Artificial Intelligence 2.0

Karl Popper and machine learning

goschinsergiu@yahoo.com (giures) — Fri, 14 Nov 2008 05:02:00 +0000

For some time I was fascinated by the correlation between Karl Popper writings in philosophy of science and some fundamental ideas in machine learning.

I first 'met' Popper when I was 20 years old and Daniel (a great mentor and friend that happens to be my uncle :) advised me to read "The Open Society and its Enemies". The Romanian society at that moment was still strongly marked by the communism years. The stories, the sufferings, the experience of totalitarianism were still alive around me and I could feel and understand them but in a confuse and sentimental way. Reading this book was a great experience because it opened my eyes to a rational account of what happened. It was like a scientific theory of tyranny, its causes and effects. It made me think and rationalize what I was feeling before.

Later on, skimming through "The Logic of Scientific Discovery" I realized that "The Open Society and its Enemies" was only an instance of a more generic view towards rationality in any scientific discipline. Popper introduced in 1934 (I marked the year to emphasize an idea later on) the theory of falsifiability, which shortly says that for a theory to be named scientific, it should offer the possibility to be negated in some clear conditions.

In a more detailed manner, Popper claims that for a discipline to be scientific (~ predictive of real world phenomenons, valuable to human knowledge) it needs to offer predictions which are potentially verifiable (and falsifiable) in an experimental manner. A good scientific theory will offer good predictions (like Einstein's general relativity - Popper's favorite example), a bad one, bad predictions. A non scientific theory will not offer a generic frame that will allow others to negate it (see astrology). If you want to know more here is an excerpt written by the author: Science as falsification.

Even later on, when first reading about machine learning I was struck by the similarity of Popper's ideas and the necessity in ML to have a hypothesis (that is obtained from after training on a training set) that is tried on a different set to test its generalization accuracy. Unless you have such a test set, you can't say that you learned a good hypothesis. It's the same thing as with scientific and non-scientific theories.

And my improbable personal knowledge circle was complete when skimming through Vladimir Vapnik's "The nature of statistical learning theory" I read: "Before continuing with the description of statistical learning theory, let me remark how amazing Popper's idea was. In the 1930s Popper suggested a general concept determining the generalization ability (in a very wide philosophical sense) that in the 1990's turned out to be one of the most crucial concepts for the analysis of consistency of the ERM inductive principle" (page 55). It seems remarkable (at least to me) how someone with no mathematical tools was able to come up with such an idea (Popper formalized it really well as Vapnik's acknowledges ; the two actually met at a certain point towards the end of Popper's life). And it is also an indication of how powerful philosophy can be.

Update: Thanks to David Corfield for pointing me to a much more informed opinion regarding the connection between statistical learning and Popper's ideas: "Popper, Falsification and the VC-dimension" by David Corfield, Bernhard Scholkopf, Vladimir Vapnik.

Random Machine Learning thoughts

goschinsergiu@yahoo.com (giures) — Thu, 13 Nov 2008 06:36:00 +0000

A part of my brain is still fighting the idea that if you don't make any kind of presupposition (if you don't have a bias whatsoever) then you won't be able to learn anything.

It's still strange that when you perfectly learn something then you are overfit and you won't be able to generalize in a good manner.

It's amazing for me that a 2 layer neural network can act as a universal continuous function approximator.

I feel there is a gap between the math in ML and the philosophy that ML should generate and that's kind of sad.

I keep wondering how come not everybody wants to study AI, machine learning and the like - I'm kidding, but deep inside I'm really stupefied :). A good friend asked me what would I do (professionally) if I would win a huge sum on the lottery (no danger here, I never buy tickets). And I felt really well thinking and saying that I would continue doing exactly what I am doing now.

I use the opportunity of an interesting post about the (lack of?) power of randomness to refer you to one of my favorite blogs: Overcoming Bias. (a blog at the intersection between philosophy, cognitive science, and a bit of AI).

NY Machine Learning Symposium

goschinsergiu@yahoo.com (giures) — Sat, 11 Oct 2008 04:23:00 +0000

I was at the 3rd Annual Machine Learning Symposium today. This a (rather new) workshop that aims to bring together researchers from NYC area that work in machine learning.

