WoC-Bots: Swarms of Biologically Inspired Prediction Agents
Dissertation
Author: Sean Grimes
Available Here
April 2023
This dissertation presents Wisdom-of-Crowds-Bots (WoC-Bots), biologically-inspired, simple, and modular agents which work together in a multi-agent environment to collectively make binary predictions. Building on the theoretical underpinnings of Wisdom of Crowds, WoC-Bots represent a knowledge-diverse crowd where each agent is trained on a subset of available information. A honeybee-derived swarm aggregation mechanism was developed to elicit a collective prediction with an associated confidence score. Due to the multi-agent architecture, WoC-Bots can be distributed across multiple compute nodes, reducing training and inference time. Importantly, this architecture demonstrates significant key advantages over traditional classification methods while maintaining comparable predictive performance. Specifically, new and previously unknown input features can be included in an existing classification problem without retraining existing agents. New input features, combined with existing features, are encapsulated into newly generated agents before agents are injected into an existing classification task. Further development led to a ``meta-swarm'', where an external prediction is used as the core belief of an agent, replacing a simple multi-layer perceptron network. The external prediction requires zero knowledge of source data, maintaining the localization and privacy of the data used to generate the prediction, enabling collaboration between institutions unable to share their data externally.
A Multi-Agent Approach to Binary Classification Using Swarm Intelligence
Future Internet, Special Issue on Modern Trends in Multi-agent Systems
Authors: Sean Grimes and David E. Breen
doi: 0.3390/fi15010036
Available Here
January 2023
Wisdom-of-Crowds-Bots (WoC-Bots) are simple, modular agents working together in a multi-agent environment to collectively make binary predictions. The agents represent a knowledge-diverse crowd, with each agent trained on a subset of available information. A honey-bee-derived swarm aggregation mechanism is used to elicit a collective prediction with an associated confidence value from the agents. Due to their multi-agent design, WoC-Bots can be distributed across multiple hardware nodes, include new features without re-training existing agents, and the aggregation mechanism can be used to incorporate predictions from other sources, thus improving overall predictive accuracy of the system. In addition to these advantages, we demonstrate that WoC-Bots are competitive with other top classification methods on three datasets and apply our system to a real-world sports betting problem, producing a consistent return on investment from 1 January 2021 through 15 November 2022 on most major sports.
An agent-based approach to predicting lymph node metastasis status in breast cancer
IEEE International Conference on Bioinformatics and Biomedicine
Authors: Sean Grimes, Mark D. Zarella, Fernando U. Garcia, David E. Breen
doi: 10.1109/BIBM52615.2021.9669624
Available Here
December 2021
We present a flexible, multi-agent approach to predictive classification problems which uses simple, modular agents that interact and share information socially in an arena with a variable number of participants. Opinion aggregation is accomplished using a honey-bee-derived optimization algorithm that improves accuracy and reduces variance compared with existing weighted and unweighted voter mechanisms. Confidence metrics may be derived from the agent interactions. We apply our system to a data set of 483 de-identified breast cancer patients to predict node-positive or node-negative disease with over 78.5% accuracy in general. When eliminating low-confidence predictions, which leaves 79.5% of patients, classification accuracy improves to 84.5%.
WoC-Bots: An Agent-Based Approach to Decision-Making
Applied Sciences
Authors: Sean Grimes and David E. Breen
doi: 10.3390/app9214653
Available Here
October 2019
We present a flexible, robust approach to predictive decision-making using simple, modular agents (WoC-Bots) that interact with each other socially and share information about the features they are trained on. Our agents form a knowledge-diverse crowd, allowing us to use Wisdom of the Crowd (WoC) theories to aggregate their opinions and come to a collective conclusion. Compared to traditional multi-layer perceptron (MLP) networks, WoC-Bots can be trained more quickly, more easily incorporate new features, and make it easier to determine why the network gives the prediction that it does. We compare our predictive accuracy with MLP networks to show that WoC-Bots can attain similar results when predicting the box office success of Hollywood movies, while requiring significantly less training time.
Directing chemotaxis-based spatial self-organisation via biased, random initial conditions
International Journal of Parallel, Emergent and Distributed Systems
Available Here
June 2018
Authors: Sean Grimes, Linge Bai, Andrew W.E. McDonald, David E. Breen
Inspired by the chemotaxis interaction of living cells, we have developed an agent-based approach for self-organising shape formation. Since all our simulations begin with a different uniform random configuration and our agents move stochastically, it has been observed that the self-organisation process may form two or more stable final configurations. These differing configurations may be characterised via statistical moments of the agents' locations. In order to direct the agents to robustly form one specific configuration, we generate biased initial conditions whose statistical moments are related to moments of the desired configuration. With this approach, we are able to successfully direct the aggregating swarms to produce a desired macroscopic shape, starting from randomised initial conditions with controlled statistical properties.
Ortus: an Emotion-Driven Approach to (artificial) Biological Intelligence
The European Conference on Artificial Life
Available Here
September 2017
Authors: Andrew W.E. McDonald, Sean Grimes, David E. Breen
Ortus is a simple virtual organism that also serves as an initial framework for investigating and developing biologically based artificial intelligence. Born from a goal to create complex virtual intelligence and an initial attempt to model C. elegans, Ortus implements a number of mechanisms observed in organic nervous systems, and attempts to fill in unknowns based upon plausible biological implementations and psychological observations. Implemented mechanisms include excitatory and inhibitory chemical synapses, bidirectional gap junctions, and Hebbian learning with its Stentian extension. We present an initial experiment that showcases Ortus fundamental principles; specifically, a cyclic respiratory circuit, and emotionally driven associative learning with respect to an input stimulus. Finally, we discuss the implications and future directions for Ortus and similar systems.