Pretraining Graph Neural Networks for Few-Shot Analog Circuit Modeling and Design

Hakhamaneshi, Kourosh; Nassar, Marcel; Abbeel, Pieter; Stojanović, Vladimir

doi:10.1109/tcad.2022.3217421

Cited by 15 publications

(14 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Examples include the integration of logic synthesis [22] [23] [24], reinforcement learning for field-programmable gate array performance [22] and transistor placement [16]. Notably, these examples do not necessarily consider the circuit and graph reversibility because the output consists of images of the component layout [16] and numerical values for the simulation [13]. A method similar to the one-hot embedding vector concatenates the real values of the circuit constants into a one-hot vector [25] [26] [27].…”

Section: A Previous Workmentioning

confidence: 99%

“…Unlike continuous values in images and natural language, graph networks have discrete values that allow for a degree of freedom in the graph transformation method. Conventionally, circuit components and wiring are defined as nodes, with edges connecting these nodes [13] [14]. Because the roles of the circuit components and wiring differ, a constraint necessitates alternating appearances of component and wiring nodes.…”

Section: Introductionmentioning

confidence: 99%

“…When a circuit component with multiple terminals is transformed into a single node, the terminal numbers are lost, resulting in a self-loop representing a short circuit between the terminals of a single-circuit component. One potential transformation involves creating a complete graph, where each terminal becomes a node interconnected by edges [13]. However, complete graphs introduce two additional challenges.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Circuit2Graph: Circuits With Graph Neural Networks

Yamakaji,

Shouno,

Fukushima

2024

IEEE Access

View full text Add to dashboard Cite

Circuit design requires trial and error in both prototyping and simulation owing to the high degrees of freedom and mutual interference between the components. In this study, we propose a novel approach to address this challenge by introducing a transformation method that converts circuits into graph networks. This transformation is achieved with no loss of information, and we evaluate its accuracy through the application of graph classification by utilizing graph neural networks. We assume that the information degradation can result from self-loops, multi-edges, and heterogeneous graphs. To mitigate these issues, we propose a method that effectively reduces their impact. The results of this study demonstrate the effectiveness of our proposed method, as it achieves an accuracy of 97.89%. This represents a significant improvement of 5.2% when compared with the conventional method. Notably, our proposed method is applicable to generalpurpose circuits. This makes it a valuable addition to the existing repertoire of circuit solution methods, alongside analytical and simulation approaches.

show abstract

Section: A Previous Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Circuit2Graph: Circuits With Graph Neural Networks

Yamakaji,

Shouno,

Fukushima

2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Feature concatenation is a numerical representation of circuit inputs and outputs that properly tuned by minimizing the loss function. Attempts has been made to include different circuit topologies and obtain predictions as in [50], where two circuit types were included in the study: the ladder circuits and two stage operational amplifier circuits, with 20k training data instances of resistor ladders with 2 to 10 branches with equal distribution weight. The model is based on DeepGEN architecture and was able to make predictions on ladder circuits with higher number of branches.…”

Section: Branchmentioning

confidence: 99%

Comprehensive Mapping of Continuous/Switching Circuits in CCM and DCM to Machine Learning Domain Using Homogeneous Graph Neural Networks

Khamis

Agamy

2023

IEEE Open J. Circuits Syst.

View full text Add to dashboard Cite

This paper proposes a method of transferring physical continuous and switching/converter circuits working in continuous conduction mode (CCM) and discontinuous conduction mode (DCM) to graph representation, independent of the connection or the number of circuit components, so that machine learning (ML) algorithms and applications can be easily applied. Such methodology is generalized and is applicable to circuits with any number of switches, components, sources and loads, and can be useful in applications such as artificial intelligence (AI) based circuit design automation, layout optimization, circuit synthesis and performance monitoring and control. The proposed circuit representation and feature extraction methodology is applied to seven types of continuous circuits, ranging from second to fourth order and it is also applied to three of the most common converters (Buck, Boost, and Buck-boost) operating in CCM or DCM. A classifier ML task can easily differentiate between circuit types as well as their mode of operation, showing classification accuracy of 97.37% in continuous circuits and 100% in switching circuits.

show abstract

“…Recently, various studies have focused on learning movements comprising multiple short-term movements [1]. A common method involves using hierarchical architectures [13], [14]. These typically consist of low-level policies that learn primitive behaviors and high-level policies that plan sequences of low-level policies.…”

Section: Related Work a Imitation Learningmentioning

confidence: 99%

Imitation Learning With Time-Varying Synergy for Compact Representation of Spatiotemporal Structures

Kutsuzawa

Hayashibe

2023

IEEE Access

View full text Add to dashboard Cite

Imitation learning is a promising approach for robots to learn complex motor skills. Recent techniques allow robots to learn long-term movements comprising multiple sub-behaviors. However, learning the temporal structures of movements from a demonstration is challenging, particularly when subbehaviors overlap and are not labeled in advance. This study applied time-varying synergies, which are representations of spatial and temporal structures in human behavior in neuroscience, to imitation learning. The proposed method extracts time-varying synergies from human demonstrations, with neural networks that learn their activation patterns. Because time-varying synergies can decompose demonstrations into linear combinations of primitives while allowing overlapping, neural networks can learn demonstrations efficiently. This would make the model compact and improve its generalization ability. The proposed method was evaluated with the task of cursive letter writing requiring overlapping sub-behaviors. Consequently, the proposed method allows a neural network to generate new movements with a higher success rate and fewer parameters than those without the proposed method. Moreover, the neural network worked robustly against control deviations and disturbances in an actual robot.

show abstract

Pretraining Graph Neural Networks for Few-Shot Analog Circuit Modeling and Design

Cited by 15 publications

References 19 publications

Circuit2Graph: Circuits With Graph Neural Networks

Circuit2Graph: Circuits With Graph Neural Networks

Comprehensive Mapping of Continuous/Switching Circuits in CCM and DCM to Machine Learning Domain Using Homogeneous Graph Neural Networks

Imitation Learning With Time-Varying Synergy for Compact Representation of Spatiotemporal Structures

Contact Info

Product

Resources

About