Learning Deterministic Finite Automata from Interleaved Strings.

Inferring Probability Distributions of Graph Size and Node Degree from Stochastic Graph Grammars.

Ensembles in adversarial classification for spam.

A Context Driven Approach for Workflow Mining.
Existing work on workflow mining ignores the dataflow aspect of the problem. This is not acceptable for service-oriented applications that use Web services with typed inputs and outputs. We propose a novel algorithm WIT (Workflow Inference from Traces) which identifies the context similarities of the observed actions based on the dataflow and uses model merging techniques to generalize the control flow and the dataflow simultaneously. We identify the class of workflows that WIT can learn correctly. We implemented WIT and tested it on a real world medical scheduling domain where WIT was able to find a good approximation of the target workflow. 1
Reports of the AAAI 2009 Spring Symposia.

Reports of the AAAI 2009 Spring Symposia.

Lexical and Grammatical Inference.

Essential Phenomena of General Intelligence.

Which Came First%2C the Grammar or the Lexicon%3F.

Learning in the Lexical%2DGrammatical Interface.

Which Came First%2C the Grammar or the Lexicon%3F.

Estimating Graph Parameters Using Graph Grammars.

Estimating Graph Parameters Using Graph Grammars.

AAAI Fall Symposium Reports.

AAAI Fall Symposium Reports.

A Self%2DHelp Guide For Autonomous Systems.

A Too%2DClever Ranking Method.

Introduction Special Issue on Applications of Grammatical Inference.

Real%2DTime Identification of Operating Room State from Video.

UNDERTOW Multi%2DLevel Segmentation of Real%2DValued Time Series.

The Marchitecture A Cognitive Architecture for a Robot Baby.

Learning by Reading by Learning to Read.

Using Ontologies and the Web to Learn Lexical Semantics.

A Review of Recent Research in Metareasoning and Metalearning.
Recent years have seen a resurgence of interest in the use of metacognition in intelligent systems. This essay is part of a small section meant to give interested researchers an overview and sampling of the kinds of work currently being pursued in this broad area. The current essay offers a review of recent research in two main topic areas: the monitoring and control of reasoning (metareasoning) and the monitoring and control of learning (metalearning). What is metacognition in computation? Rosie (the robot maid from the TV show The Jetsons) spends her days cooking, cleaning, ironing, and attending to the usual household tasks of late 21 st century life. Because of a bug in one of her memory chips, however, she almost always forgets to buy dog food when she goes out. She has an adequate recovery plan for this: she simply feeds Astro some of the Jetson’s dinner. But 21 st century human food is expensive, so this strategy is wasteful. Realizing this, and recognizing that she has forgotten several times, Rosie adopts a special strategy to help her remember: she sticks the spare dog collar in her
Detecting Spam Blogs A Machine Learning Approach.
Weblogs or blogs are an important new way to publish information, engage in discussions, and form communities on the Internet. The Blogosphere has unfortunately been infected by several varieties of spam-like content. Blog search engines, for example, are inundated by posts from splogs – false blogs with machine generated or hijacked content whose sole purpose is to host ads or raise the PageRank of target sites. We discuss how SVM models based on local and link-based features can be used to detect splogs. We present an evaluation of learned models and their utility to blog search engines; systems that employ techniques differing from those of conventional web search engines.
Inferring Grammars for Mildly Context Sensitive Languages in Polynomial%2DTime.

Discovering Patterns in Real%2DValued Time Series.
Abstract. This paper describes an algorithm for discovering variable length patterns in real-valued time series. In contrast to most existing pattern discovery algorithms, ours does not first discretize the data, runs in linear time, and requires constant memory. These properties are obtained by sampling the data stream rather than processing all of the data. Empirical results show that the algorithm performs well on both synthetic and real data when compared to an exhaustive algorithm. 1
Marcus Hutter%2C Universal Artificial Intelligence%2C Springer %282004%29.

The metacognitive loop I Enhancing reinforcement learning with metacognitive monitoring and control for improved perturbation tolerance%7C%7C.

2005 AAAI Spring Symposium on Metacognition in Computation

Discovering Domain%2DSpecific Composite Kernels.

Transfer in Learning by Doing.

