From olfaction to vision and audition, there is an increasing need, and a growing number of experiments [1]-[8] that study responses of sensory neurons to natural stimuli. Natural stimuli have specific statistical properties [9, 10], and therefore sample only a subspace of all possible spatial and temporal frequencies explored during stimulation with white noise. Observing the full dynamic range of neural responses may require using stimulus ensembles which approximate those occurring in nature, and it is an attractive hypothesis that the neural representation of these natural signals may be optimized in some way. Finally, some neuron responses are strongly nonlinear and adaptive, and may not be predicted from a combination of responses to simple stimuli. It has also been shown that the variability in neural response decreases substantially when dynamical, rather than static, stimuli are used [11, 12]. For all these reasons, it would be attractive to have a rigorous method of analyzing neural responses to complex, naturalistic inputs.

Model selection is linked to model assessment, which is the problem of comparing different models, or model parameters, for a specific learning task. For supervised learning, the standard practical technique is crossvalidation, which is not applicable for semi-supervised and unsupervised settings. In this paper, a new model assessment scheme is introduced which is based on a notion of stability. The stability measure yields an upper bound to cross-validation in the supervised case, but extends to semi-supervised and unsupervised problems. In the experimental part, the performance of the stability measure is studied for model order selection in comparison to standard techniques in this area.

Psychophysical data suggest that temporal modulations of stimulus amplitude envelopes play a prominent role in the perceptual segregation of concurrent sounds. In particular, the detection of an unmodulated signal can be significantly improved by adding amplitude modulation to the spectral envelope of a competing masking noise. This perceptual phenomenon is known as "Comodulation Masking Release" (CMR). Despite the obvious influence of temporal structure on the perception of complex auditory scenes, the physiological mechanisms that contribute to CMR and auditory streaming are not well known. A recent physiological study by Nelken and colleagues has demonstrated an enhanced cortical representation of auditory signals in modulated noise. Our study evaluates these CMR-like response patterns from the perspective of a hypothetical auditory edge-detection neuron. It is shown that this simple neural model for the detection of amplitude transients can reproduce not only the physiological data of Nelken et al., but also, in light of previous results, a variety of physiological and psychoacoustical phenomena that are related to the perceptual segregation of concurrent sounds.

Prototypes based algorithms are commonly used to reduce the computational complexityof Nearest-Neighbour (NN) classifiers. In this paper we discuss theoretical and algorithmical aspects of such algorithms. On the theory side, we present margin based generalization bounds that suggest thatthese kinds of classifiers can be more accurate then the 1-NN rule. Furthermore, we derived a training algorithm that selects a good set of prototypes using large margin principles. We also show that the 20 years old Learning Vector Quantization (LVQ) algorithm emerges naturally fromour framework.

Behavioral goals are achieved reliably and repeatedly with movements rarely reproducible in their detail. Here we offer an explanation: we show that not only are variability and goal achievement compatible, but indeed that allowing variability in redundant dimensions is the optimal control strategy in the face of uncertainty. The optimal feedback control laws for typical motor tasks obey a "minimal intervention" principle: deviations from the average trajectory are only corrected when they interfere with the task goals. The resulting behavior exhibits task-constrained variability, as well as synergetic coupling among actuators--which is another unexplained empirical phenomenon.

We focus on the problem of efficient learning of dependency trees. It is well-known that given the pairwise mutual information coefficients, a minimum-weight spanning tree algorithm solves this problem exactly and in polynomial time. However, for large data-sets it is the construction of the correlation matrix that dominates the running time. We have developed a new spanning-tree algorithm which is capable of exploiting partial knowledge about edge weights. The partial knowledge we maintain is a probabilistic confidence interval on the coefficients, which we derive by examining just a small sample of the data. The algorithm is able to flag the need to shrink an interval, which translates to inspection of more data for the particular attribute pair. Experimental results show running time that is near-constant in the number of records, without significant loss in accuracy of the generated trees. Interestingly, our spanning-tree algorithm is based solely on Tarjan's red-edge rule, which is generally considered a guaranteed recipe for bad performance.

We describe a new algorithmic framework for learning multiclass categorization problems. In this framework a multiclass predictor is composed of a pair of embeddings that map both instances and labels into a common space. In this space each instance is assigned the label it is nearest to. We outline and analyze an algorithm, termed Bunching, for learning the pair of embeddings from labeled data. A key construction in the analysis of the algorithm is the notion of probabilistic output codes, a generalization of error correcting output codes (ECOC). Furthermore, the method of multiclass categorization using ECOC is shown to be an instance of Bunching. We demonstrate the advantage of Bunching over ECOC by comparing their performance on numerous categorization problems.

We develop a systems theoretical treatment of a behavioural system that interacts with its environment in a closed loop situation such that its motor actions influence its sensor inputs. The simplest form of a feedback is a reflex. Reflexes occur always "too late"; i.e., only after a (unpleasant, painful, dangerous) reflex-eliciting sensor event has occurred. This defines an objective problem which can be solved if another sensor input exists which can predict the primary reflex and can generate an earlier reaction. In contrast to previous approaches, our linear learning algorithm allows for an analytical proof that this system learns to apply feedforward control with the result that slow feedback loops are replaced by their equivalent feed-forward controller creating a forward model. In other words, learning turns the reactive system into a proactive system. By means of a robot implementation we demonstrate the applicability of the theoretical results which can be used in a variety of different areas in physics and engineering.

We consider loopy belief propagation for approximate inference in probabilistic graphical models. A limitation of the standard algorithm is that clique marginals are computed as if there were no loops in the graph. To overcome this limitation, we introduce fractional belief propagation. Fractional belief propagation is formulated in terms of a family of approximate free energies, which includes the Bethe free energy and the naive mean-field free as special cases. Using the linear response correction of the clique marginals, the scale parameters can be tuned. Simulation results illustrate the potential merits of the approach.

The problem of extracting the relevant aspects of data, in face of multiple conflicting structures, is inherent to modeling of complex data. Extracting structure in one random variable that is relevant for another variable has been principally addressed recently via the information bottleneck method [15]. However, such auxiliary variables often contain more information than is actually required due to structures that are irrelevant for the task. In many other cases it is in fact easier to specify what is irrelevant than what is, for the task at hand. Identifying the relevant structures, however, can thus be considerably improved by also minimizing the information about another, irrelevant, variable. In this paper we give a general formulation of this problem and derive its formal, as well as algorithmic, solution. Its operation is demonstrated in a synthetic example and in two real world problems in the context of text categorization and face images. While the original information bottleneck problem is related to rate distortion theory, with the distortion measure replaced by the relevant information, extracting relevant features while removing irrelevant ones is related to rate distortion with side information.