Beyond stochastic gradient descent for large-scale machine learning

Transcription

1 Beyond stochastic gradient descent for large-scale machine learning Francis Bach INRIA - Ecole Normale Supérieure, Paris, France Joint work with Eric Moulines - October 2014

2 Big data revolution? A new scientific context Data everywhere: size does not (always) matter Science and industry Size and variety Learning from examples n observations in dimension p

3 Search engines - Advertising

4 Marketing - Personalized recommendation

5 Visual object recognition

6 Personal photos

7 Bioinformatics Protein: Crucial elements of cell life Massive data: 2 millions for humans Complex data

8 Context Machine learning for big data Large-scale machine learning: large p, large n p : dimension of each observation (input) n : number of observations Examples: computer vision, bioinformatics, advertising Ideal running-time complexity: O(pn) Going back to simple methods Stochastic gradient methods (Robbins and Monro, 1951) Mixing statistics and optimization Using smoothness to go beyond stochastic gradient descent

9 Context Machine learning for big data Large-scale machine learning: large p, large n p : dimension of each observation (input) n : number of observations Examples: computer vision, bioinformatics, advertising Ideal running-time complexity: O(pn) Going back to simple methods Stochastic gradient methods (Robbins and Monro, 1951) Mixing statistics and optimization Using smoothness to go beyond stochastic gradient descent

10 Context Machine learning for big data Large-scale machine learning: large p, large n p : dimension of each observation (input) n : number of observations Examples: computer vision, bioinformatics, advertising Ideal running-time complexity: O(pn) Going back to simple methods Stochastic gradient methods (Robbins and Monro, 1951) Mixing statistics and optimization Using smoothness to go beyond stochastic gradient descent

11 Scaling to large problems with convex optimization Retour aux sources 1950 s: computers not powerful enough IBM 1620, 1959 CPU frequency: 50 KHz Price > dollars 2010 s: Data too massive One pass through the data (Robbins et Monro, 1956) Algorithm: θ n = θ n 1 γ n l (y n, θ n 1,Φ(x n ) )Φ(x n )

12 Scaling to large problems with convex optimization Retour aux sources 1950 s: computers not powerful enough IBM 1620, 1959 CPU frequency: 50 KHz Price > dollars 2010 s: Data too massive One pass through the data (Robbins et Monro, 1951) Algorithm: θ n = θ n 1 γ n l (y n, θ n 1,Φ(x n ) )Φ(x n )

13 Outline Introduction: stochastic approximation algorithms Supervised machine learning and convex optimization Stochastic gradient and averaging Strongly convex vs. non-strongly convex O(1/(nµ)) vs. O(1/ n) for all non-smooth functions Least-squares regressions Constant step-sizes O(1/n) convergence rate for all quadratic functions Smooth losses Online Newton steps O(1/n) convergence rate for all convex functions

14 Supervised machine learning Data: n observations (x i,y i ) X Y, i = 1,...,n, i.i.d. Prediction as a linear function θ,φ(x) of features Φ(x) R p (regularized) empirical risk minimization: find ˆθ solution of 1 n min l ( y i, θ,φ(x i ) ) + µω(θ) θ R p n i=1 convex data fitting term + regularizer

15 Usual losses Regression: y R, prediction ŷ = θ,φ(x) quadratic loss 1 2 (y ŷ)2 = 1 2 (y θ,φ(x) )2

16 Usual losses Regression: y R, prediction ŷ = θ,φ(x) quadratic loss 1 2 (y ŷ)2 = 1 2 (y θ,φ(x) )2 Classification : y { 1,1}, prediction ŷ = sign( θ,φ(x) ) loss of the form l(y θ,φ(x) ) True 0-1 loss: l(y θ,φ(x) ) = 1 y θ,φ(x) <0 Usual convex losses: hinge square logistic

