Machine Learning Model Inspired by Insects’ Nervous Systems

Innovation: Academia

University of Cologne

2020

Image: Andriyko Podilnyk / Unsplash / CC BY SA - Creative Commons Attribution + ShareAlike

Learn

Living systems obtain information from their environment and use it to find mates and resources, and to escape threats. Some reactions to gathered information are instinctual or innate behaviors, not based on previous experiences. But other reactions are based on previous experiences, which means that, at some level, the living system is learning. Learning can involve a living system passing information to another living system, or a living system obtaining information and using it in a feedback loop. That information can include knowledge, experience, behavior, and skills. Learning takes place when information that is new to the recipient is used to modify existing behavior or skills. The making and use of tools by some species of birds and mammals is a learned behavior. If a chimpanzee uses a stick to successfully get ants out of an ant mound and continues to do so, it has learned that behavior. If another chimpanzee observes this technique and replicates it, it has learned from the first chimp.

Search this Function

Compute

Computation allows an organism to take in and process information to make decisions. Whether it be calculating how fast prey is moving or deciding where to build a nest, computation is an important part of an organism’s life.

Search this Function

Encode/Decode

Organisms are constantly taking in new information, processing it, and storing that information as memories. Sensory information and stimuli such as sounds, smell, time, and spatial organization are encoded along with our memories. Whether we’re learning a new skill or repeating a task we’ve done before, we can easily call upon and decode that information when we need it.

Search this Function

Transduce/Convert Signals

Sometimes a signal, such as sound or vibration, reaches a living system and needs to be converted into a different, usable form. This requires converting from one type of energy form to another. For example, organisms with eyes have photoreceptors at the back of the eyes that convert light into electrical pulses. These pulses travel to the brain and allows it to register color, shape, and detail.

Search this Function

Respond to Signals

To interact with its environment, a living system must not only sense a variety of signals, but also respond to them. To be energy- and material-efficient, those responses must be appropriate to the signal. This generally requires sensing thresholds to trigger an appropriate level of response (for example, hiding under a shrub versus running away to avoid a predator). Response strategies are tied to a specific signal and often have a response threshold, which determines how strong a signal must be to warrant expending energy to respond. One example is a plant that lives in arid regions in South Africa. Its seed capsules remain closed until rainfall triggers them to open to release the seeds. But the plant only responds to a second rainfall, thus protecting against releasing its seeds before there is enough sustained water for them to grow.

Search this Function

A network model from the University of Cologne integrates fast, parallel processing with recognition of experience to enable an artificial agent to learn more efficiently.

Benefits

Increased learning
Increased efficiency
Adaptable

Applications

Artificial intelligence
Robotics

UN Sustainable Development Goals Addressed

Goal 9: Industry Innovation & Infrastructure

The Challenge

Computers rely on a binary system, consisting of 0’s and 1’s, to operate. They follow a code to compute and complete tasks, and they do not have a dynamic memory. Because of this, computers only can perform the tasks they are made to do, and can’t learn new processes unless there is a change in programming. If a computer is faced with a decision, it can only follow its code, it cannot remember what happened during past computations before the code was changed.

Innovation Details

The researchers analyzed how fruit flies learn to find food in their environments and built a computational model that mimics the process. Normally, researchers train flies by presenting a specific scent with a reward and a second scent without a reward. The fly learns to quickly associate certain scents with rewards, and they can remember this information when searching for food in more complex environments in the future. The researchers looked at the computational processing involved in the process and applied it to a computational model that mimics the biological information processing of fruit flies. What distinguishes biological computation is that it uses fast and parallel processing with input from brief nerve pulses. In addition, it creates a distributed memory by constantly changing and modifying contacts between neurons throughout the learning process. This allows a machine learning model, such as AI or an autonomous system, to learn much more efficiently and to apply what it has learned in a changing environment, using only a small database of training samples.

Biological Model

Insects are able to cope with problems in their environments by changing their behavior based on knowledge from past experiences. Their nervous systems use fast, parallel processing with brief nerve impulses to form a distributed memory.

References

Journal article

A spiking neural program for sensorimotor control during foraging in flying insects

PNAS

embedly preview

AskNature