jmlr jmlr2010 jmlr2010-93 knowledge-graph by maker-knowledge-mining

93 jmlr-2010-PyBrain

Source: pdf

Author: Tom Schaul, Justin Bayer, Daan Wierstra, Yi Sun, Martin Felder, Frank Sehnke, Thomas Rückstieß, Jürgen Schmidhuber

Abstract: PyBrain is a versatile machine learning library for Python. Its goal is to provide ﬂexible, easyto-use yet still powerful algorithms for machine learning tasks, including a variety of predeﬁned environments and benchmarks to test and compare algorithms. Implemented algorithms include Long Short-Term Memory (LSTM), policy gradient methods, (multidimensional) recurrent neural networks and deep belief networks. Keywords: Python, neural networks, reinforcement learning, optimization

Reference: text

Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 CH Technische Universit¨ t M¨ nchen a u Garching D-86748, Germany Editor: Soeren Sonnenburg Abstract PyBrain is a versatile machine learning library for Python. [sent-12, score-0.149]

2 Its goal is to provide ﬂexible, easyto-use yet still powerful algorithms for machine learning tasks, including a variety of predeﬁned environments and benchmarks to test and compare algorithms. [sent-13, score-0.098]

3 Implemented algorithms include Long Short-Term Memory (LSTM), policy gradient methods, (multidimensional) recurrent neural networks and deep belief networks. [sent-14, score-0.615]

4 Keywords: Python, neural networks, reinforcement learning, optimization 1. [sent-15, score-0.159]

5 Introduction PyBrain is a machine learning library written in Python designed to facilitate both the application of and research on premier learning algorithms such as LSTM (Hochreiter and Schmidhuber, 1997), deep belief networks, and policy gradient algorithms. [sent-16, score-0.379]

6 Emphasizing both sequential and nonsequential data and tasks, PyBrain implements many recent learning algorithms and architectures ranging from areas such as supervised learning and reinforcement learning to direct search / optimization and evolutionary methods. [sent-17, score-0.448]

7 PyBrain is implemented in Python, with the scientiﬁc library SciPy being its only strict dependency. [sent-18, score-0.123]

8 As is typical for programming in Python/SciPy, development time is greatly reduced as compared to languages such as Java/C++, at the cost of lower speed. [sent-19, score-0.036]

9 PyBrain embodies a compositional setup, which means that it is designed to be able to connect various types of architectures and algorithms. [sent-20, score-0.218]

10 PyBrain goes beyond existing Python libraries in breadth in that it provides a toolbox for supervised, unsupervised and reinforcement learning as well as black-box and multi-objective optimization. [sent-21, score-0.255]

11 c 2010 Tom Schaul, Justin Bayer, Daan Wierstra, Yi Sun, Martin Felder, Frank Sehnke, Thomas R¨ ckstieß and J¨ rgen Schmidhuber. [sent-24, score-0.234]

12 Furthermore, it sets itself apart by its ﬂexibility for composing custom neural networks architectures, ranging from (multi-dimensional) recurrent networks to restricted Boltzmann machines or convolutional networks. [sent-26, score-0.446]

13 Library Overview The library includes different types of training algorithms, trainable architectural components, specialized data sets and standardized benchmark tasks/environments. [sent-28, score-0.149]

14 The available algorithms generally function both in sequential and non-sequential settings, and appropriate data handling tools have been developed for special applications, ranging from reinforcement learning to handwriting recognition applications. [sent-29, score-0.198]

15 Supervised Learning Training algorithms include classical gradient-based methods and extensions both for non-sequential and sequential data. [sent-32, score-0.028]

16 Reinforcement Learning The reinforcement learning algorithms of PyBrain encompass basic methods such as Q-learning, SARSA and REINFORCE, but also natural actor-critic, neural-ﬁtted Q-iteration, recurrent policy gradients, state-dependent exploration and reward-weighted regression. [sent-36, score-0.558]

17 Architectures Available architectures include standard feedforward neural networks, recurrent neural networks and LSTM, bi- and multidimensional recurrent neural networks and deep belief networks. [sent-37, score-1.07]

18 The library puts emphasis on, but is not limited to, (recurrent) neural network structures arbitrarily structured as a directed acyclic graph of modules and connections. [sent-38, score-0.186]

