Decoding visual information from a population of retinal ganglion nal ganglion cells in transmitting visual information to the cells. J. Neurophysiol. 78: 2336–2350, 1997. This work investigates brain. How do the spike trains of optic nerve fibers convey how a time-dependent visual stimulus is encoded by the collective the visual scene projected on the retina? At this stage of the activity of many retinal ganglion cells. Multiple ganglion cell spike visual system, questions regarding the neural code can be trains were recorded simultaneously from the isolated retina of the tiger salamander using a multielectrode array. The stimulus consisted phrased and answered particularly precisely for the followof photopic, spatially uniform, temporally broadband flicker. From ing reasons: the ganglion cells are the only neurons transmit-the recorded spike trains, an estimate was obtained of the stimulus ting visual information to the brain; the only variable they intensity as a function of time. This was compared with the actual encode is the time-varying image on the retina; this stimulus stimulus to assess the quality and quantity of visual information con- can be controlled experimentally using well-developed tech-veyed by the ganglion cell population. Two algorithms were used to nology for generating images; finally, the activity of multiple decode the spike trains: an optimized linear filter in which each action retinal ganglion cells can be monitored experimentally withpotential

Abstract. We propose a framework in which anonymity protocols are interpreted as particular kinds of channels, and the degree of anonymity provided by the protocol as the converse of the channel’s capacity. We also investigate how the adversary can test the system to try to infer the user’s identity, and we study how his probability of success depends on the characteristics of the channel. We then illustrate how various notions of anonymity can be expressed in this framework, and show the relation with some definitions of probabilistic anonymity in literature. 1

Abstract. We describe a technique for comparing distributions without the need for density estimation as an intermediate step. Our approach relies on mapping the distributions into a reproducing kernel Hilbert space. Applications of this technique can be found in two-sample tests, which are used for determining whether two sets of observations arise from the same distribution, covariate shift correction, local learning, measures of independence, and density estimation. Kernel methods are widely used in supervised learning [1, 2, 3, 4], however they are much less established in the areas of testing, estimation, and analysis of probability distributions, where information theoretic approaches [5, 6] have long been dominant. Recent examples include [7] in the context of construction of graphical models, [8] in the context of feature extraction, and [9] in the context of independent component analysis. These methods have by and large a common issue: to compute quantities such as the mutual information, entropy, or Kullback-Leibler divergence, we require sophisticated space partitioning and/or

It is not clear how information related to cognitive or psychological processes is carried by or represented in the responses of single neurons. One provocative proposal is that precisely timed spike patterns play a role in carrying such information. This would require that these spike patterns have the potential for carrying information that would not be available from other measures such as spike count or latency. We examined exactly timed (1-ms precision) triplets and quadruplets of spikes in the stimulus-elicited responses of lateral geniculate nucleus (LGN) and primary visual cortex (V1) neurons of the awake fixating rhesus monkey. Large numbers of these precisely timed spike patterns were found. Information theoretical analysis showed that the precisely timed spike patterns carried only information already available from spike count, suggesting that the number of precisely timed spike

Abstract. This paper puts in consideration the concepts of security and robustness in watermarking, in order to be able to establish a clear frontier between them. A new information-theoretic framework to study data-hiding and watermarking security is proposed, using the mutual information to quantify the information about the secret key that leaks from the observation of watermarked objects. This framework is applied to the analysis of a Spread-Spectrum data-hiding scheme in different scenarios. Finally, we show some interesting links between a measure proposed in previous works in the literature, which is based on Fisher Information Matrix, and our proposed measure. 1

Abstract. We consider the issue of query performance, and we propose a novel method for automatically predicting the difficulty of a query. Unlike a number of existing techniques which are based on examining the ranked lists returned in response to perturbed versions of the query with respect to the given collection or perturbed versions of the collection with respect to the given query, our technique is based on examining the ranked lists returned by multiple scoring functions (retrieval engines) with respect to the given query and collection. In essence, we propose that the results returned by multiple retrieval engines will be relatively similar for “easy ” queries but more diverse for “difficult ” queries. By appropriately employing Jensen-Shannon divergence to measure the “diversity ” of the returned results, we demonstrate a methodology for predicting query difficulty whose performance exceeds existing state-ofthe-art techniques on TREC collections, often remarkably so. 1

Abstract. There is a huge diversity of definitions of “visually meaningful” image segments, ranging from simple uniformly colored segments, textured segments, through symmetric patterns, and up to complex semantically meaningful objects. This diversity has led to a wide range of different approaches for image segmentation. In this paper we present a single unified framework for addressing this problem – “Segmentation by Composition”. We define a good image segment as one which can be easily composed using its own pieces, but is difficult to compose using pieces from other parts of the image. This non-parametric approach captures a large diversity of segment types, yet requires no pre-definition or modelling of segment types, nor prior training. Based on this definition, we develop a segment extraction algorithm – i.e., given a single point-ofinterest, provide the “best ” image segment containing that point. This induces a figure-ground image segmentation, which applies to a range of different segmentation tasks: single image segmentation, simultaneous co-segmentation of several images, and class-based segmentations. 1

The portfolio selection problem is clearly one of the most fundamental problems in the field of computational finance. Given a set of say m stocks (one of which may be "cash"), the natural online problem is to determine a portfolio for the ith trading period based on the sequence of prices (or equivalently relative prices) for the preceding i \Gamma 1 trading periods. There has been both a growing interest and a growing skepticism concerning the value of a competitive theory of online portfolio selection algorithms. Competitive analysis is based on a worst case perspective and such a perspective is inconsistent with the more widely accepted analyses and theories based on statistical assumptions. The competitive framework does (perhaps surprisingly) permit non trivial upper bounds on relative performance against CBAL-OPT, an optimal offline constant rebalancing portfolio. Perhaps more impressive are some preliminary experimental results showing that certain algorithms that enjoy "respectable" competitive (i.e. worst case) performance also seem to perform quite well on historical sequences of data. These algorithms and the emerging competitive theory are directly related to studies in information theory and computational learning theory and indeed some of these algorithms have been pioneered within the information theory and computational learning communities. One goal of this paper is to try to better understand the extent to which competitive portfolio algorithms are indeed "learning". In doing so we discuss some simple strategies which can adapt to the data sequence. We present a mixture of both theoretical and experimental results. We also present a more inclusive study of the performance of existing and new algorithms with respect to a standard ...

Abstract. Active selection of good training examples is an important approach to reducing data-collection costs in machine learning; however, most existing methods focus on maximizing classification accuracy. In many applications, such as those with unequal misclassification costs, producing good class probability estimates (CPEs) is more important than optimizing classification accuracy. We introduce novel approaches to active learning based on the algorithms Bootstrap-LV and ACTIVEDECORATE, by using Jensen-Shannon divergence (a similarity measure for probability distributions) to improve sample selection for optimizing CPEs. Comprehensive experimental results demonstrate the benefits of our approaches. 1

In this paper a novel theoretical security analysis will be presented for data hiding methods with side-information, based on Costa's dirty paper scheme. We quantify the information about the secret key that leaks from the observation of watermarked signals, using the mutual information as analytic tool for providing a fair comparison between the original Costa's scheme, Distortion Compensated - Dither Modulation and Spread Spectrum.