A century of generalization

We review data from both ethology and psychology about generalization, that is how animals respond to sets of stimuli including familiar and novel stimuli. Our main conclusion is that patterns of generalization are largely independent of systematic group (evidence is available for insects, fish, amphibians, reptiles, birds and mammals, including humans), behavioural context (feeding, drinking, courting, etc.), sensory modality (light, sound, etc.) and of whether reaction to stimuli is learned or genetically inherited. These universalities suggest that generalization originates from general properties of nervous systems, and that evolutionary strategies to cope with novelty and variability in stimulation may be limited. Two major shapes of the generalization gradient can be identified, corresponding to two types of stimulus dimensions. When changes in stimulation involve a rearrangement of a constant amount of stimulation on the sense organs, the generalization gradient peaks close to familiar stimuli, and peak responding is not much higher than responding to familiar stimuli. Contrary to what is often claimed, such gradients are better described by Gaussian curves than by exponentials. When the stimulus dimension involves a variation in the intensity of stimulation, the gradient is often monotonic, and responding to some novel stimuli is considerably stronger than responding to familiar stimuli. Lastly, when several or many familiar stimuli are close to each other predictable biases in responding occur, along all studied dimensions. We do not find differences between biases referred to as peak shift and biases referred to as supernormal stimulation. We conclude by discussing theoretical issues.     

Adaptation to Changes in Higher-Order Stimulus Statistics in the Salamander Retina

Adaptation in the retina is thought to optimize the encoding of natural light signals into sequences of spikes sent to the brain. While adaptive changes in retinal processing to the variations of the mean luminance level and second-order stimulus statistics have been documented before, no such measurements have been performed when higher-order moments of the light distribution change. We therefore measured the ganglion cell responses in the tiger salamander retina to controlled changes in the second (contrast), third (skew) and fourth (kurtosis) moments of the light intensity distribution of spatially uniform temporally independent stimuli. The skew and kurtosis of the stimuli were chosen to cover the range observed in natural scenes. We quantified adaptation in ganglion cells by studying linear-nonlinear models that capture well the retinal encoding properties across all stimuli. We found that the encoding properties of retinal ganglion cells change only marginally when higher-order statistics change, compared to the changes observed in response to the variation in contrast. By analyzing optimal coding in LN-type models, we showed that neurons can maintain a high information rate without large dynamic adaptation to changes in skew or kurtosis. This is because, for uncorrelated stimuli, spatio-temporal summation within the receptive field averages away non-gaussian aspects of the light intensity distribution.     

How robust are neural network models of stimulus generalization?

Artificial feed-forward neural networks are commonly used as a tool for modelling stimulus selection and animal signalling. A key finding of stimulus selection research has been generalization: if a given behaviour has been established to one stimulus, perceptually similar novel stimuli are likely to induce a similar response. Stimulus generalization, in feed-forward neural networks, automatically arises as a property of the network. This network property raises understandable concern regarding the sensitivity of the network to variation in its internal parameter values used in relation to its structure and to its training process. Researchers must have confidence that the predictions of their model follow from the underlying biology that they deliberately incorporated in the model, and not from often arbitrary choices about model implementation. We study how network training and parameter perturbations influence the qualitative and quantitative behaviour of a simple but general network. Specifically, for models of stimulus control we study the effect that parameter variation has on the shape of the generalization curves produced by the network. We show that certain network and training conditions produce undesirable artifacts that need to be avoided (or at least understood) when modelling stimulus selection.     

Intensity Characteristics of the Noctuid Acoustic Receptor

Spiking activity of the more sensitive acoustic receptor is described as a function of stimulus intensity. The form of the intensity characteristic depends strongly on stimulus duration. For very brief stimuli, the integral of stimulus power over stimulus duration determines the effectiveness. No response saturation is observed. With longer stimuli (50 msec), a steady firing rate is elicited. The response extends from the spontaneous rate of 20–40 spikes/sec to a saturated firing rate of nearly 700 spikes/sec. The characteristic is monotonic over more than 50 db in stimulus intensity. With very long stimuli (10 sec), the characteristics are nonmonotonic. Firing rates late in the stimulus decrease in response to an increase in stimulus intensity. The non-monotonic characteristics are attributed to intensity-related changes in response adaptation.     

