Keio University, Mitsubishi Chemical, and research partners devised a new quantum reservoir computing (QRC) model to predict the chaotic motions of a soft robot, a bendable machine with movements controlled through applied air pressure.
Researchers converted the robot’s input states for an IBM superconducting transmon device acting as a quantum reservoir. The teams ran experimental demonstrations of their QRC scheme with up to 120 qubits, consistently achieving higher accuracy and shorter execution time than comparable established methods.
This is meaningful. Reservoir computing effectively lets us model the complexities of time-dependent systems with little training overhead. Such systems figure centrally in the world’s most important research domains: synaptic sequence processing, systems biology, climatic forecasting, fluid dynamics. If we can develop better modalities, we can more confidently predict the behavior of chaotic and higher dimensional systems.
Reservoirs and Echo States
Reservoir computing refers to any computational framework in which inputs map to higher dimensional computational spaces through the dynamics of some stable system, called a “reservoir”. This reservoir can be any fixed system where input adjustments do not result in directly corresponding output changes.
Context reverberation networks were some of our first substantive examples of reservoir computing in action. The idea stems from Turing’s morphogenesis model. Essentially, echoes reverberate through a system, providing dynamic context for all future inputs. Deep Echo State Network models are direct descendants that continue to drive major advances in hierarchical processing techniques. These frameworks also enable intensive study of recursive connective webs within neural networks, in turn facilitating increasingly complex, responsive systems.
But real reservoirs abound in the physical systems and phenomena of the world around us. Sand on the surface of a cymbal. A bucket of water bouncing in the bed of a moving truck. These are physical reservoirs where complex dynamics transmute input perturbations into computationally useful states. Careful measurement allows us to extract meaningful calculations from such complex behavior.
Current reservoir computing research efforts span a wide range of industries and prospective use cases. Foci include novel machine learning techniques, time series analysis/prediction, robotics, and the design of sophisticated autonomous control systems. The unique computational properties of reservoir computing frameworks also show particular promise to propel advances in the field of wearable health monitor devices to preempt medical emergencies.
Well beyond the realm of wearable tech, data analysis is a major area of interest for the implementation of reservoir computing models. This is especially true where rapid response and spatial system efficiency are of paramount importance — like the expanding networks of smart appliances that pervade our everyday lives. Reservoir computing also features prominently in ongoing research devoted to biometric security scanning technologies, weather forecasting and network optimization problems.
New methods for meaningful problem solving
QRC is an experimental machine learning framework that could help reduce the massive training overhead associated with more widely used frameworks like neural networks, and unlock new methods for the study of complex systems in nature. And not in some far flung future. Even today’s noisy, error-prone quantum computers show promise to help address fundamental problems that have long plagued efforts to optimize temporal information processing.
At present, many new models marrying the disciplines of machine learning and quantum computation show promise as potential paths to meaningful problem solving. Processor architectures and system designs vary widely, including high-profile research efforts predicated on neutral atom arrays, trapped ion, superconducting transmon, and spin qubits.
Quantum computers are naturally well-suited to high-dimensional data processing, and may ultimately prove more computationally valuable than classical reservoirs. Forthcoming research will explore the potential of quantum computing on superconducting transmon qubit devices for hard nonlinear problems like financial risk modeling.
Check out the paper in its entirety at the link below.
Alex Lukin 