Machine Learning (Theory)

5/8/2009

Computability in Artificial Intelligence

Normally I do not blog, but John kindly invited me to do so. Since computability issues play a major role in Artificial Intelligence and Machine Learning, I would like to take the opportunity to comment on that and raise some questions.

The general attitude is that AI is about finding efficient smart algorithms. For large parts of machine learning, the same attitude is not too dangerous. If you want to concentrate on conceptual problems, simply become a statistician. There is no analogous escape for modern research on AI (as opposed to GOFAI rooted in logic).

Let me show by analogy why limiting research to computational questions is bad for any field.

Except in computer science, computational aspects play little role in the development of fundamental theories: Consider e.g. set theory with axiom of choice, foundations of logic, exact/full minimax for zero-sum games, quantum (field) theory, string theory, … Indeed, at least in physics, every new fundamental theory seems to be less computable than previous ones. Of course, once a subject has been formalized, further research (a) analyzes the structure of the theory and (b) tries to compute efficient approximations. Only in (b) do computational aspects play a role.

So my question is: Why are computational questions so prevalent in AI research? Here are some unconvincing arguments I’ve heard:

A) Because AI is a subfield of computer science, and the task of computer scientists is to find (efficient) algorithms for well-defined problems?

I think it does not do any (real-world) problem any good to confine it to computer science. Of course, philosophers and cognitive scientists also care about AI, but where are the mathematicians?

B) Because formalizing AI and finding efficient smart programs goes hand-in-hand? Separating these two issues would lead to no, or at best to results which are misleading or useless in the construction of intelligent machines?

I am not aware of any convincing argument that separating the issues of “axiomatizing a field” and “finding efficient solutions” will (likely) fail for AI. The examples above of other fields actually indicate the opposite. Of course, interaction is important to avoid both sides running wild. For instance, von Neumann’s minimax solution for games, albeit infeasible for most games, is the cornerstone of most practical approximations.

C) Because there is some deep connection between intelligence and computation which can not be disentangled?

Sure, you could say that intelligence is by definition about computationally efficient decision making. This is as unconvincing as argument (A). Pointing out that the human brain is a computational device is quite useful in many ways, but doesn’t proves (C) either. Of course, ultimately we want a “fast” smart algorithm. How is AI different from wanting a fast algorithm computing primes, which you derive from a non-algorithmic definition of primes; or drawing fractals?

D) Because AI is trivial if computational issues are ignored? All conceptual problems have already been solved?

Many have expressed ideas that some form of exhaustive search over all possible solutions and picking the “best” one does the job. This works essentially for exactly those problems that are well-defined. For instance, optimal minimax play of a zero-sum game or solving NP complete problems are conceptually trivial, i.e. if computation time is ignored. But in general AI and machine learning, there is not a universally agreed-upon objective function. The Turing test is informal (involves a human judge in the loop), maximizing expected reward (the true distribution is not known, so expectation w.r.t. to what?), etc. The AIXI model, briefly discussed at this blog, is the first complete and formal such criterion, for which, let me phrase it that way, no flaw has yet been identified. Shane Legg’s award-winning thesis gives an informal introduction and contains lots of discussion.

Conceptual and computational problems in AI should be studied jointly as well as separately, but the latter is not (yet) fashionable. When AI was more logic oriented, some good logicians helped develop the foundations of “deductive” AI. Where are the researchers giving modern “inductive” AI its foundation? I am talking about generic learning agents, not classifying i.i.d. data. Reinforcement learners? Well, most of the hard results are from adaptive control theorists, but it’s reassuring to see parts of these communities merging. It’s a pity that so few mathematicians are interested in AI. A field “mathematical AI” with the prestige of “mathematical physics” would be exciting. As a start: 40% of the COLT & ALT papers on generic learning agents, 30% induction, 20% time-series forecasting, 10% i.i.d. Currently it’s reversed.

5/6/2009

Machine Learning to AI

I recently had fun discussions with both Vikash Mansinghka and Thomas Breuel about approaching AI with machine learning. The general interest in taking a crack at AI with machine learning seems to be rising on many fronts including DARPA.