19 Compositionality The basic structure of the library enables a compositional approach to machine learning. [sent-39, score-0.142]

20 Different algorithms and architectures can be connected and composed, and then used and trained as desired. [sent-40, score-0.17]

21 For example, arbitrarily structured recurrent neural network graphs can be trained using various different algorithms such as black-box search, policy gradients and supervised learning. [sent-41, score-0.537]

22 For example, both an LSTM architecture and a deep belief network, though constituting wildly different architectures, can be trained using the same gradient implementation (e. [sent-42, score-0.152]

23 Tasks and Benchmarks For black-box and multi-objective optimization, standard benchmarks are included in the library. [sent-45, score-0.05]

24 For reinforcement learning settings, there are classical (PO)MDP mazes, (non-Markov) (double) pole balancing and various ODE environments based on physics simulations. [sent-46, score-0.18]

25 In order to make data processing as easy as possible, PyBrain features functionality for constructing, serializing and deserializing data sets to ﬁles or over network connections. [sent-47, score-0.039]

26 net = buildNetwork(1, 2, 1, bias=True, hiddenclass=TanhLayer, outclass=TanhLayer, recurrent=True) recCon = FullConnection(net[’out’], net[’hidden0’]) net. [sent-52, score-0.048]

27 sortModules() # Create a trainer for backprop and train the net. [sent-54, score-0.067]

28 trainEpochs(1000) Figure 1: Code example for solving the parity problem with a recurrent neural network. [sent-57, score-0.322]

29 Speed Optimization The development cycle of Python/SciPy is short, especially compared to languages such as Java/C/C++ and especially in scientiﬁc / machine learning settings. [sent-58, score-0.036]

30 In order to partially mitigate this potential problem, we have implemented many parts of the library in C/C++ using SWIG as a bridge. [sent-60, score-0.123]

31 An Illustrative Example In order to provide a useful example to a novice user, we construct a recurrent network that is able to solve the parity problem. [sent-64, score-0.363]

32 The network is given a sequence of binary inputs and the corresponding objective is to determine whether the network has been provided an even or an odd number of 1s so far. [sent-65, score-0.078]

33 The implemented topology and the corresponding code are shown in Figure 1. [sent-66, score-0.06]

34 We ﬁrst construct a feed forward network consisting of a single linear input cell, a layer of two hidden cells and one output cell. [sent-67, score-0.106]

35 The network also needs a bias, which is connected to the hidden and output layer. [sent-68, score-0.039]

36 The network is ﬁrst created using a ‘convenience’ function. [sent-70, score-0.039]

37 A recurrent connection from the output unit to the hidden layer is crucial to learn the problem and thus added afterwards. [sent-71, score-0.289]

38 It should be stressed that the clarity and shortness of this code example is representative for many algorithms and architectures used in actual problems. [sent-72, score-0.227]

39 Constructing networks for classiﬁcation, control or function approximation often requires even fewer lines of code. [sent-73, score-0.052]

40 Concluding Remarks In PyBrain we emphasize simplicity, compositionality and the ability to combine various architectures and algorithm types. [sent-75, score-0.237]

41 We believe this to be advantageous in the light of recent developments in sequence processing, deep belief networks, policy gradients and visual processing. [sent-76, score-0.346]

42 Pybrain is easy to use, and is well-documented, both in code and in documents and tutorials explaining the use of the basic capabilities of the library. [sent-77, score-0.088]

43 Among machine learning libraries written in Python, 745 ¨ S CHAUL , BAYER , W IERSTRA , S UN , F ELDER , S EHNKE , R UCKSTIESS AND S CHMIDHUBER PyBrain stands out for its combination of breadth, versatility and maturity of development. [sent-78, score-0.097]

44 In order to further demonstrate its practical applicability to scientiﬁc research, we could emphasize that PyBrain, so far, has already been used in fourteen scientiﬁc peer-reviewed publications, for example, Sehnke et al. [sent-79, score-0.029]

45 3 and both the entire source code and the documentation are available from the website, www. [sent-86, score-0.067]

46 The documentation includes a quickstart tutorial, installation instructions, tutorials on advanced topics, and an extensive API reference. [sent-93, score-0.093]