Decoding cortical neuronal signals: Network models,information estimation and spatial tuning

We have studied the encoding of spatial pattern information by complex cells in the primary visual cortex of awake monkeys. Three models for the conditional probabilities of different stimuli, given the neuronal response, were fit and compared using cross-validation. For our data, a feed-forward neural network proved to be the best of these models.The information carried by a cell about a stimulus set can be calculated from the estimated conditional probabilities. We performed a spatial spectroscopy of the encoding, examining how the transmitted information varies with both the average coarseness of the stimulus set and the coarseness differences within it. We find that each neuron encodes information about many features at multiple scales. Our data do not appear to allow a characterization of these variations in terms of the detection of simple single features such as oriented bars.     

Colour categorization by domestic chicks

Spectral stimuli form a physical continuum, which humans divide into discrete non-overlapping regions or categories that are designated by colour names. Little is known about whether non-verbal animals form categories on stimulus continua, but work in psychology and artificial intelligence provides models for stimulus generalization and categorization. We compare predictions of such models to the way poultry chicks (Gallus gallus) generalize to novel stimuli following appetitive training to either one or two colours. If the two training colours are (to human eyes) red and greenish-yellow or green and blue, chicks prefer intermediates, i.e. orange rather than red or yellow and turquoise rather than green or blue. The level of preference for intermediate colours implies that the chicks interpolate between the training stimuli. However, they do not extrapolate beyond the limits set by the training stimuli, at least for red and yellow training colours. Similarly, chicks trained to red and blue generalize to purple, but they do not generalize across grey after training to the complementary colours yellow and blue. These results are consistent with a modified version of a Bayesian model of generalization from multiple examples that was proposed by Shepard and show similarities to human colour categorization.     

Auditory pre-experience modulates classification of affect intensity: evidence for the evaluation of call salience by a non-human mammal,the bat <Emphasis Type="Italic">Megaderma lyra</Emphasis>

Introduction

Immediate responses towards emotional utterances in humans are determined by the acoustic structure and perceived relevance, i.e. salience, of the stimuli, and are controlled via a central feedback taking into account acoustic pre-experience. The present study explores whether the evaluation of stimulus salience in the acoustic communication of emotions is specifically human or has precursors in mammals. We created different pre-experiences by habituating bats (Megaderma lyra) to stimuli based on aggression, and response, calls from high or low intensity level agonistic interactions, respectively. Then we presented a test stimulus of opposite affect intensity of the same call type. We compared the modulation of response behaviour by affect intensity between the reciprocal experiments.

Results

For aggression call stimuli, the bats responded to the dishabituation stimuli independent of affect intensity, emphasising the attention-grabbing function of this call type. For response call stimuli, the bats responded to a high affect intensity test stimulus after experiencing stimuli of low affect intensity, but transferred habituation to a low affect intensity test stimulus after experiencing stimuli of high affect intensity. This transfer of habituation was not due to over-habituation as the bats responded to a frequency-shifted control stimulus. A direct comparison confirmed the asymmetric response behaviour in the reciprocal experiments.

Conclusions

Thus, the present study provides not only evidence for a discrimination of affect intensity, but also for an evaluation of stimulus salience, suggesting that basic assessment mechanisms involved in the perception of emotion are an ancestral trait in mammals.

     

Event-related neuronal responses in the human strio-pallido-thalamic system. I. Sensory and motor functions

Multiunit activity was recorded from strio-pallido-thalamic sites in parkinsonian patients bearing gold electrodes for diagnosis and therapy. The patients voluntarily participated in tasks designed to study neuronal correlates of both physical and semantic characteristics of stimuli as well as motor responses. Six modifications of the stimulus-response paradigm were used: visual odd-ball, visual and acoustic odd-ball tasks; tasks in which either the stimulus intensity or the meaning of non-target stimuli varied; single-stage delayed response and dual-stage delayed response tasks, respectively. In each task the patients had to evaluate some of the stimulus characteristics and to respond in a particular way according to the preliminary instructions. Peristimulus time histograms for each multiunit separately as well as profiles of reactions and profiles of reaction differences for the whole set of multiunits were calculated and subjected to statistical analysis. Two functional groups of subcortical neuronal reactions, stimulus-related and response-related activities, were separated. The stimulus-related activities of most multiunits were modality-unspecific. Their most striking feature was dependence on stimulus relevance and also its probability, the strongest reactions observed in response to task relevant stimuli occurring with low probability. The response-related activities occurred prior to initiation of movements, dependent upon the particular action and its probability. The data suggest at least two different and spatially overlapping subcortical channels responsible for goal-directed behaviour: the one related to stimulus assessment and the other to preparation for motor action.     