As a matter of history, there was a great deal of interest in AI which died down before I began research. There remain many projects and conferences spawned in this earlier AI wave, as well as a good bit of experience about what did not work, or at least did not work yet. Here are a few examples of failure modes that people seem to run into:

  1. Supply/Product confusion. Sometimes we think “Intelligences use X, so I’ll create X and have an Intelligence.” An example of this is the Cyc Project which inspires some people as “intelligences use ontologies, so I’ll create an ontology and a system using it to have an Intelligence.” The flaw here is that Intelligences create ontologies, which they use, and without the ability to create ontologies you don’t have an Intelligence. If we are lucky, the substantial effort invested in Cyc won’t be wasted, as it has a large quantity of information stored in a plausibly useful format. If we are unlucky, it fails to even be partially useful, because the format is unnatural for the internal representations of an Intelligence.
  2. Uncertainty second. Many of the older AI programs had no role for uncertainty. If you asked the people working on them, they might agree that uncertainty was an important but secondary concern to be solved after the main problem. Unfortunately, it seems that uncertainty is a primary concern in practice. One example of this is blocks world where a system for planning how to rearrange blocks on a table might easily fail in practice because the robot fails to grab a block properly. Many people think of uncertainty as a second order concern, because they don’t experience uncertainty in their daily lives. I believe this is incorrect—a mental illusion due to the effect that focusing attention on a specific subject implies reducing uncertainty on that subject. More generally, because any Intelligence is a small part of the world, the ability of any intelligence to perceive, understand, and manipulate the world is inherently limited, requiring the ability to deal with uncertainty. For statistics & ML people, it’s important to not breath a sigh of relief too easily, as the problem is pernicious. For example many ML techniques based around conditional independence routinely suffer from excess certainty.
  3. Computation second. Some people try to create an intelligence without reference to efficient computation. AIXI is an extreme example of this sort. The algorithm is very difficult to deploy in practice because there were no computational constraints other than computability designed into it’s creation. It’s important to understand that computational constraints and uncertainty go together: because there are computational constraints, an intelligence is forced to deal with uncertainty since not everything which might follow at a mathematical level can be inferred in the available computational budget.
  4. AI-Hard problems. There was a time when some people thought, “If we could just get a program that mastered chess so well it could beat the best humans, we will learn enough about AI to create an AI.” Deep Blue put that theory to rest. Current efforts on Poker and Go seem more promising, but no one believes they are “AI-Hard” for good reason. It’s not even clear that the Turing Test is a reliable indicator, because (for example) we might imagine that there is Intelligence which can not imitate a human, or that there are programs that can imitate humans well enough to fool humans without being able to achieve everything that an Intelligence could. Perhaps the best evidence is something singularity-style: AI exists when it can substantially improve it’s own abilities.
  5. Asymptopia. In machine learning there are many theorems of the form “learning algorithm A can solve any learning problem in the limit of infinite data”. Here A might be nearest neighbors, decision trees, two-layer neural networks, support vector machines, nonparametric statistics, nonparametric Bayes, or something else. These theorem are ok, but insufficient. Often the algorithms are not computationally acceptable, and even if so, they are not sufficiently efficient with respect to the amount of experience required to learn.

Solving AI is undeniably hard, as evidenced by the amount of time spent on it, and the set of approaches which haven’t succeeded. There are a couple reasons for hope this time. The first is that there is, or soon will be sufficient computation available, unlike the last time. The second is that the machine learning approach fails well, because there are industrial uses for machine learning. Consequently, we can expect a lack of success to still see substantial use in practice. This might sound like “a good downside”, but it’s actually an upside, because it implies that incremental progress has the potential for ultimate success.

Restated at an abstract level: a hard problem can generally be decomposed in many ways into subproblems. Amongst all such decompositions, a good decomposition is one with the property that solutions to the subproblems are immediately useful. The machine learning approach to AI has this goodness property, unlike many other approaches, which partially explains why the ML approach is successful despite “failing” so far to achieve AI.

