Contributions to the Theory of Finite-State Based Grammars

This dissertation is a theoretical study of finite-state based grammars used in natural language processing. The study is concerned with certain varieties of finite-state intersection grammars (FSIG) whose parsers define regular relations between surface strings and annotated surface strings. The study focuses on the following three aspects of FSIGs: (i) Computational complexity of grammars under limiting parameters In the study, the computational complexity in practical natural language processing is approached through performance-motivated parameters on structural complexity. Each parameter splits some grammars in the Chomsky hierarchy into an infinite set of subset approximations. When the approximations are regular, they seem to fall into the logarithmic-time hierarchyand the dot-depth hierarchy of star-free regular languages. This theoretical result is important and possibly relevant to grammar induction. (ii) Linguistically applicable structural representations Related to the linguistically applicable representations of syntactic entities, the study contains new bracketing schemes that cope with dependency links, left- and right branching, crossing dependencies and spurious ambiguity. New grammar representations that resemble the Chomsky-Schützenberger representation of context-free languages are presented in the study, and they include, in particular, representations for mildly context-sensitive non-projective dependency grammars whose performance-motivated approximations are linear time parseable. (iii) Compilation and simplification of linguistic constraints Efficient compilation methods for certain regular operations such as generalized restriction are presented. These include an elegant algorithm that has already been adopted as the approach in a proprietary finite-state tool. In addition to the compilation methods, an approach to on-the-fly simplifications of finite-state representations for parse forests is sketched. These findings are tightly coupled with each other under the theme of locality. I argue that the findings help us to develop better, linguistically oriented formalisms for finite-state parsing and to develop more efficient parsers for natural language processing. Avainsanat: syntactic parsing, finite-state automata, dependency grammar, first-order logic, linguistic performance, star-free regular approximations, mildly context-sensitive grammars

Comprehensive Path-sensitive Data-flow Analysis

Data-flow analysis is an integral part of any aggressive optimizing compiler. We propose a framework for improving the precision of data-flow analysis in the presence of complex control-flow. W initially perform data-flow analysis to determine those control-flow merges which cause the loss in data-flow analysis precision. The control-flow graph of the program is then restructured such that performing data-flow analysis on the resulting restructured graph gives more precise results. The proposed framework is both simple, involving the familiar notion of product automata, and also general, since it is applicable to any forward data-flow analysis. Apart from proving that our restructuring process is correct, we also show that restructuring is effective in that it necessarily leads to more optimization opportunities. Furthermore, the framework handles the trade-off between the increase in data-flow precision and the code size increase inherent in the restructuring. We show that determining an optimal restructuring is NP-hard, and propose and evaluate a greedy strategy. The framework has been implemented in the Scale research compiler, and instantiated for the specific problem of Constant Propagation. On the SPECINT 2000 benchmark suite we observe an average speedup of 4% in the running times over Wegman-Zadeck conditional constant propagation algorithm and 2% over a purely path profile guided approach.

On linear order and computation : The expressiveness of interactive computations on linear orders and computations indexed by ordinals

We solve the Dynamic Ehrenfeucht-Fra\"iss\'e Game on linear orders for both players, yielding a normal form for quantifier-rank equivalence classes of linear orders in first-order logic, infinitary logic, and generalized-infinitary logics with linearly ordered clocks. We show that Scott Sentences can be manipulated quickly, classified into local information, and consistency can be decided effectively in the length of the Scott Sentence. We describe a finite set of linked automata moving continuously on a linear order. Running them on ordinals, we compute the ordinal truth predicate and compute truth in the constructible universe of set-theory. Among the corollaries are a study of semi-models as efficient database of both model-theoretic and formulaic information, and a new proof of the atomicity of the Boolean algebra of sentences consistent with the theory of linear order -- i.e., that the finitely axiomatized theories of linear order are dense.

