In this work, we provide a characterization of the feature-learning process in two-layer ReLU networks trained by gradient descent on the logistic loss following random initialization. We consider data with binary lab...
详细信息
In this work, we provide a characterization of the feature-learning process in two-layer ReLU networks trained by gradient descent on the logistic loss following random initialization. We consider data with binary labels that are generated by an XOR-like function of the input features. We permit a constant fraction of the training labels to be corrupted by an adversary. We show that, although linear classifiers are no better than random guessing for the distribution we consider, two-layer ReLU networks trained by gradient descent achieve generalization error close to the label noise rate. We develop a novel proof technique that shows that at initialization, the vast majority of neurons function as random features that are only weakly correlated with useful features, and the gradient descent dynamics 'amplify' these weak, random features to strong, useful features.
Chandran et al. (SIAM J. Comput.'14) formally introduced the cryptographic task of position verification, where they also showed that it cannot be achieved by classical protocols. In this work, we initiate the stu...
详细信息
Successive cancellation list (SCL) decoders of polar codes excel in practical performance but pose challenges for theoretical analysis. Existing works either limit their scope to erasure channels or address general ch...
详细信息
ISBN:
(数字)9798350382846
ISBN:
(纸本)9798350382853
Successive cancellation list (SCL) decoders of polar codes excel in practical performance but pose challenges for theoretical analysis. Existing works either limit their scope to erasure channels or address general channels without taking advantage of soft information. In this paper, we propose the successive cancellation sampling (SCS) decoder. SCS hires iid “agents” to sample codewords using posterior probabilities. This makes it fully parallel and amenable for some theoretical analysis. As an example, when comparing SCS with
$\boldsymbol{a}$
agents to any list decoder with list size
$\boldsymbol{\ell}$
, we can prove that the error probability of the former is at most
$\boldsymbol{\ell}/\boldsymbol{ae}$
more than that of the latter. In this paper, we also describe how to adjust the “temperature” of agents. Warmer agents are less likely to sample the same codewords and hence can further reduce error probability.
This article considers the geometric problem of finding the center of a sphere in vector space over finite fields, given samples of random points on the sphere. We propose a quantum algorithm based on continuous-time ...
详细信息
Junta testing for Boolean functions has sparked a long line of work over recent decades in theoretical computerscience, and recently has also been studied for unitary operators in quantumcomputing. Tolerant junta te...
详细信息
Continuous-time quantum walks (CTQWs) play a crucial role in quantumcomputing, especially for designing quantum algorithms. However, how to efficiently implement CTQWs is a challenging issue. In this paper, we study ...
详细信息
Gaussian boson sampling is a popular method for experimental demonstrations of quantum advantage, but many subtleties remain in fully understanding its theoretical underpinnings. An important component in the theoreti...
详细信息
Gaussian boson sampling is a popular method for experimental demonstrations of quantum advantage, but many subtleties remain in fully understanding its theoretical underpinnings. An important component in the theoretical arguments for approximate average-case hardness of sampling is anticoncentration, which is a second-moment property of the output probabilities. In Gaussian boson sampling these are given by hafnians of generalized circular orthogonal ensemble matrices. In a companion work by Ehrenberg et al. [Phys. Rev. Lett. 134, 140601 (2025)], we develop a graph-theoretic method to study these moments and use it to identify a transition in anticoncentration. In this work, we find a recursive expression for the second moment using these graph-theoretic techniques. While we have not been able to solve this recursion by hand, we are able to solve it numerically exactly, which we do up to Fock sector 2n=80. We further derive analytical results about the second moment. These results allow us to pinpoint the transition in anticoncentration and furthermore yield the expected linear cross-entropy benchmarking score for an ideal (error-free) device.
Exotic hadrons,beyond the conventional quark model,have been discovered over the past two *** these states can lead to a deeper understanding of the nonperturbative dynamics of the strong *** this review,we focus on t...
详细信息
Exotic hadrons,beyond the conventional quark model,have been discovered over the past two *** these states can lead to a deeper understanding of the nonperturbative dynamics of the strong *** this review,we focus on the production of exotic hadrons in pp,PP^(-),and nuclear *** observations of light and hypernuclei as prototypes of hadronic molecules in heavy-ion collisions are also briefly discussed.
quantum state preparation is a fundamental and significant subroutine in quantumcomputing. In this paper, we conduct a systematic investigation on the circuit size for sparse quantum state preparation. A quantum stat...
详细信息
For the blockchain storage system, this paper proposes an I/O optimization scheme based on non-volatile memory based on the storage performance requirements of the blockchain system application. This scheme adopts the...
详细信息
暂无评论