We consider the question "How should one act when the only goal is to learn as much as possible?" Building on the theoretical results of Fedorov [1972] and MacKay [1992], we apply techniques from Optimal Experiment Design (OED) to guide the query/action selection of a neural network learner. We demonstrate that these techniques allow the learner to minimize its generalization error by exploring its domain efficiently and completely.We conclude that, while not a panacea, OED-based query/action has muchto offer, especially in domains where its high computational costs can be tolerated.

HINE Parameters X0, X1, ....Xp Output E0, E1,....Ep Error Desired Output D0, D1,...Dp Y0, Y1,...Yp Input w w0 w1 AT&T Laboratories (c) COST FUNCTION Output E0, E1,....Ep Error Desired Output D0, D1,...Dp Y0, Y1,...Yp X0, X1, ....Xp Input Parameters w B R A COMPUTING THE GRADIENT WITH BACKPROPAGATION O = A(I1, I2) dI1 = dO ¶ A ¶ I1 dI2 = dO ¶ A ¶ I2 - The learning machine is composed of modules (e.g. layers) - Each module can do two things: 1- compute its outputs from its inputs (FPROP) 2- compute gradient vectors at its inputs from gradient vectors at its outputs (BPROP) A O, dO I1, dI1 I2, dI2 AT&T Laboratories (c) AN INTERESTING SPECIAL CASE: MULTILAYER NETWORKS X0, X1, ....Xp Output Desired Output D0, D1,...Dp Y0, Y1,...Yp Input || D - Y || 2 2 1 WX F() WX F() Mean Square Error Parameters (weights + biases) w Weight matrix E0, E1,....Ep Sigmoids + Biase

We apply Stochastic Meta-Descent (SMD), a stochastic gradient optimization method with gain vector adaptation, to the training of Conditional Random Fields (CRFs). On several large data sets, the resulting optimizer converges to the same quality of solution over an order of magnitude faster than limited-memory BFGS, the leading method reported to date. We report results for both exact and inexact inference techniques. 1.

Gain adaptation algorithms for neural networks typically adjust learning rates by monitoring the correlation between successive gradients. Here we discuss the limitations of this approach, and develop an alternative by extending Sutton's work on linear systems to the general, nonlinear case. The resulting online algorithms are computationally little more expensive than other acceleration techniques, do not assume statistical independence between successive training patterns, and do not require an arbitrary smoothing parameter. In our benchmark experiments, they consistently outperform other acceleration methods, and show remarkable robustness when faced with noni. i.d. sampling of the input space.

this paper (available on the World-Wide Web; see our home pages) contains pseudo-code of an efficient implementation. It is based on fast multiplication of the Hessian and a vector due to Pearlmutter (1994) and Mller (1993). Acknowledgments

Abstract. Markov logic networks (MLNs) combine Markov networks and first-order logic, and are a powerful and increasingly popular representation for statistical relational learning. The state-of-the-art method for discriminative learning of MLN weights is the voted perceptron algorithm, which is essentially gradient descent with an MPE approximation to the expected sufficient statistics (true clause counts). Unfortunately, these can vary widely between clauses, causing the learning problem to be highly ill-conditioned, and making gradient descent very slow. In this paper, we explore several alternatives, from per-weight learning rates to second-order methods. In particular, we focus on two approaches that avoid computing the partition function: diagonal Newton and scaled conjugate gradient. In experiments on standard SRL datasets, we obtain order-of-magnitude speedups, or more accurate models given comparable learning times. 1

We propose a generic method for iteratively approximating various second-order gradient steps -- Newton, Gauss-Newton, Levenberg-Marquardt, and natural gradient -- in linear time per iteration, using special curvature matrix-vector products that can be computed in O(n). Two recent acceleration techniques for online learning, matrix momentum and stochastic meta-descent (SMD), in fact implement this approach. Since both were originally derived by very different routes, this o ers fresh insight into their operation, resulting in further improvements to SMD.

. The calculation of second derivatives is required by recent training and analyses techniques of connectionist networks, such as the elimination of superfluous weights, and the estimation of confidence intervals both for weights and network outputs. We here review and develop exact and approximate algorithms for calculating second derivatives. For networks with jwj weights, simply writing the full matrix of second derivatives requires O(jwj 2 ) operations. For networks of radial basis units or sigmoid units, exact calculation of the necessary intermediate terms requires of the order of 2h + 2 backward/forward-propagation passes where h is the number of hidden units in the network. We also review and compare three approximations (ignoring some components of the second derivative, numerical differentiation, and scoring). Our algorithms apply to arbitrary activation functions, networks, and error functions (for instance, with connections that skip layers, or radial basis functions, or ...

Bayesian treatments of learning in neural networks are typically based either on a local Gaussian approximation to a mode of the posterior weight distribution, or on Markov chain Monte Carlo simulations. A third approach, called ensemble learning, was introduced by Hinton and van Camp (1993). It aims to approximate the posterior distribution by minimizing the Kullback-Leibler divergence between the true posterior and a parametric approximating distribution. The original derivation of a deterministic algorithm relied on the use of a Gaussian approximating distribution with a diagonal covariance matrix and hence was unable to capture the posterior correlations between parameters. In this chapter we show how the ensemble learning approach can be extended to full-covariance Gaussian distributions while remaining computationally tractable. We also extend the framework to deal with hyperparameters, leading to a simple re-estimation procedure. One of the benefits of our approach is that it yields a strict lower bound on the marginal likelihood, in contrast to other approximate procedures. 1

The main challenge of tracking articulated structures like hands is their large number of degrees of freedom (DOFs). A realistic 3D model of the human hand has at least 26 DOFs. The arsenal of tracking approaches that can track such structures fast and reliably is still very small. This paper proposes a tracker based on ‘Stochastic Meta-Descent ’ (SMD) for optimizations in such highdimensional state spaces. This new algorithm is based on a gradient descent approach with adaptive and parameter-specific step sizes. The SMD tracker facilitates the integration of constraints, and combined with a stochastic sampling technique, can get out of spurious local minima. Furthermore, the integration of a deformable hand model based on linear blend skinning and anthropometrical measurements reinforce the robustness of our tracker. Experiments show the efficiency of the SMD algorithm in comparison with common optimization methods.