Online learning: theory, algorithms and applications (למידה מקוונת.)

By Shai Shalev-Shwartz. 2007

364 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Smoothness, Low Noise and Fast Rates (2010, methodology): “The proof follows from Lemma 3.1 and Theorem 1 of [28], using U 1 = B 2 and U 2 = nL ∗ in the Theorem.”
Introduction to Online Convex Optimization (2016, methodology): “In the context of OCO, the term follow-the-regularized-leader was coined in (99; 96), and at roughly the same time an essentially identical algorithm was called “RFTL” in (2).”
Regret Analysis of Stochastic and Nonstochastic Multi-armed Bandit Problems (2012, neutral): “A somewhat similar class of algorithms was rediscovered in the online learning community, see [88, 101, 119, 151].”
Katyusha: the first direct acceleration of stochastic gradient methods (2016, background): “tion V x(y) rather than the kknorm; or symbolically, (y) (x) + hr (x);y xi+ ˙V x(y) : (For instance, this is satised if !(y) def= 1 ˙ (y).) This is known as the \generalized strong convexity" [50] and is necessary for any linear-convergence result in the SC setting. Of course, in the non-SC se”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, background): “Online or sequential learning [193, 18, 194] is related to active learning in the sense that in both cases the model is iteratively and continuously updated using new data.”
Tight Bounds for Approximate Carathéodory and Beyond (2015, unknown): “We refer to [14, 28] for a proof of the converse.”
Minimizing The Misclassification Error Rate Using a Surrogate Convex Loss (2012, methodology): “the hinge loss with its corresponding analysis by (Shalev-Shwartz, 2007), or for the exact minimizer (which corresponds to the SVM) of empirical hinge loss using results in (Srebro et al.”
OTL: A Framework of Online Transfer Learning (2010, background): “More extensive surveys for online learning can be found in (Shalev-Shwartz, 2007).”
Fast Rates for Regularized Objectives (2008, unknown): “‖w‖1, as long as ‖w‖1 ≤ Bw (see [2]), yielding: Corollary 3.”
On the Interaction between Norm and Dimensionality: Multiple Regimes in Learning (2010, methodology): “(2010), which is based on Theorem 1 of Shalev-Shwartz (2007). While this bound holds only for stochastic gradient descent, we strongly suspect that it also holds for the constrained estimator (the regularized empirical risk minimizer).”
Collective Stability in Structured Prediction: Generalization from One Example (2013, background): “This includes the previous example of (approximate) marginal inference, since the negative entropy, as well as many surrogates, are strongly convex (Wainwright, 2006; Shalev-Schwartz, 2007).”
Adaptive Smoothed Online Multi-Task Learning (2016, background): “Notice when α→ 1, the above reduces to the perceptron mistake bound [17].”
DP-ADMM: ADMM-Based Distributed Learning With Differential Privacy (2018, methodology): “We deﬁne: From the Lemma 14 of [42], we have the inequality: According to the Cauchy-Schwartz inequality, we can get: By dividing both sides of the above inequality by , we can get: As we assume that k∇ ℓ ( · ) k ≤ c 1 , then:”
One-Pass AUC Optimization (2013, background): “(s1,s2,··· ,sn) > 0. Therefore, it is evident that f(xi) > f(xj) if ξi > ξj, and this theorem follows as desired. 9 Gao, Jin, Zhuand Zhou 5.2 Proof of Theorem 2 This proof is motivated from (Shalev-Shwartz, 2007; Srebro et al., 2010). Recall Lt(w) = λ 2 |w|2 + Pt−1 i=1 I[yi 6= yt](1 −yt(xt ”
Stochastic methods for l1 regularized loss minimization (2009, background): “The coordinate descent approach for solving £1 regularized problems is not new (as we survey below in Section 1.1).”