc.), and implemented search, copy and later even structure and templates support. The structure support is pretty straight-forward, and the templates are a way to identify structures dynamically.
The entire thing is programmable in a Visual BASIC dialect, allowing you to test values, read specific blocks, and all.
The tool is free, works on all platforms (Win, Mac, Linux), but as it's personal tool which I just released to the public to share it, it's not much documented.
However, if you want to give it a try, and like to give feedback, I might add more useful features.
I'd even open source it, but as it's written in REALbasic, I doubt many people will join such a project.
Link: iBored home page
I still occasionally use an old hex editor called A.X.E., Advanced Hex Editor. It seems to have largely disappeared from the Internet now, though Google should still be able to find it for you. The last version I know of was version 3.4, but I've really only used the free-for-personal-use version 2.1.
Its most interesting feature, and the one I've had the most use for deciphering various game and graphics formats, is its graphical view mode. That basically just shows you the file with each byte turned into a color-coded pixel. And as simple as that sounds, it has made my reverse-engineering attempts a lot easier at times.
I suppose doing it by eye is quite the opposite of doing automatic analysis, though, and the graphical mode won't be much use for finding and following offsets...
The later version has some features that sound like they could fit your needs (scripts, regularity finder, grammar generator), but I have no idea how good they are.
There is Hachoir which is a Python library for parsing any binary format into fields, and then browse the fields. It has lots of parsers for common formats, but you can also write own parsers for your files (eg. when working with code that reads or writes binary files, I usually write a Hachoir parser first to have a debugging aid). Looks like the project is pretty much inactive by now, though.
Kaitai is an open-source language for describing binary structures in data streams. It comes with a translator that can output parsing code for many programming languages, for inclusion in your own program code.
My project icebuddha.com supports this using python to describe the format in the browser.
A cut'n'paste of my answer to a similar question:
One tool is WinOLS, which is designed for interpreting and editing vehicle engine managment computer binary images (mostly the numeric data in their lookup tables). It has support for various endian formats (though not PDP, I think) and viewing data at various widths and offsets, defining array areas (maps) and visualising them in 2D or 3D with all kinds of scaling and offset options. It also has a heuristic/statistical automatic map finder, which might work for you.
It's a commercial tool, but the free demo will let you do everything but save changes to the binary and use engine management features you don't need.