In the visual system of primates, image information propagates across successive cortical areas, and there is also local feedback within an area and long-range feedback across areas. Recent findings suggest that the resulting temporal dynamics of neural activity are crucial in several vision tasks. In contrast, artificial neural network models of vision are typically feedforward and do not capitalize on the benefits of temporal dynamics, partly due to concerns about stability and computational costs. In this study, we focus on recurrent networks with feedback connections for visual tasks with static input corresponding to a single fixation. We demonstrate mathematically that a network's dynamics can be stabilized by four key features of biological networks: layer-ordered structure, temporal delays between layers, longer distance feedback across layers, and nonlinear neuronal responses. Conversely, when feedback has a fixed distance, one can omit delays in feedforward connections to achieve more efficient artificial implementations. We also evaluated the effect of feedback connections on object detection and classification performance using standard benchmarks, specifically the COCO and CIFAR10 datasets. Our findings indicate that feedback connections improved the detection of small objects, and classification performance became more robust to noise. We found that performance increased with the temporal dynamics, not unlike what is observed in core vision of primates. These results suggest that delays and layered organization are crucial features for stability and performance in both biological and artificial recurrent neural networks.

The sleeping brain exhibits characteristic slow-wave activity which decays over the course of the night. This decay is thought to result from homeostatic synaptic downscaling. Transcranial electrical stimulation can entrain slow-wave oscillations (SWO) in the human electro-encephalogram (EEG). A computational model of the underlying mechanism predicts that firing rates are predominantly increased during stimulation. Assuming that synaptic homeostasis is driven by average firing rates, we expected an acceleration of synaptic downscaling during stimulation, which is compensated by a reduced drive after stimulation. We show that 25 minutes of transcranial electrical stimulation, as predicted, reduced the decay of SWO in the remainder of the night. Anatomically accurate simulations of the field intensities on human cortex precisely matched the effect size in different EEG electrodes. Together these results suggest a mechanistic link between electrical stimulation and accelerated synaptic homeostasis in human sleep.

Our society's information technology advancements have resulted in the increasingly problematic issue of information overload—i.e., we have more access to information than we can possibly process. This is nowhere more apparent than in the volume of imagery and video that we can access on a daily basis—for the general public, availability of YouTube video and Google Images, or for the image analysis professional tasked with searching security video or satellite reconnaissance. Which images to look at and how to ensure we see the images that are of most interest to us, begs the question of whether there are smart ways to triage this volume of imagery. Over the past decade, computer vision research has focused on the issue of ranking and indexing imagery. However, computer vision is limited in its ability to identify interesting imagery, particularly as “interesting” might be defined by an individual. In this paper we describe our efforts in developing brain–computer interfaces (BCIs) which synergistically integrate computer vision and human vision so as to construct a system for image triage. Our approach exploits machine learning for real-time decoding of brain signals which are recorded noninvasively via electroencephalography (EEG). The signals we decode are specific for events related to imagery attracting a user's attention. We describe two architectures we have developed for this type of cortically coupled computer vision and discuss potential applications and challenges for the future.

A feedforward spiking network represents a nonlinear transformation that maps a set of input spikes to a set of output spikes. This mapping transforms the joint probability distribution of incoming spikes into a joint distribution of output spikes. We present an algorithm for synaptic adaptation that aims to maximize the entropy of this output distribution, thereby creating a model for the joint distribution of the incoming point processes. The learning rule that is derived depends on the precise pre- and postsynaptic spike timings. When trained on correlated spike trains, the network learns to extract independent spike trains, thereby uncovering the underlying statistical structure and creating a more efficient representation of the incoming spike trains.

How does one extract unknown but stereotypical events that are linearly superimposed within a signal with variable latencies and variable amplitudes? One could think of using template matching or matching pursuit to find the arbitrarily shifted linear components. However, traditional matching approaches require that the templates be known a priori. To overcome this restriction we use instead semi Non-Negative Matrix Factorization (semi-NMF) that we extend to allow for time shifts when matching the templates to the signal. The algorithm estimates templates directly from the data along with their non-negative amplitudes. The resulting method can be thought of as an adaptive template matching procedure. We demonstrate the procedure on the task of extracting spikes from single channel extracellular recordings. On these data the algorithm essentially performs spike detection and unsupervised spike clustering. Results on simulated data and extracellular recordings indicate that the method performs well for signal-to-noise ratios of 6dB or higher and that spike templates are recovered accurately provided they are sufficiently different.

