Abstract
We investigate DC (Difference of Convex functions) programming and DCA (DC Algorithm) for a class of online learning techniques, namely prediction with expert advice, where the learner’s prediction is made based on the weighted average of experts’ predictions. The problem of predicting the experts’ weights is formulated as a DC program for which an online version of DCA is investigated. The two so-called approximate/complete variants of online DCA based schemes are designed, and their regrets are proved to be logarithmic/sublinear. The four proposed algorithms for online prediction with expert advice are furthermore applied to online binary classification. Experimental results tested on various benchmark datasets showed their performance and their superiority over three standard online prediction with expert advice algorithms—the well-known weighted majority algorithm and two online convex optimization algorithms.
Similar content being viewed by others
References
Alexander L, Das SR, Ives Z, Jagadish H, Monteleoni C (2017) Research challenges in financial data modeling and analysis. Big Data 5(3):177–188
Angluin D (1988) Queries and concept learning. Mach Learn 2(4):319–342
Azoury K, Warmuth MK (2001) Relative loss bounds for on-line density estimation with the exponential family of distributions. Mach Learn 43(3):211–246
Barzdin JM, Freivald RV (1972) On the prediction of general recursive functions. Sov Math Doklady 13:1224–1228
Cesa-Bianchi N (1999) Analysis of two gradient-based algorithms for on-line regression. J Comput Syst Sci 59(3):392–411
Cesa-Bianchi N, Freund Y, Haussler D, Helmbold DP, Schapire RE, Warmuth MK (1997) How to use expert advice. J ACM 44(3):427–485
Cesa-Bianchi N, Lugosi G (2003) Potential-based algorithms in on-line prediction and game theory. Mach Learn 51(3):239–261
Cesa-Bianchi N, Lugosi G (2006) Prediction, learning, and games. Cambridge University Press, New York
Cesa-Bianchi N, Mansour Y, Stoltz G (2007) Improved second-order bounds for prediction with expert advice. Mach Learn 66(2):321–352
Chung TH (1994) Approximate methods for sequential decision making using expert advice. In: Proceedings of the seventh annual conference on computational learning theory, COLT ’94, pp 183–189. ACM, New York, NY, USA
Collobert R, Sinz F, Weston J, Bottou L (2006) Large scale transductive SVMs. J Mach Learn Res 7:1687–1712
Collobert R, Sinz F, Weston J, Bottou L (2006) Trading convexity for scalability. In: Proceedings of the 23rd international conference on machine learning, ICML ’06, pp 201–208. New York, NY, USA
Conover WJ (1999) Pratical nonparametric statistics, 3rd edn. Wiley, Hoboken
Crammer K, Dekel O, Keshet J, Shalev-Shwartz S, Singer Y (2006) Online passive-aggressive algorithms. J Mach Learn Res 7:551–585
Dadkhahi H, Shanmugam K, Rios J, Das P, Hoffman SC, Loeffler TD, Sankaranarayanan S (2020) Combinatorial black-box optimization with expert advice. In: Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery and data mining, pp 1918–1927. Association for Computing Machinery, New York, NY, USA
Demšar J (2006) Statistical comparisons of classifiers over multiple data sets. J Mach Learn Res 7:1–30
DeSantis A, Markowsky G, Wegman MN (1988) Learning probabilistic prediction functions. In: Proceedings of the first annual workshop on computational learning theory, COLT’88, pp. 312–328. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA
Devaine M, Gaillard P, Goude Y, Stoltz G (2013) Forecasting electricity consumption by aggregating specialized experts. Mach Learn 90(2):231–260
Freund Y, Schapire RE (1997) A decision-theoretic generalization of on-line learning and an application to boosting. J Comput Syst Sci 55(1):119–139
Friedman M (1937) The use of ranks to avoid the assumption of normality implicit in the analysis of variance. J Am Stat Assoc 32(200):675–701
Friedman M (1940) A comparison of alternative tests of significance for the problem of \(m\) rankings. Ann Math Stat 11(1):86–92
