Spatio-Temporal Memory for Navigation in a Mushroom Body Model
425
inherently encodes short sequences, and we show the model output is consistent
with observed behaviour of ants when familiar segments of a route occur in a
novel sequence. Our model has not yet been tested in more complex environments, but the next step will be to verify its robustness and generalisability in
a natural environment.
The current model has a limited memory capacity (see experiment 1), but
some simple modifications could increase the capacity: tuning the parameters to
make KC firing sparser and thus slow the accumulation of inhibition; or simply
increase the number of KCs. The size of MB in our model (only 10,000 KCs)
is smaller than for insect navigators (e.g. the honeybee mushroom body has
about 368,000 KCs [8]). Also, our model does lifetime learning of all incoming
information, with no selection or forgetting. Learning in insect the MB is likely
to be more sophisticated and efficient in dealing with redundancies in the sensory
stream. This might also enhance robustness, as it seems likely that a repeated
run on the same route in realistic conditions may easily exceed the 2% noise
tolerance shown in Experiment 2. Introducing additional smoothing and motion
processing layers between the event-based camera output and the MB input is
also likely to be helpful. In the insect there are multiple neural layers of visual
processing. Modelling these and other properties of the visual pathway is one of
the obvious next steps for this work.
Another required step is to design a navigation strategy which can utilise
the output of this model to generate proper route following patterns. Our model
explains how an insect would know it is on route, but not how it would re-find the
route. The novelty index results in Experiment 4 suggested that getting closer
to the route should provide some drop in novelty, so an oscillation strategy as
suggested in [10] could be effective, and will be explored in future work.
References
1. Ardin, P., Peng, F., Mangan, M., Lagogiannis, K., Webb, B.: Using an insect mushroom body circuit to encode route memory in complex natural environments. PLOS
Comput. Biol. 12(2), 1–22 (2016). https://doi.org/10.1371/journal.pcbi.1004683
2. Baddeley, B., Graham, P., Philippides, A., Husbands, P.: Holistic visual encoding of
ant-like routes: Navigation without waypoints. Adapt. Behav. 19(1), 3–15 (2011).
https://doi.org/10.1177/1059712310395410
3. Blanchard, M., Rind, F.C., Verschure, P.F.: Collision avoidance using a model of
the locust LGMD neuron. Robot. Auton. Syst. 30(1–2), 17–38 (2000)
4. Cope, A.J., Sabo, C., Gurney, K., Vasilaki, E., Marshall, J.A.: A model for an
angular velocity-tuned motion detector accounting for deviations in the corridorcentering response of the bee. PLoS Comput. Biol. 12(5), e1004887 (2016)
5. Cope, A.J., Vasilaki, E., Minors, D., Sabo, C., Marshall, J.A., Barron, A.B.:
Abstract concept learning in a simple neural network inspired by the insect brain.
PLoS Comput. Biol. 14(9), e1006435 (2018)
6. Eichler, K., et al.: The complete connectome of a learning and memory centre in
an insect brain. Nature 548(7666), 175 (2017)
7. Graham, P., Mangan, M.: Insect navigation: do ants live in the now? J. Exp.
Biol. 218(6), 819–823 (2015). https://doi.org/10.1242/jeb.065409. https://jeb.
biologists.org/content/218/6/819
425
inherently encodes short sequences, and we show the model output is consistent
with observed behaviour of ants when familiar segments of a route occur in a
novel sequence. Our model has not yet been tested in more complex environments, but the next step will be to verify its robustness and generalisability in
a natural environment.
The current model has a limited memory capacity (see experiment 1), but
some simple modifications could increase the capacity: tuning the parameters to
make KC firing sparser and thus slow the accumulation of inhibition; or simply
increase the number of KCs. The size of MB in our model (only 10,000 KCs)
is smaller than for insect navigators (e.g. the honeybee mushroom body has
about 368,000 KCs [8]). Also, our model does lifetime learning of all incoming
information, with no selection or forgetting. Learning in insect the MB is likely
to be more sophisticated and efficient in dealing with redundancies in the sensory
stream. This might also enhance robustness, as it seems likely that a repeated
run on the same route in realistic conditions may easily exceed the 2% noise
tolerance shown in Experiment 2. Introducing additional smoothing and motion
processing layers between the event-based camera output and the MB input is
also likely to be helpful. In the insect there are multiple neural layers of visual
processing. Modelling these and other properties of the visual pathway is one of
the obvious next steps for this work.
Another required step is to design a navigation strategy which can utilise
the output of this model to generate proper route following patterns. Our model
explains how an insect would know it is on route, but not how it would re-find the
route. The novelty index results in Experiment 4 suggested that getting closer
to the route should provide some drop in novelty, so an oscillation strategy as
suggested in [10] could be effective, and will be explored in future work.
References
1. Ardin, P., Peng, F., Mangan, M., Lagogiannis, K., Webb, B.: Using an insect mushroom body circuit to encode route memory in complex natural environments. PLOS
Comput. Biol. 12(2), 1–22 (2016). https://doi.org/10.1371/journal.pcbi.1004683
2. Baddeley, B., Graham, P., Philippides, A., Husbands, P.: Holistic visual encoding of
ant-like routes: Navigation without waypoints. Adapt. Behav. 19(1), 3–15 (2011).
https://doi.org/10.1177/1059712310395410
3. Blanchard, M., Rind, F.C., Verschure, P.F.: Collision avoidance using a model of
the locust LGMD neuron. Robot. Auton. Syst. 30(1–2), 17–38 (2000)
4. Cope, A.J., Sabo, C., Gurney, K., Vasilaki, E., Marshall, J.A.: A model for an
angular velocity-tuned motion detector accounting for deviations in the corridorcentering response of the bee. PLoS Comput. Biol. 12(5), e1004887 (2016)
5. Cope, A.J., Vasilaki, E., Minors, D., Sabo, C., Marshall, J.A., Barron, A.B.:
Abstract concept learning in a simple neural network inspired by the insect brain.
PLoS Comput. Biol. 14(9), e1006435 (2018)
6. Eichler, K., et al.: The complete connectome of a learning and memory centre in
an insect brain. Nature 548(7666), 175 (2017)
7. Graham, P., Mangan, M.: Insect navigation: do ants live in the now? J. Exp.
Biol. 218(6), 819–823 (2015). https://doi.org/10.1242/jeb.065409. https://jeb.
biologists.org/content/218/6/819