Receptive field types in the lateral geniculate body and their frequency characteristics]

Time amplitude -- frequency characteristics of the I and II types of receptive fields (RF) of lateral geniculate and their dependence on the contrast and spatial parameters of the light stimulus were studied. It is shown that the frequency characteristics of the RF I type depends on the contrast and area of the light stimulus, the higher being the contrast at a small area the smaller are the low frequencies. However at a large area of the stimulus the inhibition of low frequencies is greater at a small contrast. The transmitting band of frequency characteristics of RF II type does not depend on the contrast at a small area of the stimulus, at a large area a fall of low frequencies takes place at high contrasts of the stimulus. Such different behaviour of the receptive fields is explained by the models, which take into account RF spatial characteristics.     

Temporal Modulations of Contact Force during Haptic Surface Exploration

Individuals constantly modulate their exploratory movements and adapt their internal hypotheses to incoming sensory information to achieve a thorough and realistic percept. Perception depends on the exploratory movements as well as influencing them. While this seems to be common sense, scientifically we know very little about the temporal dynamics during haptic exploration. To address this, we investigated the exploratory force modulations of two groups of healthy young adults during the exploration of grated surfaces with differing detection difficulty during successive (n = 20) and random stimulus presentation (n = 20). Results showed that exploratory force depended on stimulus properties and increased with increasing detection difficulty. Both experiments yielded the same direction of results with slightly smaller effects in the random stimulus presentation group. Across exploration time average fingertip force also increased. The biggest increase occurred systematically at the beginning (within the first 40 percent) of exploration time per stimulus indicating that most critical information is received during the initial contact phase and is directly transformed into the exploration procedure and force application. Furthermore, video-analyses and comparisons to our high temporal resolution data revealed strong dynamic changes in pressure application during test stimulus exploration with differences in the force dynamics and exploration strategies of simple and difficult stimuli.     

Modelling Individual Differences in the Form of Pavlovian Conditioned Approach Responses: A Dual Learning Systems Approach with Factored Representations

Reinforcement Learning has greatly influenced models of conditioning, providing powerful explanations of acquired behaviour and underlying physiological observations. However, in recent autoshaping experiments in rats, variation in the form of Pavlovian conditioned responses (CRs) and associated dopamine activity, have questioned the classical hypothesis that phasic dopamine activity corresponds to a reward prediction error-like signal arising from a classical Model-Free system, necessary for Pavlovian conditioning. Over the course of Pavlovian conditioning using food as the unconditioned stimulus (US), some rats (sign-trackers) come to approach and engage the conditioned stimulus (CS) itself – a lever – more and more avidly, whereas other rats (goal-trackers) learn to approach the location of food delivery upon CS presentation. Importantly, although both sign-trackers and goal-trackers learn the CS-US association equally well, only in sign-trackers does phasic dopamine activity show classical reward prediction error-like bursts. Furthermore, neither the acquisition nor the expression of a goal-tracking CR is dopamine-dependent. Here we present a computational model that can account for such individual variations. We show that a combination of a Model-Based system and a revised Model-Free system can account for the development of distinct CRs in rats. Moreover, we show that revising a classical Model-Free system to individually process stimuli by using factored representations can explain why classical dopaminergic patterns may be observed for some rats and not for others depending on the CR they develop. In addition, the model can account for other behavioural and pharmacological results obtained using the same, or similar, autoshaping procedures. Finally, the model makes it possible to draw a set of experimental predictions that may be verified in a modified experimental protocol. We suggest that further investigation of factored representations in computational neuroscience studies may be useful.     