García S, Herrera F (2009) An extension on “statistical comparisons of classifiers over multiple data sets” for all pairwise comparisons. J Mach Learn Res 9:2677–2694
Gentile C (2002) A new approximate maximal margin classification algorithm. J Mach Learn Res 2:213–242
Gentile C (2003) The robustness of the \(p\)-norm algorithms. Mach Learn 53(3):265–299
Gollapudi S, Panigrahi D (2019) Online algorithms for rent-or-buy with expert advice. In: Chaudhuri K, Salakhutdinov R (eds) Proceedings of the 36th International Conference on Machine Learning, vol. 97, pp 2319–2327. PMLR, Long Beach, California, USA
Gramacy RB, Warmuth MKK, Brandt SA, Ari I (2003) Adaptive caching by refetching. In: Becker S, Thrun S, Obermayer K (eds) Advances in neural information processing systems, vol 15. MIT Press, Cambridge, pp 1489–1496
Grove AJ, Littlestone N, Schuurmans D (2001) General convergence results for linear discriminant updates. Mach Learn 43(3):173–210
Hao S, Hu P, Zhao P, Hoi SCH, Miao C (2018) Online active learning with expert advice. ACM Trans Knowl Discov Data 12(5):1–22
Haussler D, Kivinen J, Warmuth MK (1995) Tight worst-case loss bounds for predicting with expert advice. In: Vitányi P (ed) Computational learning theory, lecture notes in computer Science, vol 904. Springer, Berlin, pp 69–83
Hazan E (2016) Introduction to online convex optimization. Found Trends Optim 2(3–4):157–325
Ho VT, Le Thi HA, Bui DC (2016) Online DC optimization for online binary linear classification. In: Nguyen TN, Trawiński B, Fujita H, Hong TP (eds) Intelligent information and database systems: 8th Asian conference, ACIIDS 2016, proceedings, Part II. Springer, Berlin, pp 661–670
Hoi SCH, Wang J, Zhao P (2014) LIBOL: a library for online learning algorithms. J Mach Learn Res 15(1):495–499
Jamil W, Bouchachia A (2019) Model selection in online learning for times series forecasting. In: Lotfi A, Bouchachia H, Gegov A, Langensiepen C, McGinnity M (eds) Advances in computational intelligence systems. Springer, Cham, pp 83–95
Kivinen J, Warmuth MK (1997) Exponentiated gradient versus gradient descent for linear predictors. Inf Comput 132(1):1–63
Kivinen J, Warmuth MK (2001) Relative loss bounds for multidimensional regression problems. Mach Learn 45(3):301–329
Kveton B, Yu JY, Theocharous G, Mannor S (2008) Online learning with expert advice and finite-horizon constraints. In: Proceedings of the twenty-third AAAI conference on artificial intelligence, AAAI 2008, pp 331–336. AAAI Press
Le Thi HA (1994) Analyse numérique des algorithmes de l’optimisation d. C. Approches locale et globale. Codes et simulations numériques en grande dimension. Applications. Ph.D. thesis, University of Rouen, France
Le Thi HA (2020) DC programming and DCA for supply chain and production management: state-of-the-art models and methods. Int J Prod Res 58(20):6078–6114
Le Thi HA, Ho VT, Pham Dinh T (2019) A unified DC programming framework and efficient DCA based approaches for large scale batch reinforcement learning. J Glob Optim 73(2):279–310
Le Thi HA, Le HM, Phan DN, Tran B (2020) Stochastic DCA for minimizing a large sum of DC functions with application to multi-class logistic regression. Neural Netw 132:220–231
Le Thi HA, Moeini M, Pham Dinh T (2009) Portfolio selection under downside risk measures and cardinality constraints based on DC programming and DCA. Comput Manag Sci 6(4):459–475
Le Thi HA, Pham Dinh T (2001) DC programming approach to the multidimensional scaling problem. In: Migdalas A, Pardalos PM, Värbrand P (eds) From local to global optimization. Springer, Boston, pp 231–276
Le Thi HA, Pham Dinh T (2003) Large-scale molecular optimization from distance matrices by a D.C. optimization approach. SIAM J Optim 14(1):77–114
Le Thi HA, Pham Dinh T (2005) The DC (difference of convex functions) programming and DCA revisited with DC models of real world nonconvex optimization problems. Ann Oper Res 133(1–4):23–48