47 Since PyBrain is under active development, we encourage researchers to contribute their work and provide guidelines for how they can do so. [sent-94, score-0.026]

48 Thomas R¨ ckstieß, Martin Felder, and J¨ rgen Schmidhuber. [sent-100, score-0.234]

49 Frank Sehnke, Christian Osendorfer, Thomas R¨ ckstieß, Alex Graves, Jan Peters, and J¨ rgen u u Schmidhuber. [sent-106, score-0.234]

50 Yi Sun, Daan Wierstra, Tom Schaul, and J¨ rgen Schmidhuber. [sent-109, score-0.234]

51 Daan Wierstra, Tom Schaul, Jan Peters, and J¨ rgen Schmidhuber. [sent-112, score-0.234]

similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('pybrain', 0.536), ('recurrent', 0.251), ('idsia', 0.234), ('rgen', 0.234), ('schaul', 0.201), ('python', 0.18), ('architectures', 0.17), ('daan', 0.167), ('sehnke', 0.167), ('wierstra', 0.167), ('bayer', 0.134), ('ckstie', 0.134), ('lstm', 0.134), ('policy', 0.133), ('reinforcement', 0.132), ('tom', 0.119), ('felder', 0.114), ('library', 0.094), ('deep', 0.09), ('ch', 0.09), ('tum', 0.086), ('schmidhuber', 0.077), ('sun', 0.072), ('justin', 0.071), ('libraries', 0.071), ('chaul', 0.067), ('chmidhuber', 0.067), ('compositionality', 0.067), ('ehnke', 0.067), ('ierstra', 0.067), ('lugano', 0.067), ('reccon', 0.067), ('scipy', 0.067), ('tanhlayer', 0.067), ('trainer', 0.067), ('uckstiess', 0.067), ('belief', 0.062), ('gradients', 0.061), ('elder', 0.057), ('hochreiter', 0.057), ('tutorials', 0.057), ('evolutionary', 0.054), ('frank', 0.052), ('networks', 0.052), ('peters', 0.052), ('breadth', 0.052), ('benchmarks', 0.05), ('net', 0.048), ('environments', 0.048), ('compositional', 0.048), ('thomas', 0.047), ('parity', 0.044), ('icann', 0.044), ('martin', 0.043), ('exploration', 0.042), ('network', 0.039), ('ranging', 0.038), ('layer', 0.038), ('evolution', 0.037), ('documentation', 0.036), ('jan', 0.036), ('languages', 0.036), ('genetic', 0.035), ('ds', 0.035), ('multidimensional', 0.035), ('scienti', 0.034), ('code', 0.031), ('implemented', 0.029), ('un', 0.029), ('reinforce', 0.029), ('ref', 0.029), ('garching', 0.029), ('rain', 0.029), ('fitness', 0.029), ('gecco', 0.029), ('jurgen', 0.029), ('sarsa', 0.029), ('nchen', 0.029), ('novice', 0.029), ('architectural', 0.029), ('cognition', 0.029), ('feed', 0.029), ('fourteen', 0.029), ('po', 0.029), ('scripts', 0.029), ('sequential', 0.028), ('neural', 0.027), ('supervised', 0.026), ('feedforward', 0.026), ('versatile', 0.026), ('custom', 0.026), ('stressed', 0.026), ('versatility', 0.026), ('guidelines', 0.026), ('trainable', 0.026), ('board', 0.026), ('emphasizing', 0.026), ('modules', 0.026), ('api', 0.026), ('bsd', 0.026)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 1.0000002 93 jmlr-2010-PyBrain

Author: Tom Schaul, Justin Bayer, Daan Wierstra, Yi Sun, Martin Felder, Frank Sehnke, Thomas Rückstieß, Jürgen Schmidhuber

2 0.09773618 117 jmlr-2010-Why Does Unsupervised Pre-training Help Deep Learning?