17 Supervised machine learning Data: n observations (x i,y i ) X Y, i = 1,...,n, i.i.d. Prediction as a linear function θ,φ(x) of features Φ(x) R p (regularized) empirical risk minimization: find ˆθ solution of 1 n min l ( y i, θ,φ(x i ) ) + µω(θ) θ R p n i=1 convex data fitting term + regularizer

18 Supervised machine learning Data: n observations (x i,y i ) X Y, i = 1,...,n, i.i.d. Prediction as a linear function θ,φ(x) of features Φ(x) R p (regularized) empirical risk minimization: find ˆθ solution of 1 n min l ( y i, θ,φ(x i ) ) + µω(θ) θ R p n i=1 convex data fitting term + regularizer Empirical risk: ˆf(θ) = 1 n n i=1 l(y i, θ,φ(x i ) ) training cost Expected risk: f(θ) = E (x,y) l(y, θ,φ(x) ) testing cost Two fundamental questions: (1) computing ˆθ and (2) analyzing ˆθ May be tackled simultaneously

19 Supervised machine learning Data: n observations (x i,y i ) X Y, i = 1,...,n, i.i.d. Prediction as a linear function θ,φ(x) of features Φ(x) R p (regularized) empirical risk minimization: find ˆθ solution of 1 n min l ( y i, θ,φ(x i ) ) + µω(θ) θ R p n i=1 convex data fitting term + regularizer Empirical risk: ˆf(θ) = 1 n n i=1 l(y i, θ,φ(x i ) ) training cost Expected risk: f(θ) = E (x,y) l(y, θ,φ(x) ) testing cost Two fundamental questions: (1) computing ˆθ and (2) analyzing ˆθ May be tackled simultaneously

20 Smoothness and strong convexity A function g : R p R is L-smooth if and only if it is twice differentiable and θ R p, eigenvalues [ g (θ) ] L smooth non smooth

21 Smoothness and strong convexity A function g : R p R is L-smooth if and only if it is twice differentiable and θ R p, eigenvalues [ g (θ) ] L Machine learning with g(θ) = 1 n n i=1 l(y i, θ,φ(x i ) ) Hessian covariance matrix 1 n n i=1 Φ(x i) Φ(x i ) Bounded data

22 Smoothness and strong convexity A twice differentiable function g : R p R is µ-strongly convex if and only if θ R p, eigenvalues [ g (θ) ] µ convex strongly convex

23 Smoothness and strong convexity A twice differentiable function g : R p R is µ-strongly convex if and only if θ R p, eigenvalues [ g (θ) ] µ (large µ) (small µ)

24 Smoothness and strong convexity A twice differentiable function g : R p R is µ-strongly convex if and only if θ R p, eigenvalues [ g (θ) ] µ Machine learning with g(θ) = 1 n n i=1 l(y i, θ,φ(x i ) ) Hessian covariance matrix 1 n n i=1 Φ(x i) Φ(x i ) Data with invertible covariance matrix (low correlation/dimension)

25 Smoothness and strong convexity A twice differentiable function g : R p R is µ-strongly convex if and only if θ R p, eigenvalues [ g (θ) ] µ Machine learning with g(θ) = 1 n n i=1 l(y i, θ,φ(x i ) ) Hessian covariance matrix 1 n n i=1 Φ(x i) Φ(x i ) Data with invertible covariance matrix (low correlation/dimension) Adding regularization by µ 2 θ 2 creates additional bias unless µ is small

26 Iterative methods for minimizing smooth functions Assumption: g convex and smooth on R p Gradient descent: θ t = θ t 1 γ t g (θ t 1 ) O(1/t) convergence rate for convex functions O(e ρt ) convergence rate for strongly convex functions

27 Iterative methods for minimizing smooth functions Assumption: g convex and smooth on R p Gradient descent: θ t = θ t 1 γ t g (θ t 1 ) O(1/t) convergence rate for convex functions O(e ρt ) convergence rate for strongly convex functions Newton method: θ t = θ t 1 g (θ t 1 ) 1 g (θ t 1 ) O ( e ρ2t ) convergence rate

