TY - JOUR
T1 - Active-learning-based reconstruction of circuit model
AU - Rozenfeld, Gal
AU - Kalech, Meir
AU - Rokach, Lior
N1 - Publisher Copyright:
© 2021, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
PY - 2022/3/1
Y1 - 2022/3/1
N2 - Reconstructing the circuit model presents a challenge for circuits with unknown functional specifications. The circuit is thought of as a black box that, given an input, produces an output. The model of the circuit, on the other hand, is unknown. Given a set of inputs and their corresponding outputs, the goal is then to recover the circuit specification while maximizing reconstruction accuracy. This process is computationally difficult, and it becomes even more difficult to solve when only a subset of inputs and outputs are provided, as is the case in many large and complex circuits. This issue is addressed in this paper: Reconstructing the model of a circuit from a set of components and observations describing its inputs and outputs. Previous work proposed a decision tree approach, but this approach only works when the entire set of inputs and outputs is available. Nonetheless, for most systems, this requirement is unrealistic. To address this challenge, we propose an active learning approach and applying orthogonal arrays of fractional factorial design to sample labeled data for learning the reconstructed circuit. Evaluation on 9 well known circuits shows the benefits of the proposed algorithms in terms of accuracy, run time and the reconstructed circuit model.
AB - Reconstructing the circuit model presents a challenge for circuits with unknown functional specifications. The circuit is thought of as a black box that, given an input, produces an output. The model of the circuit, on the other hand, is unknown. Given a set of inputs and their corresponding outputs, the goal is then to recover the circuit specification while maximizing reconstruction accuracy. This process is computationally difficult, and it becomes even more difficult to solve when only a subset of inputs and outputs are provided, as is the case in many large and complex circuits. This issue is addressed in this paper: Reconstructing the model of a circuit from a set of components and observations describing its inputs and outputs. Previous work proposed a decision tree approach, but this approach only works when the entire set of inputs and outputs is available. Nonetheless, for most systems, this requirement is unrealistic. To address this challenge, we propose an active learning approach and applying orthogonal arrays of fractional factorial design to sample labeled data for learning the reconstructed circuit. Evaluation on 9 well known circuits shows the benefits of the proposed algorithms in terms of accuracy, run time and the reconstructed circuit model.
KW - Active learning
KW - Decision trees
KW - Diagnosis
KW - Orthogonal arrays
UR - http://www.scopus.com/inward/record.url?scp=85111752481&partnerID=8YFLogxK
U2 - 10.1007/s10489-021-02700-z
DO - 10.1007/s10489-021-02700-z
M3 - Article
AN - SCOPUS:85111752481
VL - 52
SP - 5125
EP - 5143
JO - Applied Intelligence
JF - Applied Intelligence
SN - 0924-669X
IS - 5
ER -