Insect-inspired pattern recognition with low computing power
Lead Research Organisation:
NEWCASTLE UNIVERSITY
Department Name: Biosciences Institute
Abstract
Background: Despite having brains with less than a million neurons, bees achieve remarkable feats of visual cognition, including pattern recognition, visual search and number learning. Little research has however, investigated the strategies underlying these behaviours. Recent research, led by the industrial partners, has begun to show that bees solve these complex cognitive tasks using active vision strategies, where movement is combined with visual input to solve a task. By approaching stimuli using stereotypical flight patterns, bees extract the information that helps them solve the task without complex computational processing. This approach has important implications for the development of small-scale robots with low computation power and the industrial partner has been at the forefront of this field. This project will investigate active vision in bees, how it facilitates complex learning and how this strategy can be adapted for technological innovation.
Aims: 1) To characterize the importance of stereotypical and flexible flight movements for successful pattern learning 2) To investigate individual variation in active vision strategies 3) To discover active vision strategies in response to incomplete pattern information.
Methodology: We will use behavioural assays to test bees in pattern discrimination tasks, ranging from relative distinct patterns to complex classification tasks. Bee flight behaviour during learning and tests will be recorded using high speed videography. Their paths will then be reconstructed in 3D using machine -learning based approaches. These paths will first be used to evaluate how stereotypical or flexible movements relate to successful learning depending on the difficulty of the task (Years 1 and 2). To understand if diverse movement and visual strategies can enable problem solving, we will then look at individual strategies (Years 2 and 3). In the real world, bees and robots face incomplete information in noisy visual environments. We will therefore test active vision strategies when the patterns are themselves moving or differentially obscured during training (Year 3 and 4). This will reveal novel strategies that are better suited to complex real-world scenarios. In collaboration with the second supervisor at the industrial partner, mathematical modelling of the underlying mechanisms (Years 2-4) will produce models that can be implemented into technology.
Novelty: This project will advance our understanding of the fundamental rules of life, by revealing how diverse neural systems can solve complex visual problems using movement-based strategies, and providing us essential information for bio-inspired technological development. The project will this advance our understand of system biology and neuroscience while simultaneously leading to new technology.
The proposed research will develop new technology and model the complexity of organismal functioning. It thus fulfils the DTP remit of Systems Biology and Technological Innovation. It also investigates the feedback between sensory and motor biology, fitting into the Neuroscience theme. The focus on bumblebees falls in the BBSRC remit for research on wild species, including pollinators. The research focus also fits the BBSRC priority of advancing the frontiers of bioscience discovery. Examining the active vision mechanisms also works towards understanding the rules of life. Finally working with industrial partner achieves goals of transformative technologies, data-intensive research and engineering biology.
Aims: 1) To characterize the importance of stereotypical and flexible flight movements for successful pattern learning 2) To investigate individual variation in active vision strategies 3) To discover active vision strategies in response to incomplete pattern information.
Methodology: We will use behavioural assays to test bees in pattern discrimination tasks, ranging from relative distinct patterns to complex classification tasks. Bee flight behaviour during learning and tests will be recorded using high speed videography. Their paths will then be reconstructed in 3D using machine -learning based approaches. These paths will first be used to evaluate how stereotypical or flexible movements relate to successful learning depending on the difficulty of the task (Years 1 and 2). To understand if diverse movement and visual strategies can enable problem solving, we will then look at individual strategies (Years 2 and 3). In the real world, bees and robots face incomplete information in noisy visual environments. We will therefore test active vision strategies when the patterns are themselves moving or differentially obscured during training (Year 3 and 4). This will reveal novel strategies that are better suited to complex real-world scenarios. In collaboration with the second supervisor at the industrial partner, mathematical modelling of the underlying mechanisms (Years 2-4) will produce models that can be implemented into technology.
Novelty: This project will advance our understanding of the fundamental rules of life, by revealing how diverse neural systems can solve complex visual problems using movement-based strategies, and providing us essential information for bio-inspired technological development. The project will this advance our understand of system biology and neuroscience while simultaneously leading to new technology.
The proposed research will develop new technology and model the complexity of organismal functioning. It thus fulfils the DTP remit of Systems Biology and Technological Innovation. It also investigates the feedback between sensory and motor biology, fitting into the Neuroscience theme. The focus on bumblebees falls in the BBSRC remit for research on wild species, including pollinators. The research focus also fits the BBSRC priority of advancing the frontiers of bioscience discovery. Examining the active vision mechanisms also works towards understanding the rules of life. Finally working with industrial partner achieves goals of transformative technologies, data-intensive research and engineering biology.
People |
ORCID iD |