On the expressive power of behavioral profiles

Behavioral profiles have been proposed as a behavioral abstraction of dynamic systems, specifically in the context of business process modeling. A behavioral profile can be seen as a complete graph over a set of task labels, where each edge is annotated with one relation from a given set of binary behavioral relations. Since their introduction, behavioral profiles were argued to provide a convenient way for comparing pairs of process models with respect to their behavior or computing behavioral similarity between process models. Still, as of today, there is little understanding of the expressive power of behavioral profiles. Via counter-examples, several authors have shown that behavioral profiles over various sets of behavioral relations cannot distinguish certain systems up to trace equivalence, even for restricted classes of systems represented as safe workflow nets. This paper studies the expressive power of behavioral profiles from two angles. Firstly, the paper investigates the expressive power of behavioral profiles and systems captured as acyclic workflow nets. It is shown that for unlabeled acyclic workflow net systems, behavioral profiles over a simple set of behavioral relations are expressive up to configuration equivalence. When systems are labeled, this result does not hold for any of several previously proposed sets of behavioral relations. Secondly, the paper compares the expressive power of behavioral profiles and regular languages. It is shown that for any set of behavioral relations, behavioral profiles are strictly less expressive than regular languages, entailing that behavioral profiles cannot be used to decide trace equivalence of finite automata and thus Petri nets.

Analysis of DNA sequence transformations on grids

Study of the evolution of species or organisms is essential for various biological applications. Evolution is typically studied at the molecular level by analyzing the mutations of DNA sequences of organisms. Techniques have been developed for building phylogenetic or evolutionary trees for a set of sequences. Though phylogenetic trees capture the overall evolutionary relationships among the sequences, they do not reveal fine-level details of the evolution. In this work, we attempt to resolve various fine-level sequence transformation details associated with a phylogenetic tree using cellular automata. In particular, our work tries to determine the cellular automata rules for neighbor-dependent mutations of segments of DNA sequences. We also determine the number of time steps needed for evolution of a progeny from an ancestor and the unknown segments of the intermediate sequences in the phylogenetic tree. Due to the existence of vast number of cellular automata rules, we have developed a grid system that performs parallel guided explorations of the rules on grid resources. We demonstrate our techniques by conducting experiments on a grid comprising machines in three countries and obtaining potentially useful statistics regarding evolutions in three HIV sequences. In particular, our work is able to verify the phenomenon of neighbor-dependent mutations and find that certain combinations of neighbor-dependent mutations, defined by a cellular automata rule, occur with greater than 90% probability. We also find the average number of time steps for mutations for some branches of phylogenetic tree over a large number of possible transformations with standard deviations less than 2.

Fast in-memory XPath search using compressed indexes

A large fraction of an XML document typically consists of text data. The XPath query language allows text search via the equal, contains, and starts-with predicates. Such predicates can be efficiently implemented using a compressed self-index of the document's text nodes. Most queries, however, contain some parts querying the text of the document, plus some parts querying the tree structure. It is therefore a challenge to choose an appropriate evaluation order for a given query, which optimally leverages the execution speeds of the text and tree indexes. Here the SXSI system is introduced. It stores the tree structure of an XML document using a bit array of opening and closing brackets plus a sequence of labels, and stores the text nodes of the document using a global compressed self-index. On top of these indexes sits an XPath query engine that is based on tree automata. The engine uses fast counting queries of the text index in order to dynamically determine whether to evaluate top-down or bottom-up with respect to the tree structure. The resulting system has several advantages over existing systems: (1) on pure tree queries (without text search) such as the XPathMark queries, the SXSI system performs on par or better than the fastest known systems MonetDB and Qizx, (2) on queries that use text search, SXSI outperforms the existing systems by 1-3 orders of magnitude (depending on the size of the result set), and (3) with respect to memory consumption, SXSI outperforms all other systems for counting-only queries.

