QiNS23: Quantum-inspired
Neuromorphic Systems

Goals

Recently, neuromorphic hardware platforms like Loihi have demonstrated superior energy-efficiency in solving certain convex and combinatorial optimization problems while achieving similar quality solutions as the solutions that can be obtained using conventional digital hardware (CPUs and GPUs). These results in conjunction with recent advances in quantum-inspired classical computing methods have opened the possibility of achieving neuromorphic supremacy where certain computational tasks are energetically more suitable for neuromorphic architectures. What are those tasks and how can we demonstrate neuromorphic supremacy? is the main goal of the proposed topic area. The topic area participants will delve into the physics of optimization that governs the dynamics in neuromorphic systems and the participants will also explore synergies between different neurodynamical principles and formulations used in quantum-mechanical systems. The participants will explore how some of these quantum-inspired neuromorphic techniques can be mapped onto existing neuromorphic simulators/hardware and can be used for solving NP-hard problems. The participants will benchmark these approaches with the current state-of-the-art and articulate possible hardware improvements that might be required to achieve neuromorphic supremacy goals. Since this topic area will cover a wide range of subjects ranging from optimization, devices, circuits, and statistical/quantum physics, it will broaden participation in the neuromorphic engineering community.

Projects

All projects will focus on different types of NP-hard problems which will range from constraint satisfaction problems (CSP), graph partitioning and routing problems, and search and maximum likelihood decoding problems. The specific problem will be finalized based on participant's expertise, availability of data, and hardware resources. Possible topic area projects are listed below: 

Physics of neuromorphic optimizers and quantum-based optimizers: In this project the group will explore synergies between formulations that are currently used in quantum-based optimizers and neuromorphic architectures. For instance, the dynamics of both formulations could be explained using a system Hamiltonian, where short-term and long-term network dynamics encodes information in an energy-efficient manner and with a high-precision. 

Spiking/oscillatory dynamics and NP-hard optimization: In this project, the group will explore different mechanisms for mapping NP-hard optimization problems on spiking neural networks. This will involve different reduction techniques, search and randomization methods, sampling methods, and their relation to the network architecture and network spiking/synchronization activity.

NP-hard optimization using neuromorphic asynchronous Hardware: In this project the group will explore how temporal dynamics can be used to encode and solve NP-hard problems. The group will explore polychronous encoding which will include time-to-first spike and n-of-m encodings.

NP-hard optimization and benchmarking using Loihi/SpiNNaker: In this project the group will benchmark different optimization algorithms using the Loihi/SpiNNaker hardware and simulation software.

Scaling of neuromorphic optimizers: In this project the group will investigate how in-memory computing and neuromorphic routing techniques could be used to address scaling challenges when solving large NP-hard problems. These include issues related to problem partitioning, the precision of the solution, model mismatch, and convergence speeds. 

Devices and circuits for neuromorphic optimizers: In this project the group will investigate different computational primitives inherent in the device physics to accelerate the search during the optimization process.

Materials, Equipment, and Tutorials:

The following hardware platforms and software packages will be used by the topic area participants for demonstration and benchmarking:

• Growth transforms-based spiking neural network Python package – for investigating physics-based, continuous-time, neuromorphic optimization algorithms.

• FPGA-based hardware and software emulation platform – for mapping the algorithms onto different types of neuronal networks.

• NSF CISE Community Research Infrastructure as a cloud-based hardware emulation platform large-scale simulation and implementation of the neuromorphic optimizers.

• Open Cloud Testbed – for executing applications on mobile platforms.

• Caltech Science Exchange – What is Quantum Computing?

Relevant Literature:

CA Mead, Collective Electrodynamics, MIT Press,  2000.

JG Cramer, CA Mead, "Symmetry, transactions, and the mechanism of wave function collapse," Symmetry 12(8):1373, 2020.

G Cauwenberghs, "Reverse engineering the cognitive brain", Proc Nat Acad  Sci, 2013.

AJ Cressman, W Wattanapanitch, I Chuang, R Sarpeshkar, "Formulation and emulation of quantum-inspired dynamical systems with classical analog circuits," Neural Computation, MIT Press, 2022.

D Marković, J Grollier, "Quantum neuromorphic computing," Applied Physics Letters, 2020.

S Czischek, A Baumbach, S Billaudelle, B Cramer, L Kades, JM Pawlowski, MK Oberthaler, J Schemmel, MA Petrovici, T Gasenzer, M Gärttner, "Spiking neuromorphic chip learns entangled quantum states," SciPost Physics, 2022.

BR La Cour, CI Ostrove, GE Ott, MJ Starkey, GR Wilson, "Classical emulation of a quantum computer", Int. J. Quantum Information, World Scientific, 2016.

MJA Schuetz, JK Brubaker, HG Katzgraber, "Combinatorial optimization with physics-inspired graph neural networks," Nat Mach Intell 4, 367–377, 2022. 

M Davies et al., "Advancing Neuromorphic Computing With Loihi: A Survey of Results and Outlook," in Proceedings of the IEEE, vol. 109, no. 5, pp. 911-934, May 2021.

S Ghosh, K Nakajima, T Krisnanda, K Fujii, TCH Liew, "Quantum neuromorphic computing with reservoir computing networks," Advanced Quantum Technologies, 2021.

LK Grover, AM Sengupta, "Classical analog of quantum search" Phys Rev A 65, 032319, 2002.

F Arute et al, "Quantum Supremacy using a Programmable Superconducting Processor," Nature, 2019.

F Pan, K Chen, P Zhang, "Solving the Sampling Problem of the Sycamore Quantum Circuits," Physical Review Letters, 2022.