28 Iterative methods for minimizing smooth functions Assumption: g convex and smooth on R p Gradient descent: θ t = θ t 1 γ t g (θ t 1 ) O(1/t) convergence rate for convex functions O(e ρt ) convergence rate for strongly convex functions Newton method: θ t = θ t 1 g (θ t 1 ) 1 g (θ t 1 ) O ( e ρ2t ) convergence rate Key insights from Bottou and Bousquet (2008) 1. In machine learning, no need to optimize below statistical error 2. In machine learning, cost functions are averages Stochastic approximation

29 Stochastic approximation Goal: Minimizing a function f defined on R p given only unbiased estimates f n(θ n ) of its gradients f (θ n ) at certain points θ n R p

30 Stochastic approximation Goal: Minimizing a function f defined on R p given only unbiased estimates f n(θ n ) of its gradients f (θ n ) at certain points θ n R p Machine learning - statistics f(θ) = Ef n (θ) = El(y n, θ,φ(x n ) ) = generalization error Loss for a single pair of observations: f n (θ) = l(y n, θ,φ(x n ) ) Expected gradient: f (θ) = Ef n(θ) = E { l (y n, θ,φ(x n ) )Φ(x n ) } Beyond convex optimization: see, e.g., Benveniste et al. (2012)

31 Convex stochastic approximation Key assumption: smoothness and/or strong convexity Key algorithm: stochastic gradient descent (a.k.a. Robbins-Monro) θ n = θ n 1 γ n f n(θ n 1 ) Polyak-Ruppert averaging: θ n = 1 n+1 n k=0 θ k Which learning rate sequence γ n? Classical setting: γ n = Cn α

32 Convex stochastic approximation Key assumption: smoothness and/or strong convexity Key algorithm: stochastic gradient descent (a.k.a. Robbins-Monro) θ n = θ n 1 γ n f n(θ n 1 ) Polyak-Ruppert averaging: θ n = 1 n+1 n k=0 θ k Which learning rate sequence γ n? Classical setting: γ n = Cn α Running-time = O(np) Single pass through the data One line of code among many

33 Convex stochastic approximation Existing work Known global minimax rates of convergence for non-smooth problems (Nemirovsky and Yudin, 1983; Agarwal et al., 2012) Strongly convex: O((µn) 1 ) Attainedbyaveragedstochasticgradientdescentwithγ n (µn) 1 Non-strongly convex: O(n 1/2 ) Attained by averaged stochastic gradient descent with γ n n 1/2

34 Convex stochastic approximation Existing work Known global minimax rates of convergence for non-smooth problems (Nemirovsky and Yudin, 1983; Agarwal et al., 2012) Strongly convex: O((µn) 1 ) Attainedbyaveragedstochasticgradientdescentwithγ n (µn) 1 Non-strongly convex: O(n 1/2 ) Attained by averaged stochastic gradient descent with γ n n 1/2 Asymptotic analysis of averaging (Polyak and Juditsky, 1992; Ruppert, 1988) All step sizes γ n = Cn α with α (1/2,1) lead to O(n 1 ) for smooth strongly convex problems A single algorithm with global adaptive convergence rate for smooth problems?

35 Convex stochastic approximation Existing work Known global minimax rates of convergence for non-smooth problems (Nemirovsky and Yudin, 1983; Agarwal et al., 2012) Strongly convex: O((µn) 1 ) Attainedbyaveragedstochasticgradientdescentwithγ n (µn) 1 Non-strongly convex: O(n 1/2 ) Attained by averaged stochastic gradient descent with γ n n 1/2 Asymptotic analysis of averaging (Polyak and Juditsky, 1992; Ruppert, 1988) All step sizes γ n = Cn α with α (1/2,1) lead to O(n 1 ) for smooth strongly convex problems A single algorithm for smooth problems with convergence rate O(1/n) in all situations?

