Adaptive and Energy-Efficient Architectures for Machine Learning: Challenges, Opportunities, and Research Roadmap

Shafique, Muhammad; Hafiz, Rehan; Javed, Muhammad Usama; Abbas, Sarmad A.; Sekanina, Lukáš; Vašíček, Zdeněk; Mrázek, Vojtěch

doi:10.1109/isvlsi.2017.124

Cited by 44 publications

(16 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the synthesis of approximate circuits and programs Vašíček and Sekanina opened up an area of research which is rich with research opportunities [89,116]. In many application areas (e.g.…”

Section: Applicationsmentioning

confidence: 99%

Cartesian genetic programming: its status and future

Miller

2019

Genet Program Evolvable Mach

View full text Add to dashboard Cite

Cartesian genetic programming, a well-established method of genetic programming, is approximately 20 years old. It represents solutions to computational problems as graphs. Its genetic encoding includes explicitly redundant genes which are wellknown to assist in effective evolutionary search. In this article, we review and compare many of the important aspects of the method and findings discussed since its inception. In the process, we make many suggestions for further work which could improve the efficiency of the CGP for solving computational problems.

show abstract

Section: Applicationsmentioning

confidence: 99%

Cartesian genetic programming: its status and future

Miller

2019

Genet Program Evolvable Mach

View full text Add to dashboard Cite

show abstract

“…In particular, millennia of evolution have enabled the brain to make sense of psychomotor situations with automaticity from comparatively few sensory inputs [ 46 , 58 ]. By contrast, electronic computers are less evolved; hence, the efforts towards brain-inspired computing paradigms [ 89 ]. Thus, we assume that the human brain is not millions of times slower than an electronic computer in psychomotor work [ 90 ].…”

Section: Heuristic Modelling Of Effects On Internal Action ( mentioning

confidence: 99%

Variational Principle of Least Psychomotor Action: Modelling Effects on Action from Disturbances in Psychomotor Work Involving Human, Cyborg, and Robot Workers

Fox

Kotelba

2019

Entropy

View full text Add to dashboard Cite

Optimal psychomotor work can be expressed in terms of the principle of least psychomotor action (PLPA). Modelling psychomotor action encompasses modelling workers, work, and interactions between them that involve different types of situated entropy. Modelling of psychomotor workers encompasses three types of workers: human, cyborg, and robot. The type of worker and the type of work interact to affect positioning actions, performing actions, and perfecting actions undertaken in psychomotor tasks. There are often disturbances in psychomotor work, for example due to weather conditions, which have a determining influence on what work can be undertaken with least psychomotor action by different types of workers. In this paper, findings are reported from a study focused on the modelling disturbances in psychomotor work. Five contributions are provided. First, a heuristic framework for modelling disturbances and their effects is provided. In addition to PLPA and situated entropy, this framework encompasses Markov processes, the theory of perturbations, and calculus of variations. Second, formulae and ratios are provided for heuristic modelling of effects on internal action (S int ) from disturbances to psychomotor work. Third, formulae and ratios are provided for heuristic modelling of effects on external action (S e ). Fourth, examples are provided of heuristic modelling of disturbances in psychomotor work. Fifth, formulae and examples show how task complexity can be modelled heuristically in terms of microstates across the cyber domain and the physical domain of cyber-physical systems. Overall, the study reported in this paper addresses variational aspects of PLPA. Entropy 2019, 21, 543 2 of 26examples are provided of heuristic modelling of disturbances in psychomotor work. Fifth, examples show how task complexity can be modelled heuristically in terms of microstates across the cyber domain and the physical domain of cyber-physical systems. Together, these contributions enable heuristic modelling of effects from disturbances on interactions between diverse psychomotor work and workers: rather than on autonomous systems [1][2][3][4].The study reported here builds upon research reported in two previous papers in Entropy. The first paper [5] provided an explanation of how resources for physical production work, such as work instructions, product components, and workstations, can be carriers of situated information, and also carriers of various types of situated entropy. The second paper [6] expanded upon this [5] by generalizing from examples to the three categories of work setting, work composition, and work uncertainty, and to three aspects of worker action: positioning, performing, and perfecting. In addition, details were provided about the state-of-the-art for psychomotor capabilities of human, cyborgs, and robot workers. Moreover, the principle of least psychomotor action (PLPA) was introduced as follows: the preferred combination of worker types is that which can carry out psychomotor work with the least internal ...

show abstract

“…The fractional length is then selected as a subtraction of the integer length from the bit width of the variable. Shafique et al [40] provide a deep analysis of hardware-oriented quantization aimed to improve accelerators energy efficiency. To do so, they show an Evolutionary Circuit Approximation to find a suitable solution to the quantization problem.…”

Section: Related Workmentioning

confidence: 99%

Quenn

Prado

Denna

Benini

et al. 2018

Proceedings of the 15th ACM International Conference on Computing Frontiers

View full text Add to dashboard Cite

Deep Learning is moving to edge devices, ushering in a new age of distributed Artificial Intelligence (AI). The high demand of computational resources required by deep neural networks may be alleviated by approximate computing techniques, and most notably reduced-precision arithmetic with coarsely quantized numerical representations. In this context, Bonseyes comes in as an initiative to enable stakeholders to bring AI to low-power and autonomous environments such as: Automotive, Medical Healthcare and Consumer Electronics. To achieve this, we introduce LPDNN, a framework for optimized deployment of Deep Neural Networks on heterogeneous embedded devices. In this work, we detail the quantization engine that is integrated in LPDNN. The engine depends on a fine-grained workflow which enables a Neural Network Design Exploration and a sensitivity analysis of each layer for quantization. We demonstrate the engine with a case study on Alexnet and VGG16 for three different techniques for direct quantization: standard fixed-point, dynamic fixed-point and k-means clustering, and demonstrate the potential of the latter. We argue that using a Gaussian quantizer with k-means clustering can achieve better performance than linear quantizers. Without retraining, we achieve over 55.64% saving for weights' storage and 69.17% for run-time memory accesses with less than 1% drop in top5 accuracy in Imagenet.

show abstract

Adaptive and Energy-Efficient Architectures for Machine Learning: Challenges, Opportunities, and Research Roadmap

Cited by 44 publications

References 25 publications

Cartesian genetic programming: its status and future

Cartesian genetic programming: its status and future

Variational Principle of Least Psychomotor Action: Modelling Effects on Action from Disturbances in Psychomotor Work Involving Human, Cyborg, and Robot Workers

Quenn

Contact Info

Product

Resources

About