One of the most challenging aspects of creating an artificial neural network is the replication, or approximation, of the variety of subtle factors that lead to firing of neurons in the brain. It may be simple to replicate the least complex aspect of binary on/off functions in the human brain, but creating a sophisticated system that comes close to mirroring the many factors that decide whether or not the pulse is fired is quite different.
At MIT, researchers Chi-Sang Poon, Guy Rachmuth, Mark Bear, and Harel Shouval have published A biophysically-based neuromorphic model of spike rate- and timing-dependent plasticity in Proceedings of the National Academy of Sciences. This paper details research that has huge implications in the field of artificial neurons and the ‘electronic brain.’ Chi-Sang Poon, senior author of the paper, consented to answer a few questions about the project for Decoded Science.
Interview with Professor Chi-Sang Poon
Decoded Science: Professor Poon, what inspired this line of research?
Professor Poon: Carver Mead at CalTech first suggested 20+ years ago the idea of using transistors to mimic neuronal spiking. We refined and extended this technique to mimic some inner workings of the neuron and synapse beyond simple spiking.
Decoded Science: What do you consider to be the most important implications of your research?
Professor Poon: It has potential major applications on three fronts.
– For basic neuroscience research it can be used as a simulation tool for studying large-scale brain networks that cannot be easily done on digital computers.
– For clinical applications, the neural simulator can be used to emulate certain brain dysfunctions that may allow neurologists to keep track of patients’ pathologies and responsiveness to therapeutic treatments. It can also be used in conjunction with the next generation of brain-machine interfaces that include a silicon brain system component in the loop, or be part of future neuroprostheses to replace malfunctional brain parts.
– For engineering applications, the brain chip provides a novel approach for building a new generation of artificial intelligence devices that can emulate human brain functions, such as pattern generation, cognition, learning and memory, and decision making.
Decoded Science: It sounds like the ‘plasticity’ effect allows for more effective learning, particularly through repetition – is that the case?
Professor Poon: Yes, learning is through repetition. It either strengthens or weakens the synapse depending on the manner of repetition.
Artificial Synapse Research and Depression
It takes 400 transistors to simulate the activity of a single brain synapse, but this model may be beneficial in the study of long term depression (LTD) and long-term potentiation (LTP). According to MIT News, “endo-cannabinoids, similar in structure to marijuana, are produced in the brain and are involved in many functions, including appetite, pain sensation and memory.” Theories among neuroscientists state suggest that “endo-cannabinoids produced in the postsynaptic cell are released into the synapse, where they activate presynaptic endo-cannabinoid receptors. If NMDA receptors are active at the same time, LTD occurs.” According to Professor Poon,
“…our model supports the idea that endo-cannabinoids are
important for long-term depression (LTD) … This model unifies the classical “spike rate-dependent” form of LTP and LTD and the more recent “spike timing-dependent” form of LTP and LTD. It allows us to construct a realistic silicon synapse that reproduces both forms of plasticity, whereas previously one can only mimic them one at a time.”
Many thanks to Professor Poon for taking the time to answer questions about this research!
Rachmuth, G., Shouval, H., Bear, M., Poon, C. S. PNAS Plus: A biophysically-based neuromorphic model of spike rate- and timing-dependent plasticity. Proceedings of the National Academy of Sciences. (2011). Published ahead of print, November 16, 2011. Doi:10.1073/pnas.1106161108.
Trafton, A. MIT News Office. Mimicking the brain, in silicon. November 15, 2011. Accessed November 17, 2011.
Decoding Science. One article at a time.