Traditional analysis methods for single-trial classification of electro-encephalography (EEG) focus on two types of paradigms: phase locked methods, in which the amplitude of the signal is used as the feature for classification, i.e. event related potentials; and second order methods, in which the feature of interest is the power of the signal, i.e event related (de)synchronization. The process of deciding which paradigm to use is ad hoc and is driven by knowledge of neurological findings. Here we propose a unified method in which the algorithm learns the best first and second order spatial and temporal features for classification of EEG based on a bilinear model. The efficiency of the method is demonstrated in simulated and real EEG from a benchmark data set for Brain Computer Interface.

Event-related potentials (ERPs) recorded at the scalp are indicators of brain activity associated with event-related information processing; hence they may be suitable for the assessment of changes in cognitive processing load. While the measurement of ERPs in a laboratory setting and classifying those ERPs is trivial, such a task presents major challenges in a "real world" setting where the EEG signals are recorded when subjects freely move their eyes and the sensory inputs are continuously, as opposed to discretely presented. Here we demonstrate that with the aid of second-order blind identification (SOBI), a blind source separation (BSS) algorithm: (1) we can extract ERPs from such challenging data sets; (2) we were able to obtain meaningful single-trial ERPs in addition to averaged ERPs; and (3) we were able to estimate the spatial origins of these ERPs. Finally, using back-propagation neural networks as classifiers, we show that these single-trial ERPs from specific brain regions can be used to determine moment-to-moment changes in cognitive processing load during a complex "real world" task.

Probabilistic models of image statistics underlie many approaches in image analysis and processing. An important class of such models have variables whose dependency graph is a tree. If the hidden variables take values on a finite set, most computations with the model can be performed exactly, including the likelihood calculation, training with the EM algorithm, etc. Crouse et al. developed one such model, the hidden Markov tree (HMT). They took particular care to limit the complexity of their model. We argue that it is beneficial to allow more complex tree-structured models, describe the use of information theoretic penalties to choose the model complexity, and present experimental results to support these proposals. For these experiments, we use what we call the hierarchical image probability (HIP) model. The differences between the HIP and the HMT models include the use of multivariate Gaussians to model the distributions of local vectors of wavelet coefficients and the use of different numbers of hidden states at each resolution. We demonstrate the broad utility of image distributions by applying the HIP model to classification, synthesis, and compression, across a variety of image types, namely, electrooptical, synthetic aperture radar, and mammograms (digitized X-rays). In all cases, we compare with the HMT.

We use unsupervised probabilistic machine learning ideas to try to ex- plain the kinds of learning observed in real neurons, the goal being to connect abstract principles of self-organisation to known biophysi- cal processes. For example, we would like to explain Spike Timing- Dependent Plasticity (see [5,6] and Figure 3A), in terms of information theory. Starting out, we explore the optimisation of a network sensitiv- ity measure related to maximising the mutual information between input spike timings and output spike timings. Our derivations are analogous to those in ICA, except that the sensitivity of output timings to input tim- ings is maximised, rather than the sensitivity of output ‘firing rates’ to inputs. ICA and related approaches have been successful in explaining the learning of many properties of early visual receptive fields in rate cod- ing models, and we are hoping for similar gains in understanding of spike coding in networks, and how this is supported, in principled probabilistic ways, by cellular biophysical processes. For now, in our initial simula- tions, we show that our derived rule can learn synaptic weights which can unmix, or demultiplex, mixed spike trains. That is, it can recover inde- pendent point processes embedded in distributed correlated input spike trains, using an adaptive single-layer feedforward spiking network. 1 Maximising Sensitivity. In this section, we will follow the structure of the ICA derivation [4] in developing the spiking theory. We cannot claim, as before, that this gives us an information maximisation algorithm, for reasons that we will delay addressing until Section 3. But for now, to first develop our approach, we will explore an interim objective function called sensitivity which we define as the log Jacobian of how input spike timings affect output spike timings. 1.1 How to maximise the effect of one spike timing on another. Consider a spike in neuron j at time tl that has an effect on the timing of another spike in neuron i at time tk. The neurons are connected by a weight wij. We use i and j to index neurons, and k and l to index spikes, but sometimes for convenience we will use spike indices in place of neuron indices. For example, wkl, the weight between an input spike l and an output spike k, is naturally understood to be just the corresponding wij.