Le Thi HA, Pham Dinh T (2014) DC programming in communication systems: challenging problems and methods. Vietnam J Comput Sci 1(1):15–28
Le Thi HA, Pham Dinh T (2018) DC programming and DCA: thirty years of developments. Math Program Spec Issue DC Program Theory Algorithms Appl 169(1):5–68
Li Y, Long P (2002) The relaxed online maximum margin algorithm. Mach Learn 46(1–3):361–387
Littlestone N, Warmuth MK (1994) The weighted majority algorithm. Inf Comput 108(2):212–261
Nayman N, Noy A, Ridnik T, Friedman I, Jin R, Zelnik-Manor L (2019) XNAS: neural architecture search with expert advice. In: Advances in neural information processing systems 32: annual conference on neural information processing systems 2019, NeurIPS 2019, pp 1975–1985
Novikoff AB (1963) On convergence proofs for perceptrons. In: Proceedings of the symposium on the mathematical theory of automata 12:615–622
Ong CS, Le Thi HA (2013) Learning sparse classifiers with difference of convex functions algorithms. Optim Methods Softw 28(4):830–854
Pereira DG, Afonso A, Medeiros FM (2014) Overview of Friedman’s test and post-hoc analysis. Commun Stat Simul Comput 44(10):2636–2653
Pham Dinh T, Le HM, Le Thi HA, Lauer F (2014) A difference of convex functions algorithm for switched linear regression. IEEE Trans Autom Control 59(8):2277–2282
Pham Dinh T, Le Thi HA (1997) Convex analysis approach to D.C. programming: theory, algorithm and applications. Acta Math Vietnam 22(1):289–355
Pham Dinh T, Le Thi HA (1998) DC optimization algorithms for solving the trust region subproblem. SIAM J Optim 8(2):476–505
Pham Dinh T, Le Thi HA (2014) Recent advances in DC programming and DCA. In: Nguyen NT, Le Thi HA (eds) Transactions on computational intelligence XIII, vol 8342. Springer, Berlin, pp 1–37
Rosenblatt F (1958) The perceptron: a probabilistic model for information storage and organization in the brain. Psychol Rev 65(6):386–408
Shalev-Shwartz S (2007) Online learning: theory, algorithms, and applications. Ph.D. thesis, The Hebrew University of Jerusalem
Shalev-Shwartz S (2012) Online learning and online convex optimization. Found Trends Mach Learn 4(2):107–194
Shor NZ (1985) Minimization methods for non-differentiable functions, 1 edn. Springer Series in Computational Mathematics 3. Springer, Berlin
Valadier M (1969) Sous-différentiels d’une borne supérieure et d’une somme continue de fonctions convexes. CR Acad. Sci. Paris Sér. AB 268:A39–A42
Van Der Malsburg C (1986) Frank rosenblatt: principles of neurodynamics: perceptrons and the theory of brain mechanisms. In: Palm G, Aertsen A (eds) Brain theory. Springer, Berlin, pp 245–248
Vovk V (1990) Aggregating strategies. In: Proceedings of the third annual workshop on computational learning theory. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, pp 371–386
Vovk V (1998) A game of prediction with expert advice. J Comput Syst Sci 56(2):153–173
Wang W, Carreira-Perpiñán MÁ (2013) Projection onto the probability simplex: an efficient algorithm with a simple proof, and an application. arxiv: 1309.1541
Wilcoxon F (1945) Individual comparisons by ranking methods. Biometr Bull 1(6):80–83
Wu P, Hoi SCH, Zhao P, Miao C, Liu Z (2016) Online multi-modal distance metric learning with application to image retrieval. IEEE Trans Knowl Data Eng 28(2):454–467
Yaroshinsky R, El-Yaniv R, Seiden SS (2004) How to better use expert advice. Mach Learn 55(3):271–309
Zinkevich M (2003) Online convex programming and generalized infinitesimal gradient ascent. In: Fawcett T, Mishra N (eds) Proceedings of the 20th international conference on machine learning (ICML-03), pp 928–936
Acknowledgements
This research is funded by Foundation for Science and Technology Development of Ton Duc Thang University (FOSTECT), website: http://fostect.tdtu.edu.vn, under Grant FOSTECT.2017.BR.10.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix 1: DC decomposition of \({f}^{(i)}_t\)
We present the following proposition to get a DC decomposition of \({f}^{(i)}_t\).