It was a very nice experience. Besides the cool ideas and the presentations, I really appreciated the atmosphere, the openness of the people and the great view (see below).

From the about 30 student posters they selected 10 finalists (that had short but informative presentations) and the winner was a NYU team with the paper "Sample Selection Bias Correction Theory" , which was really interesting. If I understood correctly, they said that usually when a test set is generated for a learning problem some samples are not presented to the learner (because of a higher cost of the labeling for instance); thus a bias is introduced in the test sample. This bias is of course unwanted and should be corrected. They presented a theoretical analysis of sample selection bias correction and presented some techniques of how to do it.

Regarding the presentations, the most interesting (and sometimes a little scary) for me was that of Tony Jebara from Columbia. The idea was that

"With machine learning algorithms applied to these human activity graphs, it becomes possible to make predictions for advertising, marketing and collaborative recommendation."

(citation from speaker abstract). They used advanced algorithms to: 1. form a graph from sparse (GPS in this case) data (based on an approximation of k-matching algorithms - of which I just found out today and seems really cool), 2. partition the graph in a 'meaningful' way, 3. interpret the information in terms of what behaviors groups of people have so as to better match (for instance) them with advertisers. This was the scary part in the sense that they tracked people movements in a city what restaurants they went, when, etc (of course the information was made anonymous but still it felt strange to see an animation with clusters of people doing different stuff) - I wasn't familiar with most of the theoretical part but was interesting to learn more.

And the picture from outside the conference room (naturally the chairs were facing the wall, otherwise I surely wouldn't have been able to concentrate on the presentations):

A new beginning

goschinsergiu@yahoo.com (giures) — Fri, 19 Sep 2008 23:03:00 +0000

It's been a long time since I gave up at writing something here. It was for good reasons.
Even though I didn't write a lot (about 10 articles, and all of them very far from the quality or the content I wished when starting this), this blog was one of the factors that made me realize an important thing: that I wasn't following my dreams. Trying to write interesting AI related things here, I discovered that I didn't have the time to really study, understand and communicate them, except on a very superficial level. And it was frustrating. So this, among other things, led me to decide to want to return to the academic realm. And after GREs, applications and the usual track of a foreign applicant, this autumn I started a PhD in Computer Science at Rutgers University with the hope to study in depth machine learning and reinforcement learning. Hopefully, I'll have more interesting things to write from now on.

Most spectacular AI applications

goschinsergiu@yahoo.com (giures) — Tue, 19 Jun 2007 20:35:00 +0000

Below I tried to compile a short list with the most impressive AI applications (any other visions regarding this are welcomed). I find it fascinating that AI is far from being dead or dead-locked as some say and instead has a direct impact on our lives. It is most of the time hidden behind things that become normal with usage, but that does not mean that it's not there.

Mars Spirit and Opportunity Rovers.
Why are they impressive? Because they wonder semi-autonomously on a different planet for 3.5 years.

What is interesting from an AI point of view is the control system - which is much better than the one for the Sojourner's mission. The rovers have stereo cameras and generate 3D maps that allow them to avoid any obstacles by choosing the "path of least resistance". The human drivers usually give them a target point situated at "big" distances (hundreds of meters) and the rovers decide how they should get to that point.

Google
It is impressive because it is the first large scale successful implementation of complex machine learning techniques. I've always considered Google not so much the success of a Web 2.0 approach but more of an AI 2.0 approach.
AI is the "hidden" secret of Google search. They attract the best machine learning scientists (the best example is that Peter Norvig is Google's Director of research).

DeepBlue
It is impressive because it was the first computer to beat to world chess champion.
Although its main characteristic was brute force (capable of analyzing 200.000.000 moves a second) he also incorporated heuristics that allowed him to beat Kasparov.
It marked the beginning of an era - the moment when people realized that computers could outsmart them at things that were considered linked to highest levels of intelligence.

Resources:

CS on the map of science

goschinsergiu@yahoo.com (giures) — Tue, 27 Mar 2007 07:47:00 +0000

Through Slashdot I got to a Map Of Science - an interesting view of the connections between 800k published papers in 776 scientific paradigms (see the complete image here).
What seemed really interesting to me were the closest domains to Computer Science: Math, BrainResearch, Astrophysics and Social Sciences. Although the first three don't come as a surprise, the last one is a bit unexpected.

