Approximate Support Recovery using Codes for Unsourced Multiple Access

Gkagkos, Michail; Pradhan, Asit Kumar; Amalladinne, Vamsi K.; Narayanan, Krishna R.; Chamberland, Jean-François; Georghiades, C.N.

doi:10.1109/isit45174.2021.9517995

Cited by 9 publications

(4 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since the time complexity of AMP scales linearly with the length of the sparse vector, a low-complexity alternative to AMP is desirable. Hence, we propose an algorithm, called PolarAir, that combines K-sparsification and CS, in a manner akin to [23] and [24], to train the ResNet20 [22] over the AWGN medium. The time complexity of the proposed scheme, and the scheme in [10], is O(K 3 + K 2 log N ) and O(M N ), respectively, where N is the length of the gradient, and K ≪ N is sparsity and M is the number of measurements.…”

Section: B Main Contributionsmentioning

confidence: 99%

“…In the present context, our goal is to minimize the length of the vector x w (t) that can be used to transmit the gradient, without losing in terms of the test accuracy. To achieve that, we apply a top-K sparsification in conjunction with a CS-based signaling scheme similar to [23,24]. We focus on an elementary channel model to highlight the main idea behind PolarAir.…”

Section: System Modelmentioning

confidence: 99%

“…Next, a CS scheme, similar to [24] and [23], is used to reduce the dimension of vector g K w (θ t ) from N to M t < N . For the time being, we treat the CS algorithm as a black box; the details of the adopted scheme are available in Section IV.…”

Section: A Workermentioning

confidence: 99%

“…Throughout this section, the dependence of t is omitted for notational convenience. Since our scheme borrows ideas from [24], we only outline the main functionality here. We want to focus on the behaviour of the algorithm in the FL problem, and not on the structure of the CS algorithm itself.…”

Section: Compressed Sensing Schemementioning

confidence: 99%

See 3 more Smart Citations

PolarAir: A Compressed Sensing Scheme for Over-the-Air Federated Learning

Gkagkos¹,

Narayanan²,

Jean-Francois³

et al. 2023

Preprint

View full text Add to dashboard Cite

We explore a scheme that enables the training of a deep neural network in a Federated Learning configuration over an additive white Gaussian noise channel. The goal is to create a low complexity, linear compression strategy, called PolarAir, that reduces the size of the gradient at the user side to lower the number of channel uses needed to transmit it. The suggested approach belongs to the family of compressed sensing techniques, yet it constructs the sensing matrix and the recovery procedure using multiple access techniques. Simulations show that it can reduce the number of channel uses by ∼30% when compared to conveying the gradient without compression. The main advantage of the proposed scheme over other schemes in the literature is its low time complexity. We also investigate the behavior of gradient updates and the performance of PolarAir throughout the training process to obtain insight on how best to construct this compression scheme based on compressed sensing.

show abstract

Section: B Main Contributionsmentioning

confidence: 99%

Section: System Modelmentioning

confidence: 99%