Tikhonov or Lasso Regularization: Which Is Better and When

Wang, Fei; Liu, Wei; Chawla, Sanjay

doi:10.1109/ictai.2013.122

Cited by 6 publications

(2 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The key insight about robust regression as defined in [8] can be derived from considering the one-dimensional case [10], Which generalizes to where Z ∈ μ is the worst case disturbance of noise and μ has…”

Section: A Lasso and Robust Regressionmentioning

confidence: 99%

On Sparse Feature Attacks in Adversarial Learning

Wang

Liu

Chawla

2014

2014 IEEE International Conference on Data Mining

Self Cite

View full text Add to dashboard Cite

Adversarial learning is the study of machine learning techniques deployed in non-benign environments. Example applications include classifications for detecting spam email, network intrusion detection and credit card scoring. In fact as the gamut of application domains of machine learning grows, the possibility and opportunity for adversarial behavior will only increase.Till now, the standard assumption about modeling adversarial behavior has been to empower an adversary to change all features of the classifiers at will. The adversary pays a cost proportional to the size of "attack". We refer to this form of adversarial behavior as a dense feature attack.However, the aim of an adversary is not just to subvert a classifier but carry out data transformation in a way such that spam continues to appear like spam to the user as much as possible. We demonstrate that an adversary achieves this objective by carrying out a sparse feature attack. We design an algorithm to show how a classifier should be designed to be robust against sparse adversarial attacks. Our main insight is that sparse feature attacks are best defended by designing classifiers which use 1 regularizers.

show abstract

Section: A Lasso and Robust Regressionmentioning

confidence: 99%

On Sparse Feature Attacks in Adversarial Learning

Wang

Liu

Chawla

2014

2014 IEEE International Conference on Data Mining

Self Cite

View full text Add to dashboard Cite

show abstract

“…Hence, an exhaustive search for the optimal regularization factor is not required. Note also that other regularization techniques can be envisioned such as the least absolute shrinkage and selection operator (LASSO) technique [22]. However, the latter, unavailable in a closed-form solution, is usually found using optimization methods such as quadratic programming or convex optimization.…”

Section: B Reduction Of Pilot Subcarriersmentioning

confidence: 99%

ML-Type EM-Based Estimation of Fast Time-Varying Frequency-Selective Channels Over SIMO OFDM Transmissions

et al. 2019

View full text Add to dashboard Cite

This paper investigates the problem of fast time-varying frequency-selective (i.e., multipath) channel estimation over single-input multiple-output orthogonal frequency-division multiplexing (SIMO OFDM)-type transmissions. We do so by tracking the variations of each complex gain coefficient using a polynomial-in-time expansion. To that end, we derive the log-likelihood function (LLF) both in the dataaided (DA) and non-data-aided (NDA) cases. The DA maximum likelihood (ML) estimates over fast SIMO OFDM channels are derived here for the first time in closed-form expressions and hereby shown to be limited to applying over each receive antenna the DA least squares (LS) estimator tailored in [1] to fast SISO OFDM channels. This DA ML is used to initialize periodically, over a relatively large number of data blocks (i.e., with further reduced and relatively close-to-negligible pilot overhead compared to DA ML), a new expectation maximization (EM) ML-type solution we developed here in the NDA case to iteratively maximize the LLF. We also introduce an alternative regularized DA ML (RDM) initialization solution no longer requesting -in contrast to DA ML -more per-carrier pilot frames than the number of paths to further reduce overhead without incurring significant performance losses. Simulation results show that the proposed hybrid ML-EM estimator (i.e., combines all new NDA ML-EM and DA ML or RDM versions) converges within few iterations, thereby providing very accurate estimates of all multipath channel gains. Most importantly, this increased estimation accuracy translates into very significant BER and link-level percarrier throughput gains over the best representative benchmark solution available so far for the problem at hand, the SISO DA LS technique in [1] with its new generalization here to SIMO systems. INDEX TERMS Channel estimation, time-varying channel (TVC), OFDM, multi-carrier, single-input multiple-output (SIMO), single-input single-output (SISO), maximum likelihood (ML), expectation maximization (EM), least squares (LS), DA (data-aided), NDA (non-data-aided), regularized DA ML (RDM), maximum a posteriori (MAP), inter-carrier interference (ICI) cancellation (ICIC).

show abstract

Adaptive echo state network based-channel prediction algorithm for the internet of things based on the IEEE 802.11ah standard

Sui

Gao

2022

Telecommun Syst

View full text Add to dashboard Cite

Tikhonov or Lasso Regularization: Which Is Better and When

Cited by 6 publications

References 14 publications

On Sparse Feature Attacks in Adversarial Learning

On Sparse Feature Attacks in Adversarial Learning

ML-Type EM-Based Estimation of Fast Time-Varying Frequency-Selective Channels Over SIMO OFDM Transmissions

Adaptive echo state network based-channel prediction algorithm for the internet of things based on the IEEE 802.11ah standard

Contact Info

Product

Resources

About