One reason why AI is hard, is that it turns out tackling general problems in the world undeniably requires a substantial number of different strategies, including learning, searching, and chunking (= constructing macros), all while respecting constraints of computation and robustness to uncertainty. Given this, a fair strategy seems to be first mastering one strategy, and then incorporating others, always checking that that incorporation properly addresses real world problems. In doing this, considering the constraint ignoring approaches as limiting cases of the real system may be helpful.

11/26/2008

Efficient Reinforcement Learning in MDPs

Tags: Reinforcement,Theory jl@ 7:29 am

Claude Sammut is attempting to put together an Encyclopedia of Machine Learning. I volunteered to write one article on Efficient RL in MDPs, which I would like to invite comment on. Is something critical missing?

3/23/2008

Interactive Machine Learning

A new direction of research seems to be arising in machine learning: Interactive Machine Learning. This isn’t a familiar term, although it does include some familiar subjects.

What is Interactive Machine Learning? The fundamental requirement is (a) learning algorithms which interact with the world and (b) learn.

For our purposes, let’s define learning as efficiently competing with a large set of possible predictors. Examples include:

  1. Online learning against an adversary (Avrim’s Notes). The interaction is almost trivial: the learning algorithm makes a prediction and then receives feedback. The learning is choosing based upon the advice of many experts.
  2. Active Learning. In active learning, the interaction is choosing which examples to label, and the learning is choosing from amongst a large set of hypotheses.
  3. Contextual Bandits. The interaction is choosing one of several actions and learning only the value of the chosen action (weaker than active learning feedback).

More forms of interaction will doubtless be noted and tackled as time progresses. I created a webpage for my own research on interactive learning which helps define the above subjects a bit more.

What isn’t Interactive Machine Learning?
There are several learning settings which fail either the interaction or the learning test.

  1. Supervised Learning doesn’t fit. The basic paradigm in supervised learning is that you ask experts to label examples, and then you learn a predictor based upon the predictions of these experts. This approach has essentially no interaction.
  2. Semisupervised Learning doesn’t fit. Semisupervised learning is almost the same as supervised learning, except that you also throw in many unlabeled examples.
  3. Bandit algorithms don’t fit. They have the interaction, but not much learning happens because the sample complexity results only allow you to choose from amongst a small set of strategies. (One exception is EXP4 (page 66), which can operate in the contextual bandit setting.)
  4. MDP learning doesn’t fit. The interaction is there, but the set of policies learned over is still too limited—essentially the policies just memorize what to do in each state.
  5. Reinforcement learning may or may not fit, depending on whether you think of it as MDP learning or in a much broader sense.

All of these not-quite-interactive-learning topics are of course very useful background information for interactive machine learning.

Why now? Because it’s time, of course.

  1. We know from other fields and various examples that interaction is very powerful.
    1. From online learning against an adversary, we know that independence of samples is unnecessary in an interactive setting—in fact you can even function against an adversary.
    2. From active learning, we know that interaction sometimes allows us to use exponentially fewer labeled samples than in supervised learning.
    3. From context bandits, we gain the ability to learn in settings where traditional supervised learning just doesn’t apply.
    4. From complexity theory we have “IP=PSPACE” roughly: interactive proofs are as powerful as polynomial space algorithms, which is a strong statement about the power of interaction.
  2. We know that this analysis is often tractable. For example, since Sanjoy‘s post on Active Learning, much progress has been made. Several other variations of interactive settings have been proposed and analyzed. The older online learning against an adversary work is essentially completely worked out for the simpler cases (except for computational issues).
  3. Real world problems are driving it. One of my favorite problems at the moment is the ad display problem—How do you learn which ad is most likely to be of interest? The contextual bandit problem is a big piece of this problem.
  4. It’s more fun. Interactive learning is essentially a wide-open area of research. There are plenty of kinds of natural interaction which haven’t been formalized or analyzed. This is great for beginnners, because it means the problems are simple, and their solution does not require huge prerequisites.
  5. It’s a step closer to AI. Many people doing machine learning want to reach AI, and it seems clear that any AI must engage in interactive learning. Mastering this problem is a next step.

Basic Questions

  1. For natural interaction form [insert yours here], how do you learn? Some of the techniques for other methods of interactive learning may be helpful.
  2. How do we blend interactive and noninteractive learning? In many applications, there is already a pool of supervised examples around.
  3. Are there general methods for reducing interactive learning problems to supervised learning problems (which we know better)?

