Replica Exchange MCMC Hardware With Automatic Temperature Selection and Parallel Trial

Dabiri, Keivan; Malekmohammadi, Mehrdad; Sheikholeslami, Ali; Tamura, Hirotaka

doi:10.1109/tpds.2020.2972359

Cited by 16 publications

(10 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…After the trial, the system state variables s i , H, and E need to be updated as shown in (7), (8), and (9) respectively, where W i, * and W * ,i represent row and column i of W respectively.…”

Section: Boltzmann Machinesmentioning

confidence: 99%

“…3b, allows for a systematic method of escaping from local minima, making PT a better choice for utilizing parallelism than simply running M disjoint replicas in parallel using SA as proven in [10,14]. In this paper, we implement a PT algorithm based on a modified version of Dabiri's work [7]. One drawback to PT algorithms such as the BM + PT system used in [7] is that their T max and T min must be manually tuned for each problem instance.…”

Section: Parallel Temperingmentioning

confidence: 99%

“…In this paper, we implement a PT algorithm based on a modified version of Dabiri's work [7]. One drawback to PT algorithms such as the BM + PT system used in [7] is that their T max and T min must be manually tuned for each problem instance. This requires considerable time and effort while having dramatic effects on the efficacy of the optimization process.…”

Section: Parallel Temperingmentioning

confidence: 99%

See 2 more Smart Citations

A Permutational Boltzmann Machine with Parallel Tempering for Solving Combinatorial Optimization Problems

Bagherbeik

Ashtari

Mousavi

et al. 2020

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Boltzmann Machines are recurrent neural networks that have been used extensively in combinatorial optimization due to their simplicity and ease of parallelization. This paper introduces the Permutational Boltzmann Machine, a neural network capable of solving permutation optimization problems. We implement this network in combination with a Parallel Tempering algorithm with varying degrees of parallelism ranging from a single-thread variant to a multi-threaded system using a 64core CPU with SIMD instructions. We benchmark the performance of this new system on Quadratic Assignment Problems, using some of the most difficult known instances, and show that our parallel system performs in excess of 100× faster than any known dedicated solver, including those implemented on CPU clusters, GPUs, and FPGAs.

show abstract

Section: Boltzmann Machinesmentioning

confidence: 99%

Section: Parallel Temperingmentioning

confidence: 99%

Section: Parallel Temperingmentioning

confidence: 99%

See 1 more Smart Citation

A Permutational Boltzmann Machine with Parallel Tempering for Solving Combinatorial Optimization Problems

Bagherbeik

Ashtari

Mousavi

et al. 2020

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

“…It has been shown that simulated annealing, when combined with the massive parallelism offered by certain hardware platforms, can lead to novel hardware architectures that provide significant speedup compared to the standard simulated annealing algorithm [10], [11].…”

Section: Introductionmentioning

confidence: 99%

“…Inspired by the prior work in [11] and the concept of Markov Chain Monte Carlo (MCMC) search, we propose and implement a new heterogeneous hardware platform, called the Hamiltonian Engine for Radiotherapy Optimization (HERO), which can be applied to complex optimization problems in radiation therapy, such as IMRT and VMAT. This platform addresses both issues of handling large numbers of variables and casting the problem in a format supported by the engine.…”

Section: Introductionmentioning

confidence: 99%

A Hamiltonian Engine for Radiotherapy Optimization

Malekmohammadi

Dabiri

Mathews

et al. 2020

2020 42nd Annual International Conference of the IEEE Engineering in Medicine &Amp; Biology Society (EMBC)

Self Cite

View full text Add to dashboard Cite

We apply a new hardware and software platform called the Hamiltonian Engine for Radiotherapy Optimization (HERO) to the problem of Intensity-Modulated Radiation Therapy (IMRT) treatment planning. HERO solves large generalform binary optimization problems by decomposing them into sub-problems and approximating them using a quadratic pseudo-boolean function. Optimizing the resulting function becomes a quadratic unconstrained binary optimization (QUBO) problem, which has been widely studied and has numerous applications in various fields. A Quantum Annealer (QA) approach has been previously investigated to solve QUBO problems, including IMRT optimization. However, the QA can only accommodate a small number of variables and requires several hours to obtain optimized plans. HERO acts as an optimizer for QUBO problems, which not only addresses these shortcomings but also relies solely on conventional hardware design while operating at room temperature. We evaluate HERO on seven prostate IMRT cases with clinical objectives, each using approximately 6000 beamlets. Our method was compared to the commercial treatment planning software, Eclipse, for both time-to-solution and plan quality. HERO solves most cases in about 30 seconds, with significantly lower objective function scores than Eclipse. The results indicate that HERO is promising for radiation therapy optimization problems. Additionally, HERO has the potential to be applied to Volumetric-Modulated Arc Therapy (VMAT) and other complex types of treatment planning.

show abstract