We present a system for recognizing human faces from single images out of a large database containing one image per person. The task is difficult because of image variation in terms of position, size, expression, and pose. The system collapses most of this variance by extracting concise face descriptions in the form of image graphs. In these, fiducial points on the face (eyes, mouth, etc.) are described by sets of wavelet components (jets). Image graph extraction is based on a novel approach, the bunch graph, which is constructed from a small set of sample image graphs. Recognition is based on a straightforward comparison of image graphs. We report recognition experiments on the FERET database as well as the Bochum database, including recognition across pose.

AbstractöTo investigate the basis of the £uctuating activity present in neocortical neurons in vivo, we have combined computational models with whole-cell recordings using the dynamic-clamp technique. A simpli¢ed`point-conductance' model was used to represent the currents generated by thousands of stochastically releasing synapses. Synaptic activity was represented by two independent fast glutamatergic and GABAergic conductances described by stochastic randomwalk processes. An advantage of this approach is that all the model parameters can be determined from voltage-clamp experiments. We show that the point-conductance model captures the amplitude and spectral characteristics of the synaptic conductances during background activity. To determine if it can recreate in vivo-like activity, we injected this point-conductance model into a single-compartment model, or in rat prefrontal cortical neurons in vitro using dynamic clamp. This procedure successfully recreated several properties of neurons intracellularly recorded in vivo, such as a depolarized membrane potential, the presence of high-amplitude membrane potential £uctuations, a low-input resistance and irregular spontaneous ¢ring activity. In addition, the point-conductance model could simulate the enhancement of responsiveness due to background activity.We conclude that many of the characteristics of cortical neurons in vivo can be explained by fast glutamatergic and GABAergic conductances varying stochastically. ß 2001 IBRO. Published by Elsevier Science Ltd. All rights reserved.Key words: computational models, pyramidal neurons, dynamic clamp, synaptic bombardment, high-conductance states, CV.Synaptic background activity is invariably present in intracellular recordings of neocortical neurons in vivo, and modeling studies have suggested that it may have important consequences on the integrative properties of these neurons (Barrett, 1975;Holmes and Woody, 1989;Bernander et al., 1991;Destexhe and Parë, 1999). Background activity is maximal during the active states of the brain, when cortical neurons ¢re spontaneously at relatively high rates (5^20 Hz in awake animals; see Hubel, 1959;Evarts, 1964;Steriade, 1978;Matsumura et al., 1988;Steriade et al., 2001). This highly £uctuating activity was simulated in vitro by injecting noisy current waveforms (Mainen and Sejnowski, 1995;Stevens and Zador, 1998;Fellous et al., 2001). This approach, however, does not take into account the conductance due to background activity.Given that the neocortex is characterized by a very dense synaptic connectivity (5000^60 000 excitatory synapses per neuron; Cragg, 1967; DeFelipe and Farin ¬ as, 1992), these cells could potentially experience considerable amounts of synaptic conductances during periods of intense network activity. A recent estimation of the electrophysiological parameters of background activity in cat parietal cortex in vivo (Parë et al., 1998) provided evidence for a`high-conductance' state (see also Borg-Graham et al., 1998). By combining intracellular recording...

A train of action potentials (a spike train) can carry information in both the average firing rate and the pattern of spikes in the train. But can such a spike-pattern code be supported by cortical circuits? Neurons in vitro produce a spike pattern in response to the injection of a fluctuating current. However, cortical neurons in vivo are modulated by local oscillatory neuronal activity and by top-down inputs. In a cortical circuit, precise spike patterns thus reflect the interaction between internally generated activity and sensory information encoded by input spike trains. We review the evidence for precise and reliable spike timing in the cortex and discuss its computational role.Reliability and precision are two different quantities. When you make an appointment with your friend, she can either keep the appointment or not show up at all. If she does show up, she might or might not be on time. The former uncertainty is related to reliability, whereas the latter is related to precision. When the same stimulus waveform is repeatedly injected at the soma of a neuron in vitro (FIG. 1a), a similar spike train is obtained on each trial 1,2 (FIG. 1b). When approximately the same number of spikes occur on each trial the neuron is said to be reliable, whereas when the spikes occur almost at the same time across trials it is said to be precise (FIG. 1c). For a single neuron, the potential information content of precise and reliable spike times is many times larger than that which is contained in the firing rate, which is averaged across a typical interval of a hundred milliseconds [3][4][5][6] . The information contained in spike timing is available immediately, rather than after an averaging period. Furthermore, the timing of patterns of spikes can potentially transmit even more information than the timing of the individual constituent spikes 3,7 . The potential relevance of spike patterns becomes apparent when we consider neurons at the population level: when a group of similar neurons (a 'pool') produces precise and reliable spike trains, the neurons they project to receive volleys of synchronous spikes 8,9 . This opens up the possibility of communicating between different cortical areas through synchronous spike volleys.In contrast to the in vitro situation described above, in the intact cortex most excitatory synaptic inputs arrive at the dendrites rather than at the soma (FIG. 1d), and synaptic transmission is typically unreliable [10][11][12][13] . Furthermore, most of these dendritic inputs are not directly related to ongoing sensory stimulation; rather, they reflect spatiotemporally structured internal activity. Therefore, when the same stimulus is presented repeatedly, the resulting spike trains are usually neither precise nor reliable when they are aligned to the stimulus onset 6 . Instead, HHMI Author ManuscriptHHMI Author Manuscript HHMI Author Manuscript neural activity in vivo might be dominated by internally generated complex reverberations or rhythmic oscillations, and precise and reliable sp...