Fluid Analogies Research

The aim of this research project is to explore and model the general mechanisms that underlie high-level cognitive processes such as reasoning, problem-solving, analogy-making and creativity.

The term "Fluid Analogies" was coined by Professor Douglas Hofstadter at Indiana University, reflecting two main ideas about cognition: that it has fluid-like properties (emerging from subtle interactions between top-down and bottom-up pressures), and that analogy-making lies deep at the heart of cognition. The interplay of subtle pressures can be seen in even the simplest of cognitive tasks. For example, an occluded object on a desk may be perceived as a telephone, supplementing the impoverished bottom-up sensory information with top-down expectations of what it is likely to be (based on previous experience). Such an act of visual recognition can also be viewed as a form of analogy-making, as it requires equating non-identical perceptions (i.e. the current object with previous exemplars) at a high-level of abstraction . From this definition, a wide range of cognitive competencies can be viewed as "Fluid Analogies" (incorporating similar, if not identical processes), ranging from simple object recognition (equating previous experience with current perceptions, ignoring surface-level differences), through to complex feats of creativity and insight (in which high-level regularities extracted from the world can be used to generate new instances).

The work of the UQ Fluid Analogies team is divided into two main areas: the development of general computational architectures that support flexible cognition, and the implementation of learning into these systems. Please follow the links below to learn more about our research, and to explore the associated videos and demo applets.

Stream 1: Cognitive Architectures
One of the main aims of this research project is to explore general cognitive architectures, representations and processes that can support flexible intelligence. Although several paradigms have been proposed in the past (such as connectionism and the traditional symbolic approach), they are limited in that they generally target either symbolic or subsymbolic computation. In contrast, flexible real-world cognition requires a seamless integration between these levels of processing. A central component of this research project is to investigate the types of representations and processes that afford the integration between low and high levels of processing, and to implement these ideas in subsequent revisions of the Fluid Analogies Engine (Bolland, 2005).
Stream 2: Learning
Stream 2a: Supervised Learning and Self-Organisation Recent work in our group has focussed on modelling the human neocortex – a structure in the brain that plays a crucial role in memory, attention, analogy-making, reasoning, language and problem-solving. Our model can be trained through a mixture of supervised and unsupervised learning, and can be used for recognition, as well as temporal prediction (a central ability required for problem solving, analogy-making and reasoning).
Stream 2b: Developmental Learning Flexible real-world problem solving often requires sensitivity to subtle task and object related features. In developing artificially intelligent “thinking systems”, it is doubtful that such subsymbolic sensitivities can be hand-coded or learned through explicit tuition. Instead, learning appropriate grounded representations through self-generated interactions with and exploration of the world is an important (and perhaps necessary) characteristic of artificially intelligent embodied systems.
Stream 2c: Evolutionary Learning To effectively learn the knowledge that is required to function in the real-world requires both the appropriate learning algorithms, as well as the appropriate architecture and local connectivity patterns of neurons. In biological systems, such structures are specified in the genome, being generated through evolution. Likewise, simulated evolution may be useful in generating appropriate architectures that support effective learning in artificially intelligent systems. The final research stream of this project will aim to investigate this issue.