36 Least-mean-square algorithm Least-squares: f(θ) = 1 2 E[ (y n Φ(x n ),θ ) 2] with θ R p SGD = least-mean-square algorithm (see, e.g., Macchi, 1995) usually studied without averaging and decreasing step-sizes with strong convexity assumption E [ Φ(x n ) Φ(x n ) ] = H µ Id

37 Least-mean-square algorithm Least-squares: f(θ) = 1 2 E[ (y n Φ(x n ),θ ) 2] with θ R p SGD = least-mean-square algorithm (see, e.g., Macchi, 1995) usually studied without averaging and decreasing step-sizes with strong convexity assumption E [ Φ(x n ) Φ(x n ) ] = H µ Id New analysis for averaging and constant step-size γ = 1/(4R 2 ) Assume Φ(x n ) R and y n Φ(x n ),θ σ almost surely No assumption regarding lowest eigenvalues of H Main result: Ef( θ n 1 ) f(θ ) 4σ2 p n + 4R2 θ 0 θ 2 n Matches statistical lower bound (Tsybakov, 2003) Non-asymptotic robust version of Györfi and Walk (1996)

38 Markov chain interpretation of constant step sizes LMS recursion for f n (θ) = 1 2 ( yn Φ(x n ),θ ) 2 θ n = θ n 1 γ ( Φ(x n ),θ n 1 y n ) Φ(xn ) The sequence (θ n ) n is a homogeneous Markov chain convergence to a stationary distribution π γ with expectation θ γ def = θπ γ (dθ) - For least-squares, θ γ = θ θ n θ γ θ 0

39 Markov chain interpretation of constant step sizes LMS recursion for f n (θ) = 1 2 ( yn Φ(x n ),θ ) 2 θ n = θ n 1 γ ( Φ(x n ),θ n 1 y n ) Φ(xn ) The sequence (θ n ) n is a homogeneous Markov chain convergence to a stationary distribution π γ with expectation θ γ def = θπ γ (dθ) For least-squares, θ γ = θ θ n θ θ 0

40 Markov chain interpretation of constant step sizes LMS recursion for f n (θ) = 1 2 ( yn Φ(x n ),θ ) 2 θ n = θ n 1 γ ( Φ(x n ),θ n 1 y n ) Φ(xn ) The sequence (θ n ) n is a homogeneous Markov chain convergence to a stationary distribution π γ with expectation θ γ def = θπ γ (dθ) For least-squares, θ γ = θ θ n θ n θ θ 0

41 Markov chain interpretation of constant step sizes LMS recursion for f n (θ) = 1 2 ( yn Φ(x n ),θ ) 2 θ n = θ n 1 γ ( Φ(x n ),θ n 1 y n ) Φ(xn ) The sequence (θ n ) n is a homogeneous Markov chain convergence to a stationary distribution π γ with expectation θ γ def = θπ γ (dθ) For least-squares, θ γ = θ θ n does not converge to θ but oscillates around it oscillations of order γ Ergodic theorem: Averaged iterates converge to θ γ = θ at rate O(1/n)

42 Simulations - synthetic examples Gaussian distributions - p = 20 0 synthetic square log 10 [f(θ) f(θ * )] /2R 2 1/8R 2 4 1/32R 2 1/2R 2 n 1/ log 10 (n)

43 Simulations - benchmarks alpha (p = 500, n = ), news (p = , n = ) log 10 [f(θ) f(θ * )] alpha square C=1 test 1/R 2 1/R 2 n 1/2 SAG log (n) alpha square C=opt test C/R 2 C/R 2 n 1/2 SAG log (n) news square C=1 test 0.2 news square C=opt test log 10 [f(θ) f(θ * )] /R 2 1/R 2 n 1/2 SAG log (n) C/R 2 C/R 2 n 1/2 SAG log 10 (n)

44 Beyond least-squares - Markov chain interpretation Recursion θ n = θ n 1 γf n(θ n 1 ) also defines a Markov chain Stationary distribution π γ such that f (θ)π γ (dθ) = 0 When f is not linear, f ( θπ γ (dθ)) f (θ)π γ (dθ) = 0

