Nicholas S. Flann

A Study of Explanation-Based Methods for Inductive Learning

This paper formalizes a new learning-from-examples problem: identifying a correct concept definition from positive examples such that the concept is some specialization of a target concept defined by a domain theory. It describes an empirical study that evaluates three methods for solving this problem: explanation-based generalization (EBG), multiple example explanation-based generalization (mEBG), and a new method, induction over explanations (IOE). The study demonstrates that the two existing methods (EBG and mEBG) exhibit two shortcomings: (a) they rarely identify the correct definition, and (b) they are brittle in that their success depends greatly on the choice of encoding of the domain theory rules. The study demonstrates that the new method, IOE, does not exhibit these shortcomings. This method applies the domain theory to construct explanations from multiple training examples as in mEBG, but forms the concept definition by employing a similarity-based generalization policy over the explanations. IOE has the advantage that an explicit domain theory can be exploited to aid the learning process, the dependence on the initial encoding of the domain theory is significantly reduced, and the correct concepts can be learned from few examples. The study evaluates the methods in the context of an implemented system, called Wyl2, which learns a variety of concepts in chess including “skewer” and “knight-fork.”

Parallel genetic simulated annealing: a massively parallel SIMD algorithm

Many significant engineering and scientific problems involve optimization of some criteria over a combinatorial configuration space. The two methods most often used to solve these problems effectively-simulated annealing (SA) and genetic algorithms (GA)-do not easily lend themselves to massive parallel implementations. Simulated annealing is a naturally serial algorithm, while GA involves a selection process that requires global coordination. This paper introduces a new hybrid algorithm that inherits those aspects of GA that lend themselves to parallelization, and avoids serial bottle-necks of GA approaches by incorporating elements of SA to provide a completely parallel, easily scalable hybrid GA/SA method. This new method, called Genetic Simulated Annealing, does not require parallelization of any problem specific portions of a serial implementation-existing serial implementations can be incorporated as is. Results of a study on two difficult combinatorial optimization problems, a 100 city traveling salesperson problem and a 24 word, 12 bit error correcting code design problem, performed on a 16 K PE MasPar MP-1, indicate advantages over previous parallel GA and SA approaches. One of the key results is that the performance of the algorithm scales up linearly with the increase of processing elements, a feature not demonstrated by any previous parallel GA or SA approaches, which enables the new algorithm to utilize massive parallel architecture with maximum effectiveness. Additionally, the algorithm does not require careful choice of control parameters, a significant advantage over SA and GA.

Parallel Simulated Annealing and Genetic Algorithms: a Space of Hybrid Methods

Simulated annealing and genetic algorithms represent powerful optimization methods with complementary strengths and weaknesses. Hence, there is an interest in identifying hybrid methods (which combine features of both SA and GA) that exhibit performance superior than either method alone. This paper introduces a systematic approach to identifying these hybrids by defining a space of methods as a nondeterministic generating grammar. This space includes SA, GA, previously introduced hybrids and many new methods. An empirical evaluation has been completed for 14 methods from this space applied to 9 diverse optimization problems. Results demonstrate that the space contains promising new methods. In particular, a new method that combines the recombinative power of GAs and annealing schedule of SA is shown to be one of the best methods for all 9 optimization problems explored.

Explanation-Based Learning and Reinforcement Learning: A Unified View

In speedup-learning problems, where full descriptions of operators are known, both explanation-based learning (EBL) and reinforcement learning (RL) methods can be applied. This paper shows that both methods involve fundamentally the same process of propagating information backward from the goal toward the starting state. Most RL methods perform this propagation on a state-by-state basis, while EBL methods compute the weakest preconditions of operators, and hence, perform this propagation on a region-by-region basis. Barto, Bradtke, and Singh (1995) have observed that many algorithms for reinforcement learning can be viewed as asynchronous dynamic programming. Based on this observation, this paper shows how to develop dynamic programming versions of EBL, which we call region-based dynamic programming or Explanation-Based Reinforcement Learning (EBRL). The paper compares batch and online versions of EBRL to batch and online versions of point-based dynamic programming and to standard EBL. The results show that region-based dynamic programming combines the strengths of EBL (fast learning and the ability to scale to large state spaces) with the strengths of reinforcement learning algorithms (learning of optimal policies). Results are shown in chess endgames and in synthetic maze tasks.

Biocellion: accelerating computer simulation of multicellular biological system models

MOTIVATION Biological system behaviors are often the outcome of complex interactions among a large number of cells and their biotic and abiotic environment. Computational biologists attempt to understand, predict and manipulate biological system behavior through mathematical modeling and computer simulation. Discrete agent-based modeling (in combination with high-resolution grids to model the extracellular environment) is a popular approach for building biological system models. However, the computational complexity of this approach forces computational biologists to resort to coarser resolution approaches to simulate large biological systems. High-performance parallel computers have the potential to address the computing challenge, but writing efficient software for parallel computers is difficult and time-consuming. RESULTS We have developed Biocellion, a high-performance software framework, to solve this computing challenge using parallel computers. To support a wide range of multicellular biological system models, Biocellion asks users to provide their model specifics by filling the function body of pre-defined model routines. Using Biocellion, modelers without parallel computing expertise can efficiently exploit parallel computers with less effort than writing sequential programs from scratch. We simulate cell sorting, microbial patterning and a bacterial system in soil aggregate as case studies. AVAILABILITY AND IMPLEMENTATION Biocellion runs on x86 compatible systems with the 64 bit Linux operating system and is freely available for academic use. Visit http://biocellion.com for additional information.