The Problem of Implementation and its Relation to the Philosophy of Cognitive Science

According to certain arguments, computation is observer-relative either in the sense that many physical systems implement many computations (Hilary Putnam), or in the sense that almost all physical systems implement all computations (John Searle). If sound, these arguments have a potentially devastating consequence for the computational theory of mind: if arbitrary physical systems can be seen to implement arbitrary computations, the notion of computation seems to lose all explanatory power as far as brains and minds are concerned. David Chalmers and B. Jack Copeland have attempted to counter these relativist arguments by placing certain constraints on the definition of implementation. In this thesis, I examine their proposals and find both wanting in some respects. During the course of this examination, I give a formal definition of the class of combinatorial-state automata , upon which Chalmers s account of implementation is based. I show that this definition implies two theorems (one an observation due to Curtis Brown) concerning the computational power of combinatorial-state automata, theorems which speak against founding the theory of implementation upon this formalism. Toward the end of the thesis, I sketch a definition of the implementation of Turing machines in dynamical systems, and offer this as an alternative to Chalmers s and Copeland s accounts of implementation. I demonstrate that the definition does not imply Searle s claim for the universal implementation of computations. However, the definition may support claims that are weaker than Searle s, yet still troubling to the computationalist. There remains a kernel of relativity in implementation at any rate, since the interpretation of physical systems seems itself to be an observer-relative matter, to some degree at least. This observation helps clarify the role the notion of computation can play in cognitive science. Specifically, I will argue that the notion should be conceived as an instrumental rather than as a fundamental or foundational one.

Absolutely expedient algorithms for learning Nash equilibria

This paper considers a multi-person discrete game with random payoffs. The distribution of the random payoff is unknown to the players and further none of the players know the strategies or the actual moves of other players. A class of absolutely expedient learning algorithms for the game based on a decentralised team of Learning Automata is presented. These algorithms correspond, in some sense, to rational behaviour on the part of the players. All stable stationary points of the algorithm are shown to be Nash equilibria for the game. It is also shown that under some additional constraints on the game, the team will always converge to a Nash equilibrium.

Improved preprocessing methods for modulo scheduling algorithms

Instruction scheduling with an automaton-based resource conflict model is well-established for normal scheduling. Such models have been generalized to software pipelining in the modulo-scheduling framework. One weakness with existing methods is that a distinct automaton must be constructed for each combination of a reservation table and initiation interval. In this work, we present a different approach to model conflicts. We construct one automaton for each reservation table which acts as a compact encoding of all the conflict automata for this table, which can be recovered for use in modulo-scheduling. The basic premise of the construction is to move away from the Proebsting-Fraser model of conflict automaton to the Muller model of automaton modelling issue sequences. The latter turns out to be useful and efficient in this situation. Having constructed this automaton, we show how to improve the estimate of resource constrained initiation interval. Such a bound is always better than the average-use estimate. We show that our bound is safe: it is always lower than the true initiation interval. This use of the automaton is orthogonal to its use in modulo-scheduling. Once we generate the required information during pre-processing, we can compute the lower bound for a program without any further reference to the automaton.

Dynamically Reconfigurable Regular Expression Matching Architecture

Regular Expressions are generic representations for a string or a collection of strings. This paper focuses on implementation of a regular expression matching architecture on reconfigurable fabric like FPGA. We present a Nondeterministic Finite Automata based implementation with extended regular expression syntax set compared to previous approaches. We also describe a dynamically reconfigurable generic block that implements the supported regular expression syntax. This enables formation of the regular expression hardware by a simple cascade of generic blocks as well as a possibility for reconfiguring the generic blocks to change the regular expression being matched. Further,we have developed an HDL code generator to obtain the VHDL description of the hardware for any regular expression set. Our optimized regular expression engine achieves a throughput of 2.45 Gbps. Our dynamically reconfigurable regular expression engine achieves a throughput of 0.8 Gbps using 12 FPGA slices per generic block on Xilinx Virtex2Pro FPGA.