Author: Dumitru Erhan, Yoshua Bengio, Aaron Courville, Pierre-Antoine Manzagol, Pascal Vincent, Samy Bengio

Abstract: Much recent research has been devoted to learning algorithms for deep architectures such as Deep Belief Networks and stacks of auto-encoder variants, with impressive results obtained in several areas, mostly on vision and language data sets. The best results obtained on supervised learning tasks involve an unsupervised learning component, usually in an unsupervised pre-training phase. Even though these new algorithms have enabled training deep models, many questions remain as to the nature of this difﬁcult learning problem. The main question investigated here is the following: how does unsupervised pre-training work? Answering this questions is important if learning in deep architectures is to be further improved. We propose several explanatory hypotheses and test them through extensive simulations. We empirically show the inﬂuence of pre-training with respect to architecture depth, model capacity, and number of training examples. The experiments conﬁrm and clarify the advantage of unsupervised pre-training. The results suggest that unsupervised pretraining guides the learning towards basins of attraction of minima that support better generalization from the training data set; the evidence from these results supports a regularization explanation for the effect of pre-training. Keywords: deep architectures, unsupervised pre-training, deep belief networks, stacked denoising auto-encoders, non-convex optimization

3 0.053681426 107 jmlr-2010-Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion

Author: Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, Pierre-Antoine Manzagol

Abstract: We explore an original strategy for building deep networks, based on stacking layers of denoising autoencoders which are trained locally to denoise corrupted versions of their inputs. The resulting algorithm is a straightforward variation on the stacking of ordinary autoencoders. It is however shown on a benchmark of classiﬁcation problems to yield signiﬁcantly lower classiﬁcation error, thus bridging the performance gap with deep belief networks (DBN), and in several cases surpassing it. Higher level representations learnt in this purely unsupervised fashion also help boost the performance of subsequent SVM classiﬁers. Qualitative experiments show that, contrary to ordinary autoencoders, denoising autoencoders are able to learn Gabor-like edge detectors from natural image patches and larger stroke detectors from digit images. This work clearly establishes the value of using a denoising criterion as a tractable unsupervised objective to guide the learning of useful higher level representations. Keywords: deep learning, unsupervised feature learning, deep belief networks, autoencoders, denoising

4 0.05047629 110 jmlr-2010-The SHOGUN Machine Learning Toolbox

Author: Sören Sonnenburg, Gunnar Rätsch, Sebastian Henschel, Christian Widmer, Jonas Behr, Alexander Zien, Fabio de Bona, Alexander Binder, Christian Gehl, Vojtěch Franc

Abstract: We have developed a machine learning toolbox, called SHOGUN, which is designed for uniﬁed large-scale learning for a broad range of feature types and learning settings. It offers a considerable number of machine learning models such as support vector machines, hidden Markov models, multiple kernel learning, linear discriminant analysis, and more. Most of the speciﬁc algorithms are able to deal with several different data classes. We have used this toolbox in several applications from computational biology, some of them coming with no less than 50 million training examples and others with 7 billion test examples. With more than a thousand installations worldwide, SHOGUN is already widely adopted in the machine learning community and beyond. SHOGUN is , implemented in C++ and interfaces to MATLABTM R, Octave, Python, and has a stand-alone command line interface. The source code is freely available under the GNU General Public License, Version 3 at http://www.shogun-toolbox.org. Keywords: support vector machines, kernels, large-scale learning, Python, Octave, R

5 0.048096843 79 jmlr-2010-Near-optimal Regret Bounds for Reinforcement Learning

Author: Thomas Jaksch, Ronald Ortner, Peter Auer

Abstract: For undiscounted reinforcement learning in Markov decision processes (MDPs) we consider the total regret of a learning algorithm with respect to an optimal policy. In order to describe the transition structure of an MDP we propose a new parameter: An MDP has diameter D if for any pair of states s, s′ there is a policy which moves from s to s′ in at most D steps (on average). √ ˜ We present a reinforcement learning algorithm with total regret O(DS AT ) after T steps for any unknown MDP with S states, A actions per state, and diameter D. A corresponding lower bound of √ Ω( DSAT ) on the total regret of any learning algorithm is given as well. These results are complemented by a sample complexity bound on the number of suboptimal steps taken by our algorithm. This bound can be used to achieve a (gap-dependent) regret bound that is logarithmic in T . Finally, we also consider a setting where the MDP is allowed to change a ﬁxed number of ℓ times. We present a modiﬁcation of our algorithm that is able to deal with this setting and show a √ ˜ regret bound of O(ℓ1/3 T 2/3 DS A). Keywords: undiscounted reinforcement learning, Markov decision process, regret, online learning, sample complexity