Sorting unsigned permutations by reversals using multi-objective evolutionary algorithms with variable size individuals

Sorting by reversals is a simplified version of the genome rearrangement problem that seeks to discover the evolutionary relationship between different genomes, and is one of the many challenging problems in Bioinformatics. Solving the problem optimally has been proved to be NP-Hard and so a selection of approximation algorithms have been developed. In this paper a new mapping order is introduced to solve the problem of sorting unsigned permutations using a specialized multi-objective genetic algorithm. Our modified genetic algorithm uses a population with variable length individuals to maintain a worst time running time complexity of O(n4 log2 n), where n is the problem size. The results show that this approach is more effective than the 3/2 heuristic method and previous genetic algorithm approaches.

Regular mosaic pattern development: A study of the interplay between lateral inhibition, apoptosis and differential adhesion

BackgroundA significant body of literature is devoted to modeling developmental mechanisms that create patterns within groups of initially equivalent embryonic cells. Although it is clear that these mechanisms do not function in isolation, the timing of and interactions between these mechanisms during embryogenesis is not well known. In this work, a computational approach was taken to understand how lateral inhibition, differential adhesion and programmed cell death can interact to create a mosaic pattern of biologically realistic primary and secondary cells, such as that formed by sensory (primary) and supporting (secondary) cells of the developing chick inner ear epithelium.ResultsFour different models that interlaced cellular patterning mechanisms in a variety of ways were examined and their output compared to the mosaic of sensory and supporting cells that develops in the chick inner ear sensory epithelium. The results show that: 1) no single patterning mechanism can create a 2-dimensional mosaic pattern of the regularity seen in the chick inner ear; 2) cell death was essential to generate the most regular mosaics, even through extensive cell death has not been reported for the developing basilar papilla; 3) a model that includes an iterative loop of lateral inhibition, programmed cell death and cell rearrangements driven by differential adhesion created mosaics of primary and secondary cells that are more regular than the basilar papilla; 4) this same model was much more robust to changes in homo- and heterotypic cell-cell adhesive differences than models that considered either fewer patterning mechanisms or single rather than iterative use of each mechanism.ConclusionPatterning the embryo requires collaboration between multiple mechanisms that operate iteratively. Interlacing these mechanisms into feedback loops not only refines the output patterns, but also increases the robustness of patterning to varying initial cell states.

A small mobile robot for security and inspection operations

Abstract The omni-directional inspection system (ODIS) is a small, man-portable mobile robotic system that can be used for autonomous or semi-autonomous inspection under vehicles in a parking area. Customers for such a system include military police (MP) and other law enforcement and security entities. The robot features three “smart wheels” in which both the speed and direction of the wheel can be independently controlled and a vehicle electronic capability that includes multiple processors and a sensor array with a laser, sonar and IR sensors, and a video camera. ODIS employs a novel parameterized command language for intelligent behavior generation. A key feature of the ODIS control system is the use of an object recognition system that fits models to sensor data. These models are then used as input parameters to the motion and behavior control commands.

Biological development of cell patterns: characterizing the space of cell chemistry genetic regulatory networks

Genetic regulatory networks (GRNs) control gene expression and are responsible for establishing the regular cellular patterns that constitute an organism. This paper introduces a model of biological development that generates cellular patterns via chemical interactions. GRNs for protein expression are generated and evaluated for their effectiveness in constructing 2D patterns of cells such as borders, patches, and mosaics. Three types of searches were performed: (a) a Monte Carlo search of the GRN space using a utility function based on spatial interestingness; (b) a hill climbing search to identify GRNs that solve specific pattern problems; (c) a search for combinatorial codes that solve difficult target patterns by running multiple disjoint GRNs in parallel. We show that simple biologically realistic GRNs can construct many complex cellular patterns. Our model provides an avenue to explore the evolution of complex GRNs that drive development.

Discovering novel cancer therapies: A computational modeling and search approach

Solid tumors must recruit new blood vessels for growth and maintenance. Discovering drugs that block this tumor-induced development of new blood vessels (angiogenesis) is an important approach in cancer treatment. However, the complexity of angiogenesis and the difficulty in implementing and evaluating rationally-designed treatments prevent the discovery of effective new therapies. This paper presents a massively parallel computational search-based approach for the discovery of novel potential cancer treatments using a high fidelity simulation of angiogenesis. Discovering new therapies is viewed as multi-objective combinatorial optimization over two competing objectives: minimizing the cost of developing the intervention while minimizing the oxygen provided to the cancer tumor by angiogenesis. Results show the effectiveness of the search process in finding interventions that are currently in use and more interestingly, discovering some new approaches that are counter intuitive yet effective.