A Probabilistic Palimpsest Model of Visual Short-term Memory

Working memory plays a key role in cognition, and yet its mechanisms remain much debated. Human performance on memory tasks is severely limited; however, the two major classes of theory explaining the limits leave open questions about key issues such as how multiple simultaneously-represented items can be distinguished. We propose a palimpsest model, with the occurrent activity of a single population of neurons coding for several multi-featured items. Using a probabilistic approach to storage and recall, we show how this model can account for many qualitative aspects of existing experimental data. In our account, the underlying nature of a memory item depends entirely on the characteristics of the population representation, and we provide analytical and numerical insights into critical issues such as multiplicity and binding. We consider representations in which information about individual feature values is partially separate from the information about binding that creates single items out of multiple features. An appropriate balance between these two types of information is required to capture fully the different types of error seen in human experimental data. Our model provides the first principled account of misbinding errors. We also suggest a specific set of stimuli designed to elucidate the representations that subjects actually employ.     

Types of receptive fields in the lateral geniculate body and their functional model

In the Type I receptive fields (RFs) changes of the luminance leads to a shift of the curve relating the response and the stimulus area along the abscissa, in the Type II RFs the maximum of a response does not shift with changes of the luminance (Types I and II on classification by Glezer et al., 1971, 1972). The transient responses were observed in the Type I RFs and sustained responses in the Type II RFs. In the Type I RFs variation of the stimulus area and intensity brings about the change in the temporal and spatial frequency characteristics. This is produced by functional reorganization of the RF. In the Type II RFs there is no functional reorganization. The data obtained indicate that the Type I RFs are non-linear. By contrast, the Type II RFs are linear systems. The analysis of the model has shown that the distinctions in the dynamic characteristics of the responses of RFs belonging to different types is mainly due to different time constants for excitation and inhibition as well as inhibition coefficients. Distinctions in the mode of dependence of the RF response on stimulus parameters have been found to result from different relationship between delay time and stimulus parameters as well as different forms of the spatial weighting functions. It is shown that the Type I RFs transmit higher frequency components of the image spectrum, i.e. they emphasise the temporal and spatial contrasts. The Type II RFs transmit low frequency components of the spectrum including information about the intensity of an input stimulus.     

The case for a cognitive approach to animal learning and behavior

The dangers of hypothesizing about unobservable cognitive mechanisms are well known to behavior analysts. I propose, however, that carefully fashioned cognitive theories that make predictions that are inconsistent with current behavioral theories can provide useful research tools for the understanding of behavior. Furthermore, even if the results of such research may be accommodated by modifying existing behavioral theories, our understanding of behavior is often advanced by the empirical findings because it is unlikely that the research would have been conducted in the absence of such cognitive hypothesizing. Two examples of the development of emergent relations are described: The first deals with the nature of a pigeon's 'representation' of two stimuli both of which are associated with correct responding to a third in a many-to-one matching task (stimulus equivalence or common representations). The second has to do with transitive inference, the emergent relation between two stimuli mediated by their relation to a common stimulus in a simultaneous discrimination.     

Learning cortical topography from spatiotemporal stimuli

Stimulus representation is a functional interpretation of early sensory cortices. Early sensory cortices are subject to stimulus-induced modifications. Common models for stimulus-induced learning within topographic representations are based on the stimuli's spatial structure and probability distribution. Furthermore, we argue that average temporal stimulus distances reflect the stimuli's relatedness. As topographic representations reflect the stimuli's relatedness, the temporal structure of incoming stimuli is important for the learning in cortical maps. Motivated by recent neurobiological findings, we present an approach of cortical self-organization that additionally takes temporal stimulus aspects into account. The proposed model transforms average interstimulus intervals into representational distances. Thereby, neural topography is related to stimulus dynamics. This offers a new time-based interpretation of cortical maps. Our approach is based on a wave-like spread of cortical activity. Interactions between dynamics and feedforward activations lead to shifts of neural activity. The psychophysical saltation phenomenon may represent an analogue to the shifts proposed here. With regard to cortical plasticity, we offer an explanation for neurobiological findings that other models cannot explain. Moreover, we predict cortical reorganizations under new experimental, spatiotemporal conditions. With regard to psychophysics, we relate the saltation phenomenon to dynamics and interaction in early sensory cortices and predict further effects in the perception of spatiotemporal stimuli. Received: 17 March 1999 / Accepted in revised form: 10 August 1999     