12/10/2007

Learning Track of International Planning Competition

The International Planning Competition (IPC) is a biennial event organized in the context of the International Conference on Automated Planning and Scheduling (ICAPS). This year, for the first time, there will a learning track of the competition. For more information you can go to the competition web-site.

The competitions are typically organized around a number of planning domains that can vary from year to year, where a planning domain is simply a class of problems that share a common action schema—e.g. Blocksworld is a well-known planning domain that contains a problem instance each possible initial tower configuration and goal configuration. Some other domains have included Logistics, Airport, Freecell, PipesWorld, and many others. For each domain the competition includes a number of problems (say 40-50) and the planners are run on each problem with a time limit for each problem (around 30 minutes). The problems are hard enough that many problems are not solved within the time limit.

Given that the planners are asked to solve many problems from individual domains, and that problems within a domain generally have common solution structures, it makes sense to consider learning from previous problem-solving experience in a domain to better solve future problems in the same domain. Here “better solve” could mean either solve the problems significantly more quickly or finding better quality plans in a similar time frame. However, no planner in any of the competitions has included a learning component. Rather, to quote Foreigner, for these planners each problem “feels like the first time”.

Perhaps one reason that planners have not incorporated learning into the competition setting is that there has not been much overlap between the ICML and ICAPS communities, although that is changing. Another reason is perhaps that the structure of the competition would deduct any “learning time” from a planner’s 30mins per problem, which could reduce the potential benefits.

The learning track for the 2008 competition is being designed so that learning time is not counted against planners. Rather, there will be a distinct learning phase and a distinct evaluation phase. During the learning phase the planners will be provided with the set of domains to be used in the competition and example problems from each. The evaluation phase will be conducted like the current competition, with the exception that the learning-based planners will be allowed to read in a learned domain-specific “knowledge file” when solving the problems. This structure is designed to help answer the following question:

Do we have techniques that can leverage a learning period to outperform state-of-the-art non-learning techniques across a wide range of domains?

My current belief is that the answer is “no”. I certainly have never seen any such demonstration. This is not because of lack of work in the area of “learning to plan” as there has been a long history dating back to some of the early planners (see my horribly outdated resource page for a taste). While many of the learning approaches have shown some degree of success, the evaluations have typically been very narrow, focusing on only 2 to 3 domains and often only a few problems. My intuition, grounded in personal experience, is that most (all) of these systems would be quite brittle when taken to new domains. The hope is that the learning track of the competition will force us to take the issue of robustness seriously and soon lead to learning systems that convincingly outperform non-learning planners across a wide range of domains given proper training experience.

I hope to see a wide range of approaches entered into the competition. I’ll mention two styles of approaches that might be particular interesting to readers of this blog.

First, one might consider applying reinforcement learning to learn “generalized policies” that can be applied to any problem from a domain. Recall that here the domain model is provided to us, so applying RL would mean that the domain model is used as a sort of simulator in which an RL algorithm is run. RL is particularly difficult in these domains due to the challenges in developing an appropriate representation for learning value functions and/or policies—the states can be viewed as sets of ground relational atoms, rather than the more typical n-dimensional vectors common in RL. Another challenge is the extremely sparse reward, which is obtained only at goal states. There has been some work on applying RL to IPC-style domains (e.g. relational reinforcement learning, approximate policy iteration, policy gradient) but much improvement is needed to compete with non-learning planners.

Second, one might consider structured-classification techniques for this problem. Here one might view the planning problem as an input X and the plan as the structured output Y. Training data can be generated by solving example planning problems using state-of-the-art planners perhaps using significant resources. This approach has been studied under the name max-margin planning, but applied to a very different class of planning problems. Other work has considered applying the Learning as Search Optimization (LaSO) framework to IPC-style domains with some success. Some of the challenges here are to automatically produce an appropriate feature set given a planning domain and ambiguity in the training data. Ambiguity here refers to the fact that there are often a huge number of equally good plans for a given problem and the training data has only one or a small number of them, making the training data incomplete.

« Newer PostsOlder Posts »

Powered by WordPress