AI academic research

goschinsergiu@yahoo.com (giures) — Sun, 04 Feb 2007 21:58:00 +0000

The idea for this post came to me after reading a great series of 3 articles on Machine Learning Thoughts about how it is to be a scientist (read here, here and here).
Do you want to do academic research? In AI maybe? At what level? (i asked myself these questions at a certain point and all the ideas the follow are only my personal opinions).
Q1: What country?
A1: USA seems the good place to be at the moment for AI research. Top level universities, top level conferences, top level firms where you can project ideas into reality and Silicon Valley where you can make your startup and change the world.
Q2: What universities / research groups?
A2: You can find a lot of rankings on the internet - general, engineering, computer science and Artificial Intelligence. For AI the ranks are old, but they give a good idea of places where interesting things happen. The only place that is missing from there is Georgia Tech. Of course in the end it matters if you find a place with a great teacher (they are not all at MIT or Carnegie) that does what you like, but your chance of finding this are much higher at the first 10 schools. Be it robotics, machine learning, text mining you can find in those universities the majority of people that make things moving in AI research.
I was impressed since i was in high school of the incredible number of robots from MIT and then in university when i heard of seemingly science-fiction projects from the same university.
Q3: How do you get in such great places?
A3: I am not an American so i only know about experiences of foreigners. You have to a good GPA (unless you are really great in other aspects), good GRE / Toefl scores, nice recommendations and a good intuition of what those that make the selection want to hear in a statement of purpose letter. Oh yes, and if possible research activity (publications at conferences, in journals).
Q4: How do you publish?
A4: Well, that's tough. You have to have an idea of what you like and try to touch the borders in the specific domain. When you do try to push it a little further and try to project that into a paper (just my opinion remember :). Then (or before) find a professor with similar interests and try to get some advices and maybe a collaboration.
Q5: Where do you publish?
A5: It depends a lot on how good your idea / paper is. You can find here the ranks for AI conferences (remember, publishing at AAAI, ICML or NIPS might be as hard as being accepted at MIT, Carnegie Mellon or Berkley).
Your chance of being accepted? See a statistic of the last years acceptance rate. It's also a good idea to look at least at the titles of the papers accepted at the conference / journal you want to apply to in the previous years to get an idea of the "hot" and preferred topics.
Q6: Is it possible to do research directly in a company as to not "lose" so much time with a PhD?
A6: I don't know since i don't have any experience with that (if you know, tell me). Logically, it would seem improbable that a top notch industrial research department would choose a John Doe instead of a bright PhD from one of the first 10 universities, so i would be inclined to say no, it's not very probable for this to happen (but again, i might be very wrong).
I'm sure there are a lot of other questions, but in the end the most important ones are: do you want to do research? is this something that suites you? are you ok with spending some years without earning so much money as in the industry and to take the risk of not making anything significant instead? are you passioned of the ideas you have? If you answer yes to all of these then it might be a good idea to start looking for an university :).

Microsoft Robotics Studio technical evaluation

goschinsergiu@yahoo.com (giures) — Mon, 18 Dec 2006 19:07:00 +0000

I've been playing in the last few days with the release from Microsoft of Microsoft Robotics Studio kit.
It's supposed to offer a common and standard platform for the development of robotic applications for hobbyists, students, researchers (for free) and enterprises (for 399$).
It is based on a collaboration with a lot of partners (among which Lego, Braintech - who announced support last week for the new platform, Kuka and others). It's interesting that the partners come from the industrial robotic business too (and not only IT oriented companies).
See this interview with Bill Gates if you want to see the supposed business perspective and the target of this release from Microsoft point of view (it is interesting but not the main point of this post).
I read the documentation and the tutorials. This is a first bad point - the documentation is poor, the concepts are described too briefly (especially since the kit is supposed to attract people that had nothing to do with computer science, to robotics ). It seems like a job done in a hurry.
The services runtime is similar to a component based approach (even if they don't say that explicitly). This is helpful because you don't have very tight links between modules and this increases modularity and re-use of software modules (at OpenPlug, we are doing a similar thing for the mobile industry based on a proprietary technology - ELIPS).
Another concept introduced is Microsoft Visual Programming Language - that should help non-programmers in doing robotic programming (even if they say this is not limited to robotics). See the picture below for a good non-verbal explanation of what this is.

