Limits...
A Unified Framework for Reservoir Computing and Extreme Learning Machines based on a Single Time-delayed Neuron.

Ortín S, Soriano MC, Pesquera L, Brunner D, San-Martín D, Fischer I, Mirasso CR, Gutiérrez JM - Sci Rep (2015)

Bottom Line: The reservoir is built within the delay-line, employing a number of "virtual" neurons.One key advantage of this approach is that it can be implemented efficiently in hardware.We show that the reservoir computing implementation, in this case optoelectronic, is also capable to realize extreme learning machines, demonstrating the unified framework for both schemes in software as well as in hardware.

View Article: PubMed Central - PubMed

Affiliation: Instituto de Física de Cantabria, CSIC-Universidad de Cantabria, E-39005 Santander, Spain.

ABSTRACT
In this paper we present a unified framework for extreme learning machines and reservoir computing (echo state networks), which can be physically implemented using a single nonlinear neuron subject to delayed feedback. The reservoir is built within the delay-line, employing a number of "virtual" neurons. These virtual neurons receive random projections from the input layer containing the information to be processed. One key advantage of this approach is that it can be implemented efficiently in hardware. We show that the reservoir computing implementation, in this case optoelectronic, is also capable to realize extreme learning machines, demonstrating the unified framework for both schemes in software as well as in hardware.

No MeSH data available.


Experimental implementation of either an ELM or an ESN in photonic hardware.The nonlinear projection is provided by a Lithium-Niobate Mach-Zehnder modulator, modulating the intensity of a standard semiconductor laser-diode. Simply by using a fiber-switch, one can select the information injected into the modulator to be γI(t) for the case of an ELM-implementation, or to be βz(t − τ) + γI(t) when implementing an ESN.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4597340&req=5

f2: Experimental implementation of either an ELM or an ESN in photonic hardware.The nonlinear projection is provided by a Lithium-Niobate Mach-Zehnder modulator, modulating the intensity of a standard semiconductor laser-diode. Simply by using a fiber-switch, one can select the information injected into the modulator to be γI(t) for the case of an ELM-implementation, or to be βz(t − τ) + γI(t) when implementing an ESN.

Mentions: Our optoelectronic hardware-implementation is schematically illustrated in Fig. 2. An integrated telecommunication Mach-Zehnder modulator (MZM, LiNbO3) implements a nonlinear transformation, which modulates the emission of a standard telecommunication laser-diode (20 mW), emitting at 1550 nm. A long optical fiber (50.4 km) combined with an electronic feedback circuit implements the delayed feedback loop, with the feedback-delay being τ = 247.2 μs. The electronic circuit acts as a low pass filter, with a characteristic response time of T = 240 ns. The experimental set-up has a switch that controls if the recurrent feedback loop is open or closed. If the switch closes the feedback loop, the circuit allows to combine the input information γI(t) with the delay signal βz(t − τ) and the system works as an ESN. The signal is amplified to allow for sufficiently nonlinear operation. If the switch does not close the feedback loop then the nonlinear operation is only performed over the input information γI(t) and the implementation works as an ELM. Our experimental system provides direct access to key parameters, e.g. the nonlinearity gain κ and the offset phase of the MZM ϕ, enabling easy tunability of nonlinearity and dynamical behaviors. The product κβ is equivalent to the spectral radius in ESNs1214. The parameter κ is controlled via the laser diode power, while ϕ is controlled by the DC-bias input of the MZM. Further details about the experimental setup can be found in9 where a similar implementation was used as a reservoir computer. Much higher processing speed, reaching 10 GSamples/s, have already been achieved by a comparable system17.


A Unified Framework for Reservoir Computing and Extreme Learning Machines based on a Single Time-delayed Neuron.

Ortín S, Soriano MC, Pesquera L, Brunner D, San-Martín D, Fischer I, Mirasso CR, Gutiérrez JM - Sci Rep (2015)

Experimental implementation of either an ELM or an ESN in photonic hardware.The nonlinear projection is provided by a Lithium-Niobate Mach-Zehnder modulator, modulating the intensity of a standard semiconductor laser-diode. Simply by using a fiber-switch, one can select the information injected into the modulator to be γI(t) for the case of an ELM-implementation, or to be βz(t − τ) + γI(t) when implementing an ESN.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4597340&req=5

f2: Experimental implementation of either an ELM or an ESN in photonic hardware.The nonlinear projection is provided by a Lithium-Niobate Mach-Zehnder modulator, modulating the intensity of a standard semiconductor laser-diode. Simply by using a fiber-switch, one can select the information injected into the modulator to be γI(t) for the case of an ELM-implementation, or to be βz(t − τ) + γI(t) when implementing an ESN.
Mentions: Our optoelectronic hardware-implementation is schematically illustrated in Fig. 2. An integrated telecommunication Mach-Zehnder modulator (MZM, LiNbO3) implements a nonlinear transformation, which modulates the emission of a standard telecommunication laser-diode (20 mW), emitting at 1550 nm. A long optical fiber (50.4 km) combined with an electronic feedback circuit implements the delayed feedback loop, with the feedback-delay being τ = 247.2 μs. The electronic circuit acts as a low pass filter, with a characteristic response time of T = 240 ns. The experimental set-up has a switch that controls if the recurrent feedback loop is open or closed. If the switch closes the feedback loop, the circuit allows to combine the input information γI(t) with the delay signal βz(t − τ) and the system works as an ESN. The signal is amplified to allow for sufficiently nonlinear operation. If the switch does not close the feedback loop then the nonlinear operation is only performed over the input information γI(t) and the implementation works as an ELM. Our experimental system provides direct access to key parameters, e.g. the nonlinearity gain κ and the offset phase of the MZM ϕ, enabling easy tunability of nonlinearity and dynamical behaviors. The product κβ is equivalent to the spectral radius in ESNs1214. The parameter κ is controlled via the laser diode power, while ϕ is controlled by the DC-bias input of the MZM. Further details about the experimental setup can be found in9 where a similar implementation was used as a reservoir computer. Much higher processing speed, reaching 10 GSamples/s, have already been achieved by a comparable system17.

Bottom Line: The reservoir is built within the delay-line, employing a number of "virtual" neurons.One key advantage of this approach is that it can be implemented efficiently in hardware.We show that the reservoir computing implementation, in this case optoelectronic, is also capable to realize extreme learning machines, demonstrating the unified framework for both schemes in software as well as in hardware.

View Article: PubMed Central - PubMed

Affiliation: Instituto de Física de Cantabria, CSIC-Universidad de Cantabria, E-39005 Santander, Spain.

ABSTRACT
In this paper we present a unified framework for extreme learning machines and reservoir computing (echo state networks), which can be physically implemented using a single nonlinear neuron subject to delayed feedback. The reservoir is built within the delay-line, employing a number of "virtual" neurons. These virtual neurons receive random projections from the input layer containing the information to be processed. One key advantage of this approach is that it can be implemented efficiently in hardware. We show that the reservoir computing implementation, in this case optoelectronic, is also capable to realize extreme learning machines, demonstrating the unified framework for both schemes in software as well as in hardware.

No MeSH data available.