45 Beyond least-squares - Markov chain interpretation Recursion θ n = θ n 1 γf n(θ n 1 ) also defines a Markov chain Stationary distribution π γ such that f (θ)π γ (dθ) = 0 When f is not linear, f ( θπ γ (dθ)) f (θ)π γ (dθ) = 0 θ n oscillates around the wrong value θ γ θ θ n θ n θ γ θ 0 θ

46 Beyond least-squares - Markov chain interpretation Recursion θ n = θ n 1 γf n(θ n 1 ) also defines a Markov chain Stationary distribution π γ such that f (θ)π γ (dθ) = 0 When f is not linear, f ( θπ γ (dθ)) f (θ)π γ (dθ) = 0 θ n oscillates around the wrong value θ γ θ moreover, θ θ n = O p ( γ) Ergodic theorem averaged iterates converge to θ γ θ at rate O(1/n) moreover, θ θ γ = O(γ) (Bach, 2013)

47 Simulations - synthetic examples Gaussian distributions - p = 20 0 synthetic logistic 1 log 10 [f(θ) f(θ * )] 1 2 1/2R 2 3 1/8R 2 4 1/32R 2 5 1/2R 2 n 1/ log 10 (n)

48 Restoring convergence through online Newton steps Known facts 1. Averaged SGD with γ n n 1/2 leads to robust rate O(n 1/2 ) for all convex functions 2. Averaged SGD with γ n constant leads to robust rate O(n 1 ) for all convex quadratic functions 3. Newton s method squares the error at each iteration for smooth functions 4. A single step of Newton s method is equivalent to minimizing the quadratic Taylor expansion Online Newton step Rate: O((n 1/2 ) 2 +n 1 ) = O(n 1 ) Complexity: O(p) per iteration

49 Restoring convergence through online Newton steps Known facts 1. Averaged SGD with γ n n 1/2 leads to robust rate O(n 1/2 ) for all convex functions 2. Averaged SGD with γ n constant leads to robust rate O(n 1 ) for all convex quadratic functions O(n 1 ) 3. Newton s method squares the error at each iteration for smooth functions O((n 1/2 ) 2 ) 4. A single step of Newton s method is equivalent to minimizing the quadratic Taylor expansion Online Newton step Rate: O((n 1/2 ) 2 +n 1 ) = O(n 1 ) Complexity: O(p) per iteration

50 Restoring convergence through online Newton steps The Newton step for f = Ef n (θ) def = E [ l(y n, θ,φ(x n ) ) ] at θ is equivalent to minimizing the quadratic approximation g(θ) = f( θ)+ f ( θ),θ θ θ θ,f ( θ)(θ θ) = f( θ)+ Ef n( θ),θ θ θ θ,ef n( θ)(θ θ) [ = E f( θ)+ f n( θ),θ θ θ θ,f n( θ)(θ θ) ]

51 Restoring convergence through online Newton steps The Newton step for f = Ef n (θ) def = E [ l(y n, θ,φ(x n ) ) ] at θ is equivalent to minimizing the quadratic approximation g(θ) = f( θ)+ f ( θ),θ θ θ θ,f ( θ)(θ θ) = f( θ)+ Ef n( θ),θ θ θ θ,ef n( θ)(θ θ) [ = E f( θ)+ f n( θ),θ θ θ θ,f n( θ)(θ θ) ] Complexity of least-mean-square recursion for g is O(p) θ n = θ n 1 γ [ f n( θ)+f n( θ)(θ n 1 θ) ] f n( θ) = l (y n, θ,φ(x n ) )Φ(x n ) Φ(x n ) has rank one New online Newton step without computing/inverting Hessians

52 Choice of support point for online Newton step Two-stage procedure (1) Run n/2 iterations of averaged SGD to obtain θ (2) Run n/2 iterations of averaged constant step-size LMS Reminiscent of one-step estimators (see, e.g., Van der Vaart, 2000) Provable convergence rate of O(p/n) for logistic regression Additional assumptions but no strong convexity