The Cruncher Automatic Concept Formation Using Minimum Description Length.
Abstract. We present The Cruncher, a simple representation framework and algorithm based on minimum description length for automatically forming an ontology of concepts from attribute-value data sets. Although unsupervised, when The Cruncher is applied to an animal data set, it produces a nearly zoologically accurate categorization. We demonstrate The Cruncher’s utility for finding useful macro-actions in Reinforcement Learning, and for learning models from uninterpreted sensor data. We discuss advantages The Cruncher has over concept lattices and hierarchical clustering. 1
Reports on the 2005 AAAI Spring Symposium Series.
This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author&#039;s copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder. © 2005 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE. For more information, please see www.ieee.org/portal/pages/about/documentation/copyright/polilink.html.
R. Siegwart and I. Nourbakhsh

On the Relationship between Lexical Semantics and Syntax for the Inference of Context%2DFree Grammars.

The 2004 AAAI Spring Symposium Series.

Generating Web Graphs with Embedded Communities.
Web graphs have been very useful in the structural and statistical analysis of the web. Various models have been proposed to simulate web graphs that generate degree distri-butions similar to the web. Real world blog networks resemble many properties of web graphs. But the dynamic nature of the blogosphere and the link structure evolving due to blog readership and social interactions is not well expressed by the existing models. In this research we propose a model for a blogger to construct blog graphs. We com-bine the existing preferential attachment and random attachment model to generate blog graphs which are type of scale-free networks. The blogger is modeled using read, write, idle states and finite read memory. The combination of these techniques helps in evolution of time stamped blog-blog and post-post network through citations within the blog-blog network. Other parameters like the growth function and the randomness in reading and writing posts help in the formation of graphs with different structural properties. We empirically show that these simulated blog graph exhibits properties similar to the real world blog networks in their degree distributions, degree correlations and clustering coefficient. We believe that this model will help researchers to evaluate and analyze the properties of the blogosphere and facilitate the testing of new algorithms.
The 2004 AAAI Spring Symposium Series.

Finding aliases on the web using latent semantic analysis.

Automatically Generated DAML Markup for Semistructured Documents.
The semantic web is becoming a realizable  technology due to the efforts of researchers to  develop semantic markup languages such as the  DARPA Agent Markup Language (DAML). A  major problem that faces the semantic web  community is that most information sources on the  web today lack semantic markup. To fully realize  the potential of the semantic web, we must find a  way to automatically upgrade information sources  with semantic markup. We have developed a  system based on the STALKER algorithm that  automatically generates DAML markup for a set  of documents based on previously seen labeled  training documents. Our ultimate goal is to  develop a program that performs DAML markup  for arbitrary documents given a particular  ontology.


Estimating Maximum Likelihood Parameters for Stochastic Context%2DFree Graph Grammars.

Contentful Mental States for Robot Baby.
In this paper we claim that meaningful representations can  be learned by programs, although today they are almost always  designed by skilled engineers. We discuss several kinds  of meaning that representations might have, and focus on a  functional notion of meaning as appropriate for programs to  learn. Specifically, a representation is meaningful if it incorporates  an indicator of external conditions and if the indicator  relation informs action. We survey methods for inducing  kinds of representations we call structural abstractions. Prototypes  of sensory time series are one kind of structural abstraction,  and though they are not denoting or compositional,  they do support planning. Deictic representations of objects  and prototype representations of words enable a program to  learn the denotational meanings of words. Finally, we discuss  two algorithms designed to find the macroscopic structure of  episodes in a domain-independent way.
PERUSE An Unsupervised Algorithm for Finding Recurrig Patterns in Time Series.

Estimating Grammar Parameters Using Bounded Memory.
Estimating the parameters of stochastic context-free  grammars (SCFGs) from data (i.e., strings) is an important,  well-studied problem. Almost without exception,  existing approaches make repeated passes over the  training data. The memory requirements of such algorithms  are ill-suited for embedded agents exposed to  large amounts of training data over long periods of time.
Learning k%2DReversible Context%2DFree Grammars from Positive Structural Examples.
Inference of formal languages from positive examples is more difficult than inference from positive and negative examples. The entire class of regular languages can be identified in the limit from positive and negative examples, but not from positive examples alone (Gold, 1967). Angluin introduced an algorithm that efficiently identifies k-reversible regular languages for k _ 0, a subset of the class of regular languages, from positive examples (Angluin, 1982). Sakakibara defined a subset of context-free grammars that can be characterized as zero-reversible and extended Angluin&#039;s algorithm to learn them from positive structural examples (Sakakibara, 1992).
The Thing that we Tried Didn%27t Work very Well Deictic Representation in Reinforcement Learning.

