Michael Smith

EAAI Journal 2024 Journal Article

Data engineering for predictive machine learning of stormwater infrastructure conditions

• Nicole Roberts
• Michael Smith
• Jingyi Qi

Culverts are stormwater structures that require ongoing condition assessment and proactive maintenance for optimal performance to reduce flooding, especially in urban areas. Managing these structures can be complex and costly regarding time, money, and personnel resources. Machine learning (ML) tools are a powerful means of predicting culvert conditions via learning from their existing data records. However, the data available on such subjects are usually low in quantity and of various qualities; thus, they need significant data processing for ML. Existing stormwater infrastructure condition prediction studies rarely detail their data processing methods. Hence, this work uses a comprehensive case study to illustrate essential data preparation techniques to address common data issues and enhance the performance of commonly used ML algorithms. The study shows methods for exploratory data analysis to understand the dataset, data wrangling methods for preprocessing data of various quality, and data engineering procedures for addressing insufficient data issues. After the data processing procedures are applied, F-1 scores are used to evaluate the ML models' performance. The random forest classifier model, one of the four ML models, performed best after applying the data wrangling and data engineering methods. Overall, this work provides transparency of methods and applications to encourage ML use in the water resources engineering field.

NeurIPS Conference 2019 Conference Paper

Multi-task Learning for Aggregated Data using Gaussian Processes

• Fariba Yousefi
• Michael Smith
• Mauricio Álvarez

Aggregated data is commonplace in areas such as epidemiology and demography. For example, census data for a population is usually given as averages defined over time periods or spatial resolutions (cities, regions or countries). In this paper, we present a novel multi-task learning model based on Gaussian processes for joint learning of variables that have been aggregated at different input scales. Our model represents each task as the linear combination of the realizations of latent processes that are integrated at a different scale per task. We are then able to compute the cross-covariance between the different tasks either analytically or numerically. We also allow each task to have a potentially different likelihood model and provide a variational lower bound that can be optimised in a stochastic fashion making our model suitable for larger datasets. We show examples of the model in a synthetic example, a fertility dataset and an air pollution prediction application.

AIJ Journal 2007 Journal Article

PickPocket: A computer billiards shark

• Michael Smith

Billiards is a game of both strategy and physical skill. To succeed, a player must be able to select strong shots, and then execute them accurately and consistently on the table. Several robotic billiards players have recently been developed. These systems address the task of executing shots on a physical table, but so far have incorporated little strategic reasoning. They require artificial intelligence to select the ‘best’ shot taking into account the accuracy of the robot, the noise inherent in the domain, the continuous nature of the search space, the difficulty of the shot, and the goal of maximizing the chances of winning. This article describes the program PickPocket, the winner of the simulated 8-ball tournaments at the 10th and 11th Computer Olympiad competitions. PickPocket is based on the traditional search framework, familiar from games such as chess, adapted to the continuous stochastic domain of billiards. Experimental results are presented exploring the properties of two search algorithms, Monte-Carlo search and Probabilistic search.

AAAI Conference 2006 Conference Paper

Running the Table: An AI for Computer Billiards

• Michael Smith

Billiards is a game of both strategy and physical skill. To succeed, a player must be able to select strong shots, and then execute them accurately and consistently. Several robotic billiards players have recently been developed. These systems address the task of executing shots on a physical table, but so far have incorporated little strategic reasoning. They require AI to select the ‘best’ shot taking into account the accuracy of the robotics, the noise inherent in the domain, the continuous nature of the search space, the difficulty of the shot, and the goal of maximizing the chances of winning. This paper develops and compares several approaches to establishing a strong AI for billiards. The resulting program, PickPocket, won the first international computer billiards competition.

AAMAS Conference 2005 Conference Paper

A Framework for Decomposing Reputation in MAS into Competence and Integrity

• Michael Smith
• Marie des Jardins

YNIMG Journal 2003 Journal Article

Assessing brain activity through spatial bayesian variable selection

• Michael Smith
• Benno Pütz
• Dorothee Auer
• Ludwig Fahrmeir

Statistical parametric mapping (SPM), relying on the general linear model and classical hypothesis testing, is a benchmark tool for assessing human brain activity using data from fMRI experiments. Friston et al. (Neuroimage 16 (2002a), 484) discuss some limitations of this frequentist approach and point out promising Bayesian perspectives. In particular, a Bayesian formulation allows explicit modeling and estimation of activation probabilities. In this study, we directly address this issue and develop a new regression based approach using spatial Bayesian variable selection. Our method has several advantages. First, spatial correlation is directly modeled for activation probabilities and indirectly for activation amplitudes. As a consequence, there is no need for spatial adjustment in a postprocessing step. Second, anatomical prior information, such as the distribution of grey matter or expert knowledge, can be included as part of the model. Third, the method has superior edge-preservation properties as well as being fast to compute. When applied to data from a simple visual experiment, the results demonstrate improved sensitivity for detecting activated cortical areas and for better preserving details of activated structures.