6 0.04491172 37 jmlr-2010-Evolving Static Representations for Task Transfer

7 0.038579747 118 jmlr-2010-libDAI: A Free and Open Source C++ Library for Discrete Approximate Inference in Graphical Models

8 0.037848089 39 jmlr-2010-FastInf: An Efficient Approximate Inference Library

9 0.035966899 4 jmlr-2010-A Generalized Path Integral Control Approach to Reinforcement Learning

10 0.033123538 2 jmlr-2010-A Convergent Online Single Time Scale Actor Critic Algorithm

11 0.028541744 8 jmlr-2010-A Surrogate Modeling and Adaptive Sampling Toolbox for Computer Based Design

12 0.025501516 116 jmlr-2010-WEKA−Experiences with a Java Open-Source Project

13 0.024576364 41 jmlr-2010-Gaussian Processes for Machine Learning (GPML) Toolbox

14 0.024554534 109 jmlr-2010-Stochastic Composite Likelihood

15 0.022748049 28 jmlr-2010-Continuous Time Bayesian Network Reasoning and Learning Engine

16 0.022619834 64 jmlr-2010-Learning Non-Stationary Dynamic Bayesian Networks

17 0.018301889 52 jmlr-2010-Incremental Sigmoid Belief Networks for Grammar Learning

18 0.018239219 89 jmlr-2010-PAC-Bayesian Analysis of Co-clustering and Beyond

19 0.017175037 48 jmlr-2010-How to Explain Individual Classification Decisions

20 0.01689712 62 jmlr-2010-Learning Gradients: Predictive Models that Infer Geometry and Statistical Dependence

similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, -0.082), (1, 0.018), (2, -0.07), (3, -0.032), (4, 0.022), (5, 0.076), (6, 0.051), (7, -0.105), (8, 0.15), (9, 0.164), (10, 0.039), (11, -0.108), (12, -0.039), (13, -0.064), (14, 0.067), (15, 0.115), (16, -0.004), (17, -0.077), (18, -0.058), (19, 0.008), (20, -0.096), (21, 0.007), (22, 0.049), (23, -0.059), (24, -0.099), (25, -0.033), (26, -0.147), (27, -0.008), (28, 0.177), (29, -0.089), (30, -0.203), (31, -0.031), (32, -0.036), (33, -0.189), (34, -0.001), (35, -0.117), (36, -0.049), (37, 0.087), (38, -0.18), (39, 0.033), (40, -0.078), (41, -0.095), (42, -0.054), (43, 0.122), (44, 0.05), (45, 0.177), (46, -0.034), (47, 0.131), (48, -0.029), (49, 0.06)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.97125185 93 jmlr-2010-PyBrain

Author: Tom Schaul, Justin Bayer, Daan Wierstra, Yi Sun, Martin Felder, Frank Sehnke, Thomas Rückstieß, Jürgen Schmidhuber

2 0.56733257 37 jmlr-2010-Evolving Static Representations for Task Transfer

Author: Phillip Verbancsics, Kenneth O. Stanley

Abstract: An important goal for machine learning is to transfer knowledge between tasks. For example, learning to play RoboCup Keepaway should contribute to learning the full game of RoboCup soccer. Previous approaches to transfer in Keepaway have focused on transforming the original representation to ﬁt the new task. In contrast, this paper explores the idea that transfer is most effective if the representation is designed to be the same even across different tasks. To demonstrate this point, a bird’s eye view (BEV) representation is introduced that can represent different tasks on the same two-dimensional map. For example, both the 3 vs. 2 and 4 vs. 3 Keepaway tasks can be represented on the same BEV. Yet the problem is that a raw two-dimensional map is high-dimensional and unstructured. This paper shows how this problem is addressed naturally by an idea from evolutionary computation called indirect encoding, which compresses the representation by exploiting its geometry. The result is that the BEV learns a Keepaway policy that transfers without further learning or manipulation. It also facilitates transferring knowledge learned in a different domain, Knight Joust, into Keepaway. Finally, the indirect encoding of the BEV means that its geometry can be changed without altering the solution. Thus static representations facilitate several kinds of transfer.

