Goto

Collaborating Authors

 Industry


Overfitting in Neural Nets: Backpropagation, Conjugate Gradient, and Early Stopping

Neural Information Processing Systems

The conventional wisdom is that backprop nets with excess hidden units generalize poorly. We show that nets with excess capacity generalize well when trained with backprop and early stopping. Experiments suggest tworeasons for this: 1) Overfitting can vary significantly in different regions of the model. Excess capacity allows better fit to regions of high non-linearity, and backprop often avoids overfitting the regions of low non-linearity.


Text Classification using String Kernels

Neural Information Processing Systems

HUlna Lodhi John Shawe-Taylor N ello Cristianini Chris Watkins Department of Computer Science Royal Holloway, University of London Egham, Surrey TW20 OEX, UK {huma, john, nello, chrisw}Cdcs.rhbnc.ac.uk Abstract We introduce a novel kernel for comparing two text documents. The kernel is an inner product in the feature space consisting of all subsequences of length k. A subsequence is any ordered sequence ofk characters occurring in the text though not necessarily contiguously. The subsequences are weighted by an exponentially decaying factor of their full length in the text, hence emphasising those occurrences which are close to contiguous. A direct computation ofthis feature vector would involve a prohibitive amount of computation even for modest values of k, since the dimension of the feature space grows exponentially with k. The paper describes how despite this fact the inner product can be efficiently evaluated by a dynamic programming technique.


Reinforcement Learning with Function Approximation Converges to a Region

Neural Information Processing Systems

Many algorithms for approximate reinforcement learning are not known to converge. In fact, there are counterexamples showing that the adjustable weights in some algorithms may oscillate within a region rather than converging to a point. This paper shows that, for two popular algorithms, such oscillation is the worst that can happen: the weights cannot diverge, but instead must converge to a bounded region. The algorithms are SARSA(O) and V(O); the latter algorithm was used in the well-known TD-Gammon program. 1 Introduction Although there are convergent online algorithms (such as TD()') [1]) for learning the parameters of a linear approximation to the value function of a Markov process, no way is known to extend these convergence proofs to the task of online approximation ofeither the state-value (V*) or the action-value (Q*) function of a general Markov decision process. In fact, there are known counterexamples to many proposed algorithms.For example, fitted value iteration can diverge even for Markov processes [2]; Q-Iearning with linear function approximators can diverge, even when the states are updated according to a fixed update policy [3]; and SARSA(O) can oscillate between multiple policies with different value functions [4].



Bayesian Video Shot Segmentation

Neural Information Processing Systems

Prior knowledge about video structure can be used both as a means to improve the peiformance of content analysis and to extract features that allow semantic classification. We introduce statistical models for two important components of this structure, shot duration and activity, and demonstrate the usefulness of these models by introducing a Bayesian formulation for the shot segmentation problem. The new formulations is shown to extend standard thresholding methods in an adaptive and intuitive way, leading to improved segmentation accuracy.


Higher-Order Statistical Properties Arising from the Non-Stationarity of Natural Signals

Neural Information Processing Systems

The first is that a variety of natural signals can be related through a common modelof spherically invariant random processes, which have the attractive property that the joint densities can be constructed from the one dimensional marginal. The second is that in some cases thenon-stationarity assumption and only second order methods can be explicitly exploited to find a linear basis that is equivalent to independent components obtained with higher-order methods. This is demonstrated on spectro-temporal components of speech. 1 Introduction Recently, considerable attention has been paid to understanding and modeling the non-Gaussian or "higher-order" properties of natural signals, particularly images. Several non-Gaussian properties have been identified and studied. For example, marginal densities of features have been shown to have high kurtosis or "heavy tails", indicating a non-Gaussian, sparse representation. Another example is the "bowtie" shape of conditional distributions of neighboring features, indicating dependence ofvariances [11]. These non-Gaussian properties have motivated a number of image and signal processing algorithms that attempt to exploit higher-order s tatistics of the signals, e.g., for blind source separation. In this paper we show that these previously observed higher-order phenomena are ubiquitous and can be accounted for by a model which simply varies the scale of an otherwise stationary Gaussianprocess. This enables us to relate a variety of natural signals to one another and to spherically invariant random processes, which are well-known in the signal processing literature [6, 3].