We present an algorithm for blindly recovering constituent source spectra from magnetic resonance (MR) chemical shift imaging (CSI) of the human brain. The algorithm, which we call constrained nonnegative matrix factorization (cNMF), does not enforce independence or sparsity, instead only requiring the source and mixing matrices to be nonnegative. It is based on the nonnegative matrix factorization (NMF) algorithm, extending it to include a constraint on the positivity of the amplitudes of the recovered spectra. This constraint enables recovery of physically meaningful spectra even in the presence of noise that causes a significant number of the observation amplitudes to be negative. We demonstrate and characterize the algorithm's performance using 31P volumetric brain data, comparing the results with two different blind source separation methods: Bayesian spectral decomposition (BSD) and nonnegative sparse coding (NNSC). We then incorporate the cNMF algorithm into a hierarchical decomposition framework, showing that it can be used to recover tissue-specific spectra given a processing hierarchy that proceeds coarse-to-fine. We demonstrate the hierarchical procedure on 1H brain data and conclude that the computational efficiency of the algorithm makes it well-suited for use in diagnostic work-up.

Convolutive blind source separation and adaptive beamforming have a similar goal-extracting a source of interest (or multiple sources) while reducing undesired interferences. A benefit of source separation is that it overcomes the conventional cross-talk or leakage problem of adaptive beamforming. Beamforming on the other hand exploits geometric information which is often readily available but not utilized in blind algorithms. We propose to join these benefits by combining cross-power minimization of second-order source separation with geometric linear constraints used in adaptive beamforming. We find that the geometric constraints resolve some of the ambiguities inherent in the independence criterion such as frequency permutations and degrees of freedom provided by additional sensors. We demonstrate the new method in performance comparisons for actual room recordings of two and three simultaneous acoustic sources.

We formulate a model for probability distributions on image spaces. We show that any distribution of images can be factored exactly into conditional distributions of feature vectors at one resolution (pyramid level) conditioned on the image information at lower resolutions. We would like to factor this over positions in the pyramid levels to make it tractable, but such factoring may miss long-range dependencies. To capture long-range dependencies, we introduce hidden class labels at each pixel in the pyramid. The result is a hierarchical mixture of conditional probabilities, similar to a hidden Markov model on a tree. The model parameters can be found with maximum likelihood estimation using the EM algorithm. We have obtained encouraging preliminary results on the problems of detecting various objects in SAR images and target recognition in optical aerial images.

We present evidence that several higher-order statistical proper(cid:173) ties of natural images and signals can be explained by a stochastic model which simply varies scale of an otherwise stationary Gaus(cid:173) sian process. We discuss two interesting consequences. The first is that a variety of natural signals can be related through a com(cid:173) mon model of 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 cas(cid:173) es the non-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.

Acoustic signals recorded simultaneously in a reverberant environment can be described as sums of differently convolved sources. The task of source separation is to identify the multiple channels and possibly to invert those in order to obtain estimates of the underlying sources. We tackle the problem by explicitly exploiting the nonstationarity of the acoustic sources. Changing cross correlations at multiple times give a sufficient set of constraints for the unknown channels. A least squares optimization allows us to estimate a forward model, identifying thus the multipath channel. In the same manner we can find an FIR backward model, which generates well separated model sources. Furthermore, for more than three channels we have sufficient conditions to estimate underlying additive sensor noise powers. We show the good performance in a real room environments and demonstrate the algorithm's utility for automatic speech recognition.

Using a theory of list-mode maximum-likelihood (ML) source reconstruction presented recently by Barrett et al., this paper formulates a corresponding expectation-maximization (EM) algorithm, as well as a method for estimating noise properties at the ML estimate. List-mode ML is of interest in cases where the dimensionality of the measurement space impedes a binning of the measurement data. It can be advantageous in cases where a better forward model can be obtained by including more measurement coordinates provided by a given detector. Different figures of merit for the detector performance can be computed from the Fisher information matrix (FIM). This paper uses the observed FIM, which requires a single data set, thus, avoiding costly ensemble statistics. The proposed techniques are demonstrated for an idealized two-dimensional (2-D) positron emission tomography (PET) [2-D PET] detector. We compute from simulation data the improved image quality obtained by including the time of flight of the coincident quanta.

According to Barlow (1989), feature extraction can be understood as finding a statistically independent representation of the probability distribution underlying the measured signals. The search for a statistically independent representation can be formulated by the criterion of minimal mutual information, which reduces to decorrelation in the case of gaussian distributions. If nongaussian distributions are to be considered, minimal mutual information is the appropriate generalization of decorrelation as used in linear Principal Component Analyses (PCA). We also generalize to nonlinear transformations by only demanding perfect transmission of information. This leads to a general class of nonlinear transformations, namely symplectic maps. Conservation of information allows us to consider only the statistics of single coordinates. The resulting factorial representation of the joint probability distribution gives a density estimation. We apply this concept to the real world problem of electrical motor fault detection treated as a novelty detection task.