3 0.56062317 110 jmlr-2010-The SHOGUN Machine Learning Toolbox

Author: Sören Sonnenburg, Gunnar Rätsch, Sebastian Henschel, Christian Widmer, Jonas Behr, Alexander Zien, Fabio de Bona, Alexander Binder, Christian Gehl, Vojtěch Franc

4 0.36215433 117 jmlr-2010-Why Does Unsupervised Pre-training Help Deep Learning?

Author: Dumitru Erhan, Yoshua Bengio, Aaron Courville, Pierre-Antoine Manzagol, Pascal Vincent, Samy Bengio

5 0.25658506 107 jmlr-2010-Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion

Author: Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, Pierre-Antoine Manzagol

6 0.22852278 2 jmlr-2010-A Convergent Online Single Time Scale Actor Critic Algorithm

7 0.21489485 35 jmlr-2010-Error-Correcting Output Codes Library

8 0.19697817 8 jmlr-2010-A Surrogate Modeling and Adaptive Sampling Toolbox for Computer Based Design

9 0.17984954 109 jmlr-2010-Stochastic Composite Likelihood

10 0.17653878 118 jmlr-2010-libDAI: A Free and Open Source C++ Library for Discrete Approximate Inference in Graphical Models

11 0.17646489 39 jmlr-2010-FastInf: An Efficient Approximate Inference Library

12 0.17543778 77 jmlr-2010-Model-based Boosting 2.0

13 0.16076858 4 jmlr-2010-A Generalized Path Integral Control Approach to Reinforcement Learning

14 0.15902901 79 jmlr-2010-Near-optimal Regret Bounds for Reinforcement Learning

15 0.13791411 116 jmlr-2010-WEKA−Experiences with a Java Open-Source Project

16 0.1180839 41 jmlr-2010-Gaussian Processes for Machine Learning (GPML) Toolbox

17 0.11552599 108 jmlr-2010-Stochastic Complexity and Generalization Error of a Restricted Boltzmann Machine in Bayesian Estimation

18 0.11025656 115 jmlr-2010-Using Contextual Representations to Efficiently Learn Context-Free Languages

19 0.10297163 52 jmlr-2010-Incremental Sigmoid Belief Networks for Grammar Learning

20 0.10024789 28 jmlr-2010-Continuous Time Bayesian Network Reasoning and Learning Engine

similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(2, 0.018), (3, 0.01), (8, 0.012), (21, 0.011), (24, 0.013), (25, 0.559), (32, 0.016), (36, 0.027), (37, 0.056), (56, 0.016), (75, 0.103), (85, 0.042), (96, 0.03)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.76072049 93 jmlr-2010-PyBrain

Author: Tom Schaul, Justin Bayer, Daan Wierstra, Yi Sun, Martin Felder, Frank Sehnke, Thomas Rückstieß, Jürgen Schmidhuber

2 0.29739842 46 jmlr-2010-High Dimensional Inverse Covariance Matrix Estimation via Linear Programming

Author: Ming Yuan

Abstract: This paper considers the problem of estimating a high dimensional inverse covariance matrix that can be well approximated by “sparse” matrices. Taking advantage of the connection between multivariate linear regression and entries of the inverse covariance matrix, we propose an estimating procedure that can effectively exploit such “sparsity”. The proposed method can be computed using linear programming and therefore has the potential to be used in very high dimensional problems. Oracle inequalities are established for the estimation error in terms of several operator norms, showing that the method is adaptive to different types of sparsity of the problem. Keywords: covariance selection, Dantzig selector, Gaussian graphical model, inverse covariance matrix, Lasso, linear programming, oracle inequality, sparsity

3 0.22008485 103 jmlr-2010-Sparse Semi-supervised Learning Using Conjugate Functions

