Well I have finally discovered the fundamental starting point for NASA’s interest in QUANTUM ANNEALING. I had not been going back far enough but here it is in 2012:
A Near-Term Quantum Computing Approach for Hard Computational Problems in Space Exploration
Vadim N. Smelyanskiy,∗ Eleanor G. Rieffel, and Sergey I. Knysh NASA Ames Research Center, Mail Stop 269-3, Moffett Field, CA 94035
Colin P. Williams
Jet Propulsion Laboratory, California Institute of Technology,Pasadena, CA 91109-8099
Mark W. Johnson, Murray C. Thom, William G. Macready
D-Wave Systems Inc., 100-4401 Still Creek Drive, Burnaby, BC, Canada V5C 6G9
Kristen L. Pudenz†
Ming Hsieh Department of Electrical Engineering,
Center for Quantum Information Science and Technology, and Information Sciences Institute,
University of Southern California, Los Angeles, CA 90089
Abstract
The future of Space Exploration is entwined with the future of artificial intelligence (AI) and machine learning. Autonomous rovers, unmanned spacecraft, and remote space habitats must all make intelligent decisions with little or no human guidance. The decision-making required of such NASA assets stretches machine intelligence to its limits. Currently, AI problems are tackled using a variety of heuristic approaches, and practitioners are constantly trying to find new and better techniques. To achieve a radical breakthrough in AI, radical new approaches are needed. Quantum computing is one such approach.
Many of the hard combinatorial problems in space exploration are instances of NP-complete or NP-hard problems. Neither traditional computers nor quantum computers are expected to be able to solve all instances of such problems efficiently. Many heuristic algorithms, such as simulated annealing, support vector machines, and SAT solvers, have been developed to solve or approximate solutions to practical instances of these problems. The efficacy of these approaches is generally determined by running them on benchmark sets of problem instances. Such empirical testing for quantum algorithms requires the availability of quantum hardware.
Quantum annealing machines, analog quantum computational devices, are designed to solve dis- crete combinatorial optimization problems using properties of quantum adiabatic evolution. We are now on the cusp of being able to run small-scale examples of these problems on actual quantum annealing hardware which will enable us to test empirically the performance of quantum annealing on these problems. For example, D-Wave builds quantum annealing machines based on supercon- ducting qubits. While at present noise and decoherence in quantum annealing devices cannot be easily controlled or corrected, these devices have been shown to display multi-spin tunneling, a distinct quantum phenomenon at the root of the quantum annealing process. In order to attack an optimization problem on these machines, the problem must be formulated in quadratic uncon- strained binary optimization form in which the cost function is strictly quadratic in bit assignments (in physics applications this form is often referred to as an Ising model). The above limitation is not fundamental: all NP-complete problems can be mapped to this form. However, an optimal mapping involving small or no overhead in terms of additional bits is of significant practical interest because of the limited size of early quantum annealing machines.
In this article, we discuss a sampling of the hardest artificial intelligence problems in space explo- ration in the context in which they emerge. We show how to map them onto equivalent Ising models
that then can be attacked using quantum annealing. We review existing quantum annealing results on supervised learning algorithms for classification and clustering and discuss their application to planetary feature identification and satellite image analysis. We present quantum annealing algo- rithms for unsupervised learning for clustering and discuss its application to anomaly detection in space systems. We introduce quantum annealing algorithms for data fusion and image matching for remote sensing applications. We overview planning problems for space exploration missions applica- tions and introduce algorithms for planning problems using quantum annealing of Ising models. We describe algorithms for diagnostics and recovery as well as their applications to NASA deep space missions and show how a fault tree analysis problem can be mapped onto an Ising model and solved with quantum annealing. We discuss combinatorial optimization algorithms for task assignment in the context of autonomous unmanned exploration that take into account constraints due to physical limitations of the vehicles. We show how these algorithms can be presented in the framework of Ising model optimization with application to quantum annealing. Finally, we discuss ways to circumvent the need to map practical optimization problems onto the Ising model. We demonstrate how this can be done in principle using a “blackbox" approach based on ideas from probabilistic computing. In particular, we provide initial results on Monte Carlo sampling for solving non-Ising problems.
In this article, we describe the architecture, duty cycle times and energy consumption of the D- Wave One quantum annealing machine. We report on benchmark scalability studies of D-Wave One run times and compare to state of the art classical algorithms for solving Ising optimization problems on a uniform random ensemble of problems. Results on problems in the range of up to 96 qubits show improved scaling for median core quantum annealing time compared with simulated annealing and iterative tabu search, though how it will scale as the number of qubits increases remains an open question. We also review existing results of D-Wave One benchmarking studies for solving binary classification problems with a quantum boosting algorithm. The error rates on synthetic data sets show that quantum boosting algorithm consistently outperforms the AdaBoost classical machine learning algorithm. We review quantum algorithms for structured learning for multi-label classification and describe how the problem of finding an optimal labeling can be mapped onto quantum annealing with Ising models, and then introduce a hybrid classical/quantum approach for learning the weights. We review results of D-Wave One benchmarking studies for learning structured labels on four different data sets. The first data set is Scene, a standard image benchmark set. The second data set, the RCV1 subset of the Reuters corpus of labeled news stories, has a significantly larger number of labels, and more complex relationships between the labels. The other two are synthetic data sets generated using MAX-3 SAT problem instances. On all four data sets, quantum annealing was compared with an independent Support Vector Machine (SVM) approach with linear kernel and exhibited a better performance”
So now we have the why?
We have the yes SNN can process with Quantum Annealing algorithms.
We have NASA experimenting with SNN and Quantum Annealing in 2021.
We have random statements from Professor Iliadis at the Democratis University of Thrace and Rob Telson that AKIDA is being used for autonomous space flight.
Who will find the final piece of this puzzle???
My opinion only DYOR
FF
AKIDA BALLISTA
PS: Just imagine how a system that can autonomously navigate in space unconnected without satellite navigation or any form of geolocation could revolutionise autonomous vehicle/robots of every description in the air, on land, under land, on and under the sea.