Competition and Arbors in Ocular Dominance

Neural Information Processing Systems

Hebbian and competitive Hebbian algorithms are almost ubiquitous in modeling pattern formation in cortical development. We analyse in theoretical detaila particular model (adapted from Piepenbrock & Obermayer, 1999) for the development of Id stripe-like patterns, which places competitive and interactive cortical influences, and free and restricted initial arborisationonto a common footing. 1 Introduction Cats, many species of monkeys, and humans exibit ocular dominance stripes, which are alternating areas of primary visual cortex devoted to input from (the thalamic relay associated with)just one or the other eye (see Erwin et aI, 1995; Miller, 1996; Swindale, 1996 for reviews of theory and data). These well-known fingerprint patterns have been a seductive targetfor models of cortical pattern formation because of the mix of competition and cooperation they suggest. A wealth of synaptic adaptation algorithms has been suggested to account for them (and also the concomitant refinement of the topography of the map between the eyes and the cortex), many of which are based on forms of Hebbian learning. Critical issues for the models are the degree of correlation between inputs from the eyes, the nature of the initial arborisation of the axonal inputs, the degree and form of cortical competition, and the nature of synaptic saturation (preventing weights from changing sign or getting too large) and normalisation (allowing cortical and/or thalamic cells to support only a certain total synaptic weight).


Generalized Belief Propagation

Neural Information Processing Systems

Belief propagation (BP) was only supposed to work for treelike networks but works surprisingly well in many applications involving networks with loops, including turbo codes. However, there has been little understanding of the algorithm or the nature of the solutions it finds for general graphs. We show that BP can only converge to a stationary point of an approximate free energy, known as the Bethe free energy in statistical physics.This result characterizes BP fixed-points and makes connections with variational approaches to approximate inference. More importantly, our analysis lets us build on the progress made in statistical physics since Bethe's approximation was introduced in 1935. Kikuchi and others have shown how to construct more accurate freeenergy approximations, of which Bethe's approximation is the simplest.


Smart Vision Chip Fabricated Using Three Dimensional Integration Technology

Neural Information Processing Systems

H.Kurino, M.Nakagawa, K.W.Lee, T.Nakamura, Y.Yamada, K.T.Park and M.Koyanagi Dept. of Machine Intelligence and Systems Engineering, Tohoku University 01, Aza-Aramaki, Aoba-ku, Sendai 980-8579, Japan kurino@sd.mech.tohoku.ac.jp Abstract The smart VISIOn chip has a large potential for application in general purpose high speed image processing systems. In order to fabricate smart vision chips including photo detector compactly, we have proposed the application of three dimensional LSI technology for smart vision chips. Three dimensional technology has great potential to realize new neuromorphic systems inspired by not only the biological function but also the biological structure. In this paper, we describe our three dimensional LSI technology for neuromorphic circuits and the design of smart vision chips. 1 Introduction Recently, the demand for very fast image processing systems with real time operation capability has significantly increased. Conventional image processing systems based on the system level integration of a camera and a digital processor, do not have the potential for application in general purpose consumer electronic products.


Dopamine Bonuses

Neural Information Processing Systems

Substantial data support a temporal difference (TO) model of dopamine (OA) neuron activity in which the cells provide a global error signal for reinforcement learning. However, in certain circumstances, OAactivity seems anomalous under the TO model, responding to non-rewarding stimuli. We address these anomalies bysuggesting that OA cells multiplex information about reward bonuses,including Sutton's exploration bonuses and Ng et al's non-distorting shaping bonuses. We interpret this additional role for OA in terms of the unconditional attentional and psychomotor effectsof dopamine, having the computational role of guiding exploration. 1 Introduction Much evidence suggests that dopamine cells in the primate midbrain play an important rolein reward and action learning. Electrophysiological studies support a theory that OA cells signal a global prediction error for summed future reward in appetitive conditioning tasks (Montague et al, 1996; Schultz et al, 1997), in the form of a temporal difference prediction error term.