53 Choice of support point for online Newton step Two-stage procedure (1) Run n/2 iterations of averaged SGD to obtain θ (2) Run n/2 iterations of averaged constant step-size LMS Reminiscent of one-step estimators (see, e.g., Van der Vaart, 2000) Provable convergence rate of O(p/n) for logistic regression Additional assumptions but no strong convexity Update at each iteration using the current averaged iterate Recursion: θ n = θ n 1 γ [ f n( θ n 1 )+f n( θ n 1 )(θ n 1 θ n 1 ) ] No provable convergence rate (yet) but best practical behavior Note (dis)similarity with regular SGD: θ n = θ n 1 γf n(θ n 1 )

54 Simulations - synthetic examples Gaussian distributions - p = 20 0 synthetic logistic 1 0 synthetic logistic 2 log 10 [f(θ) f(θ * )] /2R 2 3 1/8R 2 3 every 2 p 4 1/32R 2 every iter. 4 2 step 5 1/2R 2 n 1/2 5 2 step dbl log 10 (n) log 10 (n) log 10 [f(θ) f(θ * )] 1

55 Simulations - benchmarks alpha (p = 500, n = ), news (p = , n = ) alpha logistic C=1 test alpha logistic C=opt test log 10 [f(θ) f(θ * )] /R /R 2 n 1/2 SAG 2 Adagrad Newton log (n) C/R C/R 2 n 1/2 SAG 2 Adagrad Newton log (n) news logistic C=1 test 0.2 news logistic C=opt test log 10 [f(θ) f(θ * )] /R /R 2 n 1/2 SAG 0.8 Adagrad 1 Newton log (n) C/R C/R 2 n 1/2 SAG 0.8 Adagrad 1 Newton log 10 (n)

56 Conclusions Constant-step-size averaged stochastic gradient descent Reaches convergence rate O(1/n) in all regimes Improves on the O(1/ n) lower-bound of non-smooth problems Efficient online Newton step for non-quadratic problems Robustness to step-size selection Extensions and future work Going beyond a single pass Pre-conditioning Proximal extensions fo non-differentiable terms kernels and non-parametric estimation(dieuleveut and Bach, 2014) line-search parallelization Non-convex problems

57 References A. Agarwal, P. L. Bartlett, P. Ravikumar, and M. J. Wainwright. Information-theoretic lower bounds on the oracle complexity of stochastic convex optimization. Information Theory, IEEE Transactions on, 58(5): , F. Bach. Adaptivity of averaged stochastic gradient descent to local strong convexity for logistic regression. Technical Report , HAL, Albert Benveniste, Michel Métivier, and Pierre Priouret. Adaptive algorithms and stochastic approximations. Springer Publishing Company, Incorporated, L. Bottou and O. Bousquet. The tradeoffs of large scale learning. In Adv. NIPS, A. Dieuleveut and F. Bach. Non-parametric Stochastic Approximation with Large Step sizes. Technical report, ArXiv, L. Györfi and H. Walk. On the averaged stochastic approximation for linear regression. SIAM Journal on Control and Optimization, 34(1):31 61, O. Macchi. Adaptive processing: The least mean squares approach with applications in transmission. Wiley West Sussex, A. S. Nemirovsky and D. B. Yudin. Problem complexity and method efficiency in optimization. Wiley & Sons, B. T. Polyak and A. B. Juditsky. Acceleration of stochastic approximation by averaging. SIAM Journal on Control and Optimization, 30(4): , 1992.

58 H. Robbins and S. Monro. A stochastic approximation method. Ann. Math. Statistics, 22: , ISSN D. Ruppert. Efficient estimations from a slowly convergent Robbins-Monro process. Technical Report 781, Cornell University Operations Research and Industrial Engineering, A. B. Tsybakov. Optimal rates of aggregation A. W. Van der Vaart. Asymptotic statistics, volume 3. Cambridge Univ. press, 2000.