Using latent semantic analysis to find different names for the same entity in free text.
A common problem faced when gathering information from the web is the use of different names to  refer to the same entity. For example, the city in India referred to as Bombay in some documents may  be referred to as Mumbai in others because its name officially changed from the former to the latter in  1995. Multiplicity of names can cause relevant documents to be missed by search engines. Our goal  is to develop an automated system that discovers additional names for an entity given just one of its  names. Latent semantic analysis (LSA) is generally thought to be well-suited for this task (Berry &amp;  Fierro 1996). We demonstrate empirically that under a broad range of circumstances LSA performs  poorly, and describe a two-stage algorithm based on LSA that performs significantly better.
Robot Baby 2001.

The Importance of Being Discrete Learning Classes of Actions and Outcomes through Interaction.

Robot Baby 2001.

Using Dynamic Time Warping to Bootstrap HMM%2DBased Clustering of Time Series.

A Method for Clustering the Experiences of a Mobile Robot that Accords with Human Judgments.
If robotic agents are to act autonomously they must have the  ability to construct and reason about models of their physical  environment. For example, planning to achieve goals requires  knowledge of how the robot&#039;s actions affect the state  of the world over time. The traditional approach of handcoding  this knowledge is often quite difficult, especially for  robotic agents with rich sensing abilities that exist in dynamic  and uncertain environments. Ideally, robots would acquire  knowledge of their environment and then use this knowledge  to act. We present an unsupervised learning method that allows  a robotic agent to identify and represent qualitatively  different outcomes of actions. Experiments with a Pioneer-1  mobile robot demonstrate the utility of the approach with respect  to capturing the structure and dynamics of a complex,  real-world environment, and show that the models acquired  by the robot correlate surprisingly well with human models  of the environment.  
Learning Planning Operators in Real%2DWorld%2C Partially Observable Environments.
We are interested in the development of activities in  situated, embodied agents such as mobile robots. Central  to our theory of development is means-ends analysis  planning, and as such, we must rely on operator  models that can express the eects of a robot&#039;s action  in a dynamic, partially-observable environment. This  paper presents a two-step process which employs clustering  and decision tree induction to perform unsupervised  learning of operator models from simple interactions  between an agent and its environment. We report  our ndings with an implementation of this system on  a Pioneer-1 mobile robot.  
Identifying qualitatively different outcomes of actions gaining autonomy through learning.
This paper presents an unsupervised method for learning  models of environmental dynamics based on clustering multivariate time series. Experiments with a Pioneer-1 mobile robot demonstrate the utility of the method and show that the models acquired by the robot correlate surprisingly well with human models of the environment. Individual time series are obtained by recording the output of a subset of an agent&#039;s sensors. We call these time series experiences. An example of a sensor subset on the Pioneer-1 robot is its array of seven sonars. Each sonar returns the distance to the closest object in the direction that it points. Recording of time series is usually triggered by events, such as the initiation of a particular action. Each time a given event occurs, the time series that was recorded is added to a bucket associated with that event. Once a sucient number of experiences are recorded, clusters can be formed. Clustering requires a measure of similarity between multi
Toward natural language interfaces for robotic agents grounding linguistic meaning in sensors.
This paper explores one problem that lies in the way of this communicative ideal for robotic agents - the acquisition and grounding of linguistic meaning. How is it possible for a robot to identify and represent features of the physical world, as mediated by its ability to sense the world, that are relevant to understanding natural language utterances?  This paper describes the results of an experiment in which human subjects generated unrestricted natural language utterances to describe the activities of a Pioneer-1 mobile robot. A combination of word clus- tering applied to the utterances and a common subsequence algorithm applied to the time series of sensor values recorded by the robot made it possible for the Pioneer-1 to identify and represent a variety of features of its environment to which the utterances referred
Toward a Theoretical Understanding of Why and When Decision Tree Pruning Algorithms Fail.
Recent empirical studies revealed two surprising  pathologies of several common decision tree pruning  algorithms. First, tree size is often a linear function  of training set size, even when additional tree structure  yields no increase in accuracy. Second, building  trees with data in which the class label and the attributes  are independent often results in large trees.
Efficient Mining of Statistical Dependencies.
this paper, we describe three methods for reducing the size of the search space that msdd considers and empirically evaluate their utility. The remainder of this section discusses the core msdd algorithm. Section 2 describes the three search space reduction methods and section 3 summarizes our empirical work with them. Finally, section 4 summarizes. Let D be a database containing T records: D = fR 1 ; : : : ; R T g. Each record is a set of unique tokens taken from the alphabet \Sigma, and the number of tokens may vary from record to record:
Efficient Progressive Sampling.
Having access to massive amounts of data does not necessarily imply that induction algorithms must use them all.  Samples often provide the same accuracy with far less computational cost. However, the correct sample size rarely is obvious. We analyze methods for progressive sampling---  using progressively larger samples as long as model accuracy improves. We explore several notions of efficient progressive sampling. We analyze efficiency relative to induction with all instances; we show that a simple, geometric sampling schedule is asymptotically optimal, and we describe how best to take into account prior expectations of accuracy convergence. We then describe the issues involved in instantiating an efficient progressive sampler, including how to detect convergence. Finally, we provide empirical results comparing a variety of progressive sampling methods. We conclude that progressive sampling can be remarkably efficient.   1 Introduction  Induction algorithms face competing requiremen...
Identifying Distinctive Subsequences in Multivariate Time Series by Clustering.
Most time series comparison algorithms attempt to discover what the members of a  set of time series have in common. We investigate a different problem, determining  what distinguishes time series in that set from other time series obtained from the  same source. In both cases the goal is to identify shared patterns, though in the latter  case those patterns must be distinctiveaswell. An efficient incremental algorithm for  identifying distinctive subsequences in multivariate, real-valued time series is described  and evaluated with data from two very different sources: the response of a set of  bandpass filters to human speech and the sensors of a mobile robot.
Learning Deterministic Finite Automaton with a Recurrent Neural Network.