Author: Shiliang Sun, John Shawe-Taylor

Abstract: In this paper, we propose a general framework for sparse semi-supervised learning, which concerns using a small portion of unlabeled data and a few labeled data to represent target functions and thus has the merit of accelerating function evaluations when predicting the output of a new example. This framework makes use of Fenchel-Legendre conjugates to rewrite a convex insensitive loss involving a regularization with unlabeled data, and is applicable to a family of semi-supervised learning methods such as multi-view co-regularized least squares and single-view Laplacian support vector machines (SVMs). As an instantiation of this framework, we propose sparse multi-view SVMs which use a squared ε-insensitive loss. The resultant optimization is an inf-sup problem and the optimal solutions have arguably saddle-point properties. We present a globally optimal iterative algorithm to optimize the problem. We give the margin bound on the generalization error of the sparse multi-view SVMs, and derive the empirical Rademacher complexity for the induced function class. Experiments on artiﬁcial and real-world data show their effectiveness. We further give a sequential training approach to show their possibility and potential for uses in large-scale problems and provide encouraging experimental results indicating the efﬁcacy of the margin bound and empirical Rademacher complexity on characterizing the roles of unlabeled data for semi-supervised learning. Keywords: semi-supervised learning, Fenchel-Legendre conjugate, representer theorem, multiview regularization, support vector machine, statistical learning theory

4 0.21976863 118 jmlr-2010-libDAI: A Free and Open Source C++ Library for Discrete Approximate Inference in Graphical Models

Author: Joris M. Mooij

Abstract: This paper describes the software package libDAI, a free & open source C++ library that provides implementations of various exact and approximate inference methods for graphical models with discrete-valued variables. libDAI supports directed graphical models (Bayesian networks) as well as undirected ones (Markov random ﬁelds and factor graphs). It offers various approximations of the partition sum, marginal probability distributions and maximum probability states. Parameter learning is also supported. A feature comparison with other open source software packages for approximate inference is given. libDAI is licensed under the GPL v2+ license and is available at http://www.libdai.org. Keywords: probabilistic graphical models, approximate inference, open source software, factor graphs, Markov random ﬁelds, Bayesian networks

5 0.21864817 104 jmlr-2010-Sparse Spectrum Gaussian Process Regression

Author: Miguel Lázaro-Gredilla, Joaquin Quiñonero-Candela, Carl Edward Rasmussen, Aníbal R. Figueiras-Vidal

Abstract: We present a new sparse Gaussian Process (GP) model for regression. The key novel idea is to sparsify the spectral representation of the GP. This leads to a simple, practical algorithm for regression tasks. We compare the achievable trade-offs between predictive accuracy and computational requirements, and show that these are typically superior to existing state-of-the-art sparse approximations. We discuss both the weight space and function space representations, and note that the new construction implies priors over functions which are always stationary, and can approximate any covariance function in this class. Keywords: Gaussian process, probabilistic regression, sparse approximation, power spectrum, computational efﬁciency

6 0.21708632 91 jmlr-2010-Posterior Regularization for Structured Latent Variable Models

7 0.2168242 117 jmlr-2010-Why Does Unsupervised Pre-training Help Deep Learning?

8 0.21575986 42 jmlr-2010-Generalized Expectation Criteria for Semi-Supervised Learning with Weakly Labeled Data

9 0.2148423 57 jmlr-2010-Iterative Scaling and Coordinate Descent Methods for Maximum Entropy Models

10 0.214802 63 jmlr-2010-Learning Instance-Specific Predictive Models

11 0.21455584 74 jmlr-2010-Maximum Relative Margin and Data-Dependent Regularization

12 0.21244107 17 jmlr-2010-Bayesian Learning in Sparse Graphical Factor Models via Variational Mean-Field Annealing

13 0.21237136 89 jmlr-2010-PAC-Bayesian Analysis of Co-clustering and Beyond

14 0.21196537 87 jmlr-2010-Online Learning for Matrix Factorization and Sparse Coding

15 0.21153282 107 jmlr-2010-Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion

16 0.21149155 111 jmlr-2010-Topology Selection in Graphical Models of Autoregressive Processes

17 0.21132886 109 jmlr-2010-Stochastic Composite Likelihood

18 0.21105175 51 jmlr-2010-Importance Sampling for Continuous Time Bayesian Networks

19 0.21083462 59 jmlr-2010-Large Scale Online Learning of Image Similarity Through Ranking

20 0.21012044 66 jmlr-2010-Linear Algorithms for Online Multitask Classification