Michael Lawrence - Research

Coordination of Data betweeen Autonomous, Heterogeneous Sources

We address the problem of coordinating updates from a base data source B to an autonomous, contingent data source C. E.g. when a building design is updated, the contractor's cost estimate must be updated, too. We propose a data coordination system which allows C to coordinate its data without imposing significant requirements on B. We use declarative mappings between B and C and coordinate data in two stages: (1) View Differencing, which finds changes to an intermediate view of B based on its mapping to C, and (2) Update Translation, which translates the view differencing result into updates on C. We describe two approaches to view differencing based on established results in incremental view maintenance. We give a solution for update translation using methods from incomplete information, as well as an optimized variant which uses novel structures to minimize the amount of wasted effort. We experimentally evaluate our solutions to both stages and demonstrate their feasibility on real world problems.

Grid-Enabled OLAP

In this work we propose a grid-based OLAP application which distributes query computation across an enterprise grid. Our application follows a two-tiered process for answering queries based on sharing cached OLAP data between the users at the local grid site, and using grid scheduling approaches to execute the remaining parts of a query amongst a distributed set of OLAP servers. A new technique for extraction and aggregation of shared cached OLAP data is proposed, along with an efficient, aggregate-aware cache controller. An experimental evaluation of the proposed query processing and cooperative caching methods shows a significant reduction in query times compared to previously proposed methods.

(Research performed for masters thesis).

View Selection in OLAP Data Warehouses

In a data warehousing environment, aggregate views are often materialized in order to speed up aggregate queries of online analytical processing (OLAP). Due to the increasing size of data warehouses, it is often infeasible to materialize all views. View selection, the task of selecting a subset of views to materialize based on updates and expectations of the query load, is an important and challenging problem. In this research, we explore dynamic view selection in which the distribution of queries changes over time and the set of materialized views must be tuned by replacing some of the previously materialized views with new ones.

(Research performed for undergraduate honours thesis).

Modular Associator Networks

Biologically inspired neural networks which perform temporal sequence learning and generation are frequently based on hetero-associative memories. Recent work by Jensen and Lisman has suggested that a model which connects an auto-associator module to a hetero-associator module can perform this function. We modify this architecture in a simplified model which in contrast uses a pair of connected auto-associative networks with hetero-associatively trained synapses in one of the paths between them. We simulate both models, finding that accurate and robust recall of learned sequences can easily be performed with the modified model introduced here, strongly outperforming the previous architecture.