It looks very much like the Lego Mindstorms visual programming language (so much that i would say that's where they took the idea from). I hated that language back when i played for the first time with a RCX brick (the controller of the Lego Mindstorms robots) because it wasn't flexible enough for what i wanted to do (i looked and found on the Internet C libraries that gave much more flexibility). Only time will say if this is really easier to use than just do programming in an old fashion way.
And since we got here, you can program the robots using C# and VB.net (using .net framework 3.0). Somehow it doesn't feel good to program robots in C# ...
I will talk only about the simulation part (since i don't have any robots available at the moment).
It looks and moves bad. It consumes a lot of resources. Even when creating a simple world with a box and a sphere, my system barely moves (512M RAM, 64M dedicated video card, 1.8GHZ Centrino - not so bad...). This is not normal. I tried three times to actually do a simulation tutorial from start to end, and my system completely blocked when i tried to open a Firefox window (hey maybe that's the problem :) ) to change some simulation parameters for a simulated robot in the environment. The user interface is far from being nice and easy to use (if you want to load a scene, an object or something else you have the option of loading only xml anonymous files - good luck in finding what you need).
The simulation as a whole looks like a college semester project (i've seen open source 3D engines that moved a lot better with fewer resources - and that also contained physics simulators). I only hope they concentrated on the hardware part and that's why the simulation is so drafty.
From a high level view, i also think there is something very important missing - robot controllers samples for autonomous robots and a library to facilitate their development. This is from where a lot of excitement in robotics appears and there is nothing at the moment in this kit (again, maybe a partner in the future).
In conclusion, i couldn't agree more with the business perspective (of having a standard platform to unify robotics development and speed its evolution), but this release is far from being what the industry needs. I only hope to see a much better future version and maybe partners that will improve the experience.
If you had any personal evaluations of the simulation and hardware part please tell me, i am very curious to hear about your experiments.

AI News week 50

goschinsergiu@yahoo.com (giures) — Sat, 16 Dec 2006 14:43:00 +0000

Here are the most interesting AI news this week:

Microsoft launched this week a new platform for robotics development. It's aim is to standardize robotics software development (which is considered one of the main reasons why robotics isn't developing at a faster rate at the moment). This initiative is supported by more than 30 partners from the robotics industry (among which Lego, Roomba) and two top AI universities from USA (Georgia Tech and Carnegie Mellon). I'll be back with a quick evaluation of the package (it's downloadble free of charge for non-commercial use here).
A pretty interesting Final Cut movie type of idea - computers are said to be able to record the entire life of an individual in 20 years time due to the development of storage technologies to the point where each individual minute of life can be stored for later "usage". Combine this with the current work in AI to automate recognition of certain features in pictures and video footage and you get an automatic "judgment" of a person's life. Scary...
Linked to losing private life ownership (there were some articles about UK super video surveillance system in the news last week), via Interspies, an interesting article about automating video surveillance. The reason is the high cost of having humans supervise hundreds of video cameras. The system is trained to detect features in a video (like human postures, gestures, face expressions) in order to automatically signal the potential threats. It's the kind of system that it's great to work on, but has bad social effects in my opinion. A supervised society with a nanny state is not the kind of world i'd like to live in. It's surprising how people don't react more to the implementation of such systems.
Via AAAI News, an analysis on the business perspectives of replacing some of the call center humans with virtual agents to answer questions from users. From a technological point of view this is not entirely possible at the moment, but it's surely a reason to increase the money invested in AI research due to lower costs of having such systems (bad news for Indian call centers on the long term...).
Via AAAI News, a system that will do automatic trading based on news events will be offered as a solution by Reuters. The advantage is clear - the possibility to have a rule-based system capable of reacting in milliseconds to events in the market (higher than the speed of any human). This is not the first system build with this in mind. It's interesting to see how the human involvement is such decisions will evolve in the future.

