Dr Ben Evans
Bio-inspired approaches to machine intelligence
In our group, we are interested in how neural systems can self-organise to produce robust, adaptive and intelligent behaviour, which we study from both a biological and a computational perspective through modelling. We seek to enhance artificial neural networks, particularly models of vision, by drawing inspiration from their biological counterparts and reverse-engineering the solutions discovered through evolution. We are especially interested in how the inductive biases of biological and neural systems, such as information bottlenecks and genetically determined features, lead to better generalisation than “blank slate” approaches.
Beyond rate-coded approaches, I am also interested in how the unusual properties of information processing with spikes (action potentials) may be crucial to achieving such impressive perceptual and cognitive abilities from varied, noisy and unreliable neurons. I am happy to supervise projects in this bio-physically detailed temporal-coding paradigm as well as deep neural networks.
Potential co-supervisors: , , , , , , , .
Key references
- Evans B.D., Malhotra G., Bowers J.S. (2022) Biological convolutions improve DNN robustness to noise and generalisation. Neural Networks, 148:96–110.
- Tsvetkov C., Malhotra G., Evans B.D., Bowers J.S. (2022) The role of capacity constraints in Convolutional Neural Networks for learning random versus natural data. bioRxiv,
- Malhotra G., Evans B.D., Bowers J.S. (2020) Hiding a plane with a pixel: examining shape-bias in CNNs and the benefit of building in biological constraints. Vision Research, 174:57–68.
- Evans B.D., Stringer, S.M. (2012) Transformation-invariant visual representations in self-organizing spiking neural networks. Frontiers in computational neuroscience, 6(46).
Visit Ben's for a full list of publications.