Proposition 3
Let \(a > 0\), b, c and x be three constants and a given vector, respectively. The function
is a DC function with DC components
Proof
Knowing from [54] that if \(f = g - h\) is a DC function, then the function
is DC too. We see that the function
is DC. Therefore, applying (23) for \(g = b\langle w, x \rangle + c\) and \(h = \max \{0,b\langle w, x \rangle + c - a \}\), we get immediately a DC decomposition of f, that is,
However, since \(a > 0\), we have
Hence, we complete the proof. \(\square\)
Appendix 2: Proof of Lemma 1
Proof
First, we observe that when \(\tilde{\mathtt {p}}_{t-1} = 0\) or \(\tau ^{(1)}_{t-1} > \langle w^{t-1}, \tilde{\mathtt {p}}_{t-1} \rangle\) (resp. \(\tau ^{(2)}_{t-1} < \langle w^{t-1}, \tilde{\mathtt {p}}_{t-1} \rangle\)), our algorithms do not update the weight, and thus, the result in Lemma 1 is straightforward. Hence, we need to consider only the cases where, at each step \(t \in \mathcal {M}\),
Since \(u^*\) satisfies (18) and \(\tilde{\mathtt {p}}_t \ne 0\), Assumption 1 (i) is satisfied for DC functions (8). We see that \(\Vert u^* - w^t \Vert _2 \ne 0\) for all t because if there is some step t such that \(u^* = w^t\), then \(\langle u^*,\tilde{\mathtt {p}}_t \rangle = \langle w^t,\tilde{\mathtt {p}}_t \rangle\), which contradicts (18).
Next, we verify Assumptions 1 (ii)–(iv) for the DC functions \(f^{(1)}_t\).
\(\bullet\) Let us define the function \(\overline{g}^{(1)}_t := {g}^{(1)}_t - \langle z^t, \cdot \rangle .\) From (24), we derive
Thus, there exists a positive number \(\alpha \le \min \limits _{t \in \mathcal {M}}{\frac{2 }{ \Vert u^*-w^t\Vert _2^2}}\) such that Assumption 1 (ii) is satisfied.
\(\bullet\) For any \(\beta \ge 0\), we have
Thus, Assumption 1 (iii) is satisfied.
\(\bullet\) Assumption 1 (iv) is also verified with \(\gamma \le \min \limits _{t \in \mathcal {M}}{\frac{2(\langle u^*, \tilde{\mathtt {p}}_t \rangle -\rho )}{(\rho - \tau ^{(1)}_{t})\Vert u^*-w^t\Vert _2^2}}\) since
Similarly, as for the DC functions \(f_t^{(i)}\) (\(i=2,3\)), Assumptions 1 (i)–(iv) are satisfied if
The proof of Lemma 1 is established. \(\square\)
Appendix 3: Proof of Theorem 1
Proof
First of all, we analyze the regret bound of ODCA-SG.
From the definition (17), we have
It derives from Assumption 1 (i) that
where the convex function \(\overline{g}_t := g_t - \langle z^t, \cdot \rangle\) for \(t=1,\ldots ,T\).
From (25), (26) and Assumptions 1 (ii)–(iii) with the choice \(\beta = \alpha\), we obtain
The last inequality holds as \(s^{t,k} \in \partial \overline{g}_t(w^{t,k})\), \(k=0,1,\ldots ,K_t-1\).