Large Datasets Lead to Overly Complex Models An Explanation and a Solution.
This paper explores unexpected results that lie at  the intersection of two common themes in the KDD  community: large datasets and the goal of building  compact models. Experiments with many different  datasets and several model construction algorithms  (including tree learning algorithms suchasc4.5 with  three different pruning methods, and rule learning algorithms  such as c4.5rules and ripper) show that  increasing the amount of data used to build a model often  results in a linear increase in model size, even when  that additional complexity results in no significantincrease  in model accuracy. Despite the promise of better  parameter estimation held out by large datasets, as a  practical matter, models built with large amounts of  data are often needlessly complex and cumbersome. In  the case of decision trees, the cause of this pathology is  identified as a bias inherentinseveral common pruning  techniques. Pruning errors made low in the tree, where  there is insufficient data to make accurate parameter  estimates, are propagated and magnified higher in the  tree, working against the accurate parameter estimates  that are made possible there by abundant data. We  propose a general solution to this problem based on a  statistical technique known as randomization testing,  and empirically evaluate its utility.
Neo Learning Conceptual Knowledge by Sensorimotor Interaction with an Environment.
Recent developments in philosophy, linguistics, developmental  psychology and artificial intelligence make it  possible to envision a developmental path for an artificial  agent, grounded in activity-based sensorimotor  representations. This paper describes how Neo, an artificial  agent, learns concepts by interacting with its  simulated environment. Relatively little prior structure  is required to learn fairly accurate representations  of objects, activities, locations and other aspects of  Neo&#039;s experience. We show how classes (categories) can  be abstracted from these representations, and discuss  how our representation might be extended to express  physical schemas, general, domain-independent activities  that could be the building blocks of concept formation.
Parallel and Distributed Search for Structure in Multivariate Time Series.
Efficient data mining algorithms are crucial for effective knowledge discovery. We  present the Multi-Stream Dependency Detection (MSDD) data mining algorithm that  performs a systematic search for structure in multivariate time series of categorical  data. The systematicity of MSDD&#039;s searchmakes implementation of both parallel and  distributed versions straightforward. Distributing the search for structure over multiple  processors or networked machines makes mining of large numbers of databases or very  large databases feasible. We present results showing that MSDD efficiently finds complex  structure in multivariate time series, and that the distributed version finds the same  structure in approximately 1/n of the time required by MSDD, where n  is the number of  machines across which the search is distributed. MSDD differs
The Effects of Training Set Size on Decision Tree Complexity.
This paper presents experiments with 19 datasets and 5 decision tree  pruning algorithms that show that increasing training set size often  results in a linear increase in tree size, even when that additional  complexity results in no significant increase in classification accuracy.
Cooperative information%2Dgathering a distributed problem%2Dsolving approach.
We contrast two approaches to the problem of information gathering that maybe  characterized as distributed processing and distributed problem solving. The former  is characteristic of most existing information gathering systems, while the latter is  central to research in multi-agent systems. We examine features of complex information  carrying environments and the information gathering task that demonstrate both the  utility of viewing information gathering as distributed problem solving and difficulties  with viewing it as distributed processing. We propose a new approach to information  gathering based on the distributed problem solving paradigm and its attendant body of  research in multi-agent systems and distributed artificial intelligence. This approach,  called Cooperative Information Gathering, involves concurrent, asynchronous discovery  and composition of information spread across a network of information servers. Top  level queries drive the creation of partially elaborated information gathering plans,  resulting in the employment of multiple semi-autonomous, cooperative agents for the  purpose of achieving goals and subgoals within those plans. The system as a whole  satisfices, trading off solution quality and search cost while respecting user-imposed  deadlines. We also survey currentwork on distributed and agent-based approaches to  information gathering.
Building Simple Models A Case Study with Decision Trees.