AI News week 49

goschinsergiu@yahoo.com (giures) — Sat, 09 Dec 2006 14:18:00 +0000

I'll start from now a periodic weekly posting of the best news (in my view) linked to AI that i read on the web. I'll try to be very selective and present only things that really deserve passing some minutes reading and thinking about them (and let them cool down for a few days to see if they are really interesting). Doing that will also stop me from posting a new article each time something nice appears. Enough talk, let's start for this week:

"The Future Evolution of Consciousness" - a great article published on robots.net about how to improve ourselves as individuals, augmenting our consciousness. The article gives a very interesting view on our evolution. From the abstract: "the prime function of consciousness is to develop novel adaptive responses. Consciousness does this by putting together new combinations of knowledge, skills and other disparate resources that are recruited from throughout the brain. [...] A number of sources (including the practices of religious and contemplative traditions) are drawn on to investigate how [...] the potential for adaptability [can] be improved by the conscious adaptation of key processes that constitute consciousness.".
Togelius: Physical car control - very interesting experiment of a physical car controlled by a hard-coded 15 lines controller (the tools that appear in the video look pretty cool). Togelius research is linked to Evolutionary Robotics and shows that this approach works in reality and not only in simulation (i might be a little subjective here since that is a topic i like a lot).
Raja Chatila - Robot navigation - as usual a very interesting podcast on Talking Robots.
Man loses to machine again - just one of the articles describing this week Kramnik's loss to Deep Fritz. It's becoming a rule to see the world chess champion beaten by a machine. I am wondering when will this happen in Go or Poker (as you probably know, the best programs / machines at the moment have no chance in front of the Go or Poker world champions). I know there are some good research teams out there that are struggling with this.

AI methods: good and bad

goschinsergiu@yahoo.com (giures) — Mon, 04 Dec 2006 16:55:00 +0000

Via Intelligent Machines, i found an interesting article regarding miss usage of neural networks as methods to solve a problem.
I don't know how accurate the content of the article is and i usually don't like the news that encourage a pessimistic view on AI (i am always thinking how research in Neural Networks was discouraged for 2 decades because someone discovered that a simple neuron can't learn XOR - which was true and useful, but not as important as to drop research on the whole idea). Nevertheless the article is a very good example of how fashion and lack of competence can influence technology in a bad way.
As in a lot of areas in Computer Science, AI is filled with buzz words (be it Neural Networks, Machine Learning or others) that a lot of times promises much more than they can offer at the moment (in the end a Neural Network , even written with caps, is just a very abstract model that is very far - very = orders of magnitude - from the real thing). And for someone who studied AI only superficially (or not at all) these words probably seem to represent the solution to a lot of hard problems (well i hope they will in the future, but they don't at the moment).
There is a lot of talk in the preparatory courses and on the Internet about the methods (genetic algorithms, neural networks, etc) and much less on the applications (besides the very well known). And even if i agree that new ideas for applications appear when you get to know and understand the basis of some domain, it's also very true that you need to know what to do with something in a pragmatic way in order to generate useful applications.

AI resources on the web

goschinsergiu@yahoo.com (giures) — Tue, 28 Nov 2006 22:02:00 +0000

I read today a post on Machine Learning Thoughts about a nice initiative to create a page with videos linked to Machine learning. And i remembered of one of the first problems i encountered when trying to find out useful (=original and well presented information, if possible audio and video) links to AI resources on the web (google is not enough for now - maybe when they will have a real AI that will understand human language :) )
AAAI AI news seems like a good place to start, but there are some things i don't like about it:

they've only recently introduced RSS feeds and not all the news are loaded there (and of course not with the entire content) - i hate that since i'm an RSS client addicted.
the site as a whole is pretty hard to use and find useful information.
a lot of the links are broken (try to look at Reinforcement page for instance).
a lot of the information links to not really useful pages (nothing more than introduction).
it seems that only some hot topics are covered really good and this is not really nice for a site that says it should cover AI.

With all these problems, AAAI AI News somehow loses the chance to be a good portal for AI.

Another good try was Generation5 , but at a certain point this year they gave up creating new content (lack of time probably) - sad, since they had a big community behind and a lot of interesting initiatives.

There are good sites, nevertheless, but more specialized. The first example that comes to mind is Pascal Virtual Playground, a rich resource of videos with Machine Learning courses from conferences, workshops etc.