Faster Algorithms for Incremental Topological Ordering.Robert Tarjan

We present two online algorithms for maintaining a topological order of a directed acyclic graph as arcs are added, and detecting a cycle when one is created. Our first algorithm takes O(m 1/2) amortized time per arc and our second algorithm takes O(n 2.5/m) amortized time per arc, where n is the number of vertices and m is the total number of arcs. For sparse graphs, our O(m 1/2) bound improves the best previous bound by a factor of logn and is tight to within a constant factor for a natural class of algorithms that includes all the existing ones. Our main insight is that the two-way search method of previous algorithms does not require an ordered search, but can be more general, allowing us to avoid the use of heaps (priority queues). Instead, the deterministic version of our algorithm uses (approximate) median-finding; the randomized version of our algorithm uses uniform random sampling. For dense graphs, our O(n 2.5/m) bound improves the best previously published bound by a factor of n 1/4 and a recent bound obtained independently of our work by a factor of logn. Our main insight is that graph search is wasteful when the graph is dense and can be avoided by searching the topological order space instead. Our algorithms extend to the maintenance of strong components, in the same asymptotic time bounds.

Phylogenetic Predictions on Grids

A phylogenetic or evolutionary tree is constructed from a set of species or DNA sequences and depicts the relatedness between the sequences. Predictions of future sequences in a phylogenetic tree are important for a variety of applications including drug discovery, pharmaceutical research and disease control. In this work, we predict future DNA sequences in a phylogenetic tree using cellular automata. Cellular automata are used for modeling neighbor-dependent mutations from an ancestor to a progeny in a branch of the phylogenetic tree. Since the number of possible ways of transformations from an ancestor to a progeny is huge, we use computational grids and middleware techniques to explore the large number of cellular automata rules used for the mutations. We use the popular and recurring neighbor-based transitions or mutations to predict the progeny sequences in the phylogenetic tree. We performed predictions for three types of sequences, namely, triose phosphate isomerase, pyruvate kinase, and polyketide synthase sequences, by obtaining cellular automata rules on a grid consisting of 29 machines in 4 clusters located in 4 countries, and compared the predictions of the sequences using our method with predictions by random methods. We found that in all cases, our method gave about 40% better predictions than the random methods.

An (O)over-bar(m(2)n) randomized algorithm to compute a minimum cycle basis of a directed graph

We consider the problem of computing a minimum cycle basis in a directed graph G. The input to this problem is a directed graph whose arcs have positive weights. In this problem a {- 1, 0, 1} incidence vector is associated with each cycle and the vector space over Q generated by these vectors is the cycle space of G. A set of cycles is called a cycle basis of G if it forms a basis for its cycle space. A cycle basis where the sum of weights of the cycles is minimum is called a minimum cycle basis of G. The current fastest algorithm for computing a minimum cycle basis in a directed graph with m arcs and n vertices runs in O(m(w+1)n) time (where w < 2.376 is the exponent of matrix multiplication). If one allows randomization, then an (O) over tilde (m(3)n) algorithm is known for this problem. In this paper we present a simple (O) over tilde (m(2)n) randomized algorithm for this problem. The problem of computing a minimum cycle basis in an undirected graph has been well-studied. In this problem a {0, 1} incidence vector is associated with each cycle and the vector space over F-2 generated by these vectors is the cycle space of the graph. The fastest known algorithm for computing a minimum cycle basis in an undirected graph runs in O(m(2)n + mn(2) logn) time and our randomized algorithm for directed graphs almost matches this running time.

A neural network based automatic generation control design through reinforcement learning

This paper presents the design and implementation of a learning controller for the Automatic Generation Control (AGC) in power systems based on a reinforcement learning (RL) framework. In contrast to the recent RL scheme for AGC proposed by us, the present method permits handling of power system variables such as Area Control Error (ACE) and deviations from scheduled frequency and tie-line flows as continuous variables. (In the earlier scheme, these variables have to be quantized into finitely many levels). The optimal control law is arrived at in the RL framework by making use of Q-learning strategy. Since the state variables are continuous, we propose the use of Radial Basis Function (RBF) neural networks to compute the Q-values for a given input state. Since, in this application we cannot provide training data appropriate for the standard supervised learning framework, a reinforcement learning algorithm is employed to train the RBF network. We also employ a novel exploration strategy, based on a Learning Automata algorithm,for generating training samples during Q-learning. The proposed scheme, in addition to being simple to implement, inherits all the attractive features of an RL scheme such as model independent design, flexibility in control objective specification, robustness etc. Two implementations of the proposed approach are presented. Through simulation studies the attractiveness of this approach is demonstrated.