Similarly to Theorem 3.1 in [30], we can derive from (13) an upper bound of \(\langle s^{t,K_t-1}, w^{t,K_t-1} - u^* \rangle\) as follows:
Combining (19) and the fact that
we yield
It implies
Similarly, we get
We deduce from (28), (31), (32) that
where, by convention, \(\frac{1}{\eta _0}:=0\).
Let us define \(\eta _t := \dfrac{1}{\sqrt{t}\sqrt{3K^2-4K+2}}\) for all \(t=1,\ldots ,T\). We have
As for ODCA-SGk, since Assumption 1 (iv) is also satisfied, we can derive from (27) that
Defining \(\eta _t := \frac{1}{\gamma t}\) for all \(t=1,\ldots ,T\), we obtain
The proof of Theorem 1 is established. \(\square\)
Appendix 4: Proof of Proposition 1
Proof
(i) From the condition (7) and Theorem 1, we have, for any \(w \in \mathcal {S}\),
Here, \(|\mathcal {M}|\) is the number of the steps in \(\mathcal {M}\).
It implies that \(|\mathcal {M}| \le \overline{a} + \overline{c}\sqrt{|\mathcal {M}|}\). Thus, the proof of (i) is complete.
(ii) From the definition of \(\gamma _{\mathrm {SG}}\), we derive that for any \(w \in \mathcal {S}\),
It implies
Considering the strictly convex function \(r : (0,+\infty ) \rightarrow \mathbb {R}\),
Since \(\lim \nolimits _{x \rightarrow 0^+}r(x) = \lim \nolimits _{x \rightarrow +\infty }r(x) = +\infty\) and \(r(\overline{b}) \le 0\), equation \(r(x) = 0\) has two roots \(\overline{x}_1\), \(\overline{x}_2\) such that \(0 < \overline{x}_2 \le \overline{b} \le \overline{x}_1\). The proof of (ii) is established. \(\square\)
Appendix 5: Euclidean projection onto the probability simplex
Appendix 6: Description of the experts
We give here a description of the five experts used in the numerical experiments. They are well-known online classification algorithms: perceptron [50, 57, 62], relaxed online maximum margin algorithm [47], approximate maximal margin classification algorithm [23], passive-aggressive learning algorithm [14], classic online gradient descent algorithm [69].
Note that, in this paper, we consider the outcome space \(\mathcal {Y} = \{0,1\}\) and the prediction label \(\tilde{p}_{i,t} = \mathbb {1}_{\langle u_i, x \rangle \ge 0}(x_t) \in \{0,1\}\), where \(u_i\) is the linear classifier of ith expert. Therefore, in the description below, the \(\{0,1\}\) label is used instead to \(\{-1,1\}\) which are often used in linear classification algorithms.
First, the perceptron algorithm is known as the earliest, simplest approach for online binary linear classification [57].
Second, the relaxed online maximum margin algorithm [47] is an incremental algorithm for classification using a linear threshold function. It can be seen as a relaxed version of the algorithm that searches for the separating hyperplane which maximizes the minimum distance from previous instances classified correctly.
Third, the approximate maximal margin classification algorithm [23] consists in approximating the maximal margin hyperplane with respect to \(\ell _p\)-norm (\(p \ge 2\)) for a set of linearly separable data. The proposed algorithm in [23] is called Approximate Large Margin Algorithm.
Fourth, the passive-aggressive learning algorithm [14] computes the classifier based on analytical solution to simple constrained optimization problem which minimizes the distance from the current classifier \(u^t\) to the half-space of vectors which are of the zero hinge-loss on the current sample.
Finally, the classic online gradient descent algorithm [69] uses the gradient descent method for minimizing the hinge-loss function.
Rights and permissions
About this article
Cite this article
Le Thi, H.A., Ho, V.T. DCA for online prediction with expert advice. Neural Comput & Applic 33, 9521–9544 (2021). https://doi.org/10.1007/s00521-021-05709-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00521-021-05709-0