Bottom line it would be great to have a real AI portal with easy access to rich, meaningfull and maintained information like:

AI research groups.
AI publications (links to citeseer categorized in some way).
AI news (with RSS with full content).
AI interviews.
AI tutorials and applications (text, video, audio).
and many more.

If no one will do it, maybe i'll try some day ;).
If anyone knows of other good portals please let me know.

A new post on Talking Robots

goschinsergiu@yahoo.com (giures) — Fri, 24 Nov 2006 10:45:00 +0000

A new really interesting post on Talking Robots about the evolution of language - Luc Steels - Evolution of Communication and Language.
I like a lot the initiative of Dario Floreano - which papers influenced me a lot in adopting Evolutionary Robotics as my theme for the bachelor degree final project - but about this in a later post.

Self replicating machines

goschinsergiu@yahoo.com (giures) — Wed, 22 Nov 2006 18:21:00 +0000

Today i'd like to describe and make a summary about something which obsesses me for some time: self-replicating systems (SRS).

I found out about them when reading stuff linked to Cellular Automaton (no, i won't speak about the Game of Life - not yet anyway).
As the name says, they are machines capable of producing, besides usefull outputs, identical copies of themselves. The advantage is the generation, in an exponential manner, of products with diverse utilities.
The idea came to John von Neumann in the 50's.
NASA conducted a study in the 80's to evaluate the possibility of constructing such machines to explore the universe. Their conclusions were "The basic concept of physical machine systems capable of self-replication appears credible both from a theoretical and a practical engineering standpoint." and "It is feasible to begin immediate work on the development of a simple demonstration SRS on a laboratory scale, with phased steps to more sophisticated levels as the technology is proven and matures.". These things were more or less known before (cellular automaton specialists showed simulated systems that could successfully reproduce themselves) but, as far as i know, NASA didn't pursue this research. Sad, since this is really, in my opinion, the only way to start conquering the solar system at affordable economic costs and a great way to increase the level of life for everybody.
How? The generic mechanism for an SRS is:

place an auto-reproducible kernel in an area with some specific resources
abandon it for a period for self-replication
the kernel produces useful artifacts (NASA envisioned a factory on the Moon that could produce a things like living modules, diverse components for space ships, and then replicate itself to increase the speed)
gather the results.

The initial investment would be very small compared to the huge benefits.
I know this sounds pretty science-fiction, and of course we don't have the technology to build "perfectly" SRS at the moment (capable of self replicating in chaotic environments = that don't contain finished modules), but on one hand i believe it is possible to build "partial" SRS and on the other hand it really seems feasible to do it in the future (a big part of the bricks needed already exist).
Do you know of any other initiatives of bringing all this to reality (i know there are a lot of computer simulations, but is there a roadmap for really doing it)?

Resources:
http://www.islandone.org/MMSG/aasm/AASMIndex.html - NASA study
http://www.zyvex.com/nanotech/selfRepJBIS.html - a good presentation on SRS
http://www.foresight.org/Conferences/MNT6/Papers/Hall/index.html - architecture of SRS
http://www.cs.bgu.ac.il/~sipper/selfrep/ - the artificial self replication page

Update: Alex sent me some interesting links about recent work done in this domain linked to nano-technology. The idea presented in the post could be applied to fabricate nanorobots that can self-replicate in an adaptive way (generating copies and other usefull objects according to the needs in the environment) and self-repair. You can read about this here and see an interesting video here.

Intention

goschinsergiu@yahoo.com (giures) — Thu, 16 Nov 2006 20:08:00 +0000

This blog is about any breakthrough that will enable us to have truly intelligent systems that can autonomously exist in current human environments (artificial or real). There are more than 50 years since researchers started working on AI and we are still very far from having such systems. Nevertheless, there are a number of breakthroughs that will potentially make them possible. I would like to talk about these bricks that (hopefully) will lead to a new era in AI - let's call it AI 2.0.
I will also post about my personal research interests and projects (linked to AI - just a hobby right now). I am very passionate about Neural Networks, Evolutionary Robotics, Genetic Algorithms as methods and about Autonomous Agents / Robots as applications of these methods.