Cooperative information%2Dgathering a distributed problem%2Dsolving approach.
We contrast two approaches to the problem of information gathering that maybe  characterized as distributed processing and distributed problem solving. The former  is characteristic of most existing information gathering systems, while the latter is  central to research in multi-agent systems. We examine features of complex information  carrying environments and the information gathering task that demonstrate both the  utility of viewing information gathering as distributed problem solving and difficulties  with viewing it as distributed processing. We propose a new approach to information  gathering based on the distributed problem solving paradigm and its attendant body of  research in multi-agent systems and distributed artificial intelligence. This approach,  called Cooperative Information Gathering, involves concurrent, asynchronous discovery  and composition of information spread across a network of information servers. Top  level queries drive the creation of partially elaborated information gathering plans,  resulting in the employment of multiple semi-autonomous, cooperative agents for the  purpose of achieving goals and subgoals within those plans. The system as a whole  satisfices, trading off solution quality and search cost while respecting user-imposed  deadlines. We also survey currentwork on distributed and agent-based approaches to  information gathering.
Searching for Planning Operators with Context%2DDependent and Probabilistic Effects.
Providing a complete and accurate domain model  for an agent situated in a complex environmentcan  be an extremely difficult task. Actions mayhave  different effects depending on the context in which  they are taken, and actions mayormay not induce  their intended effects, with the probability of success  again depending on context. We present an algorithm  for automatically learning planning operators  with context-dependent and probabilistic effects in environments  where exogenous events change the state  of the world. Empirical results show that the algorithm  successfully finds operators that capture the  true structure of an agent&#039;s interactions with its environment,  and avoids spurious associations between  actions and exogenous events.
Searching for Structure in Multiple Streams of Data.
Finding structure in multiple streams of data  is an important problem. Consider the  streams of data flowing from a robot&#039;s sensors,  the monitors in an intensive care unit,  or periodic measurements of various indicators  of the health of the economy. There  is clearly utility in determining howcurrent  and past values in those streams are related  to future values. We formulate the problem  of finding structure in multiple streams  of categorical data as searchover the space  of dependencies, unexpectedly frequentor  infrequent co-occurrences, between complex  patterns of values that can appear in the  streams. Based on that formulation, we develop  the Multi-Stream Dependency Detection   (msdd) algorithm that performs an efficient  systematic searchover the space of all  possible dependencies. Dependency strength  is evaluated with a statistical measure of nonindependence,  and bounds that we derivefor  the value of that measure allow the search  to be pruned. Due to the pruning,    can find the k strongest dependencies in the  streams by examining only a fraction of the  search space.
Toward a Plan Steering Agent Experiments with Schedule Maintenance.
When a plan involves hundreds or thousands of events over time  it can be difficult or impossible to tell whether those events are unfolding  &#034;according to plan&#034; and to assess the impact of dynamic plan  modifications. Pathological states may arise in which goals cannot be  attained or are attained too slowly. Plan steering is an agent-based  approach to this problem. The agent monitors an unfolding plan, detects  and predicts pathological situations, and develops dynamic plan  modifications that will steer the plan around the problem. Wepresent  results for such a system that performs the related task of schedule  maintenance in the transportation planning domain. The agentuses  limited domain knowledge and simple heuristics and is able to improve  throughput significantly.