Sparsity and Compressed Coding in Sensory Systems

Considering that many natural stimuli are sparse, can a sensory system evolve to take advantage of this sparsity? We explore this question and show that significant downstream reductions in the numbers of neurons transmitting stimuli observed in early sensory pathways might be a consequence of this sparsity. First, we model an early sensory pathway using an idealized neuronal network comprised of receptors and downstream sensory neurons. Then, by revealing a linear structure intrinsic to neuronal network dynamics, our work points to a potential mechanism for transmitting sparse stimuli, related to compressed-sensing (CS) type data acquisition. Through simulation, we examine the characteristics of networks that are optimal in sparsity encoding, and the impact of localized receptive fields beyond conventional CS theory. The results of this work suggest a new network framework of signal sparsity, freeing the notion from any dependence on specific component-space representations. We expect our CS network mechanism to provide guidance for studying sparse stimulus transmission along realistic sensory pathways as well as engineering network designs that utilize sparsity encoding.     

Isometric force generated by locust skeletal muscle: responses to single stimuli

A mathematical model of the locust hind leg extensor muscle is described. The model accounts for the force response of the muscle to well-separated input stimuli under isometric conditions. Experimental data was collected by stimulating the extensor muscle and measuring the force generated at the tibia. In developing a model it was assumed that the response to a single isolated stimulus was linear. A linear model was found to fit well to the response to an isolated stimulus. No assumptions were made about the model order and models of various order were fitted to data in the frequency domain, using a least squares fit. The stimulus can be approximated as an impulse, with the response to each stimulus well described by a linear second-order system. Using a third-order model provided a better fit to data, but the improvement in fit was marginal and the model uses one extra parameter. A fourth-order model, which is often used to describe the behaviour of isometric muscle was found to overfit the data. Using a second-order model provides a simpler way of describing the behaviour of an isometric twitch.     

On the electrodetection threshold of aquatic vertebrates with ampullary or mucous gland electroreceptor organs

Reinterpretation of research on the electric sense in aquatic organisms with ampullary organs results in the following conclusions. The detection limit of limnic vertebrates with ampullary organs is 1 microV cm(-1), and of marine fish is 20 nV cm(-1). Angular movements are essential for stimulation of the ampullary system in uniform d.c. fields. Angular movements in the geomagnetic field also generate induction voltages, which exceed the 20 nV cm(-1) limit in marine fish. As a result, marine electrosensitive fish are sensitive to motion in the geomagnetic field, whereas limnic fish are not. Angular swimming movements generate a.c. stimuli, which act like the noise in a stochastic resonance system, and result in a detection threshold in marine organisms as low as 1 nV cm(-1). Fish in the benthic space are exposed to stronger electric stimuli than fish in the pelagic space. Benthic fish scan the orientation plane for the maximum potential difference with their raster of electroreceptor organs, in order to locate bioelectric prey. This behaviour explains why the detection threshold does not depend on fish size. Pelagic marine fish are mainly exposed to electric fields caused by movements in the geomagnetic field. The straight orientation courses found in certain shark species might indicate that the electric sense functions as a simple bisensor system. Symmetrical stimulation of the sensory raster would provide an easy way to keep a straight course with respect to a far-field stimulus. The same neural mechanism would be effective in the location of a bioelectric prey generating a near-field stimulus. The response criteria in conditioning experiments and in experiments with spontaneous reactions are discussed.     

Understanding the temporal codes of intra-cellular signals

The health of organisms and cells depends on appropriate responses to diverse internal and external cues, stimuli, or challenges, such as changes in hormone or cytokine levels, or exposure to a pathogen. Cellular responses must be tailored to the identity and intensity of the stimulus and therefore intra-cellular signals must carry information about both. However, signaling mediators often form intricate networks that react to multiple stimuli yet manage to produce stimulus-specific responses. The multi-functionality ('functional pleiotropism') of signaling nodes suggests that biological networks have evolved ways of passing physiologically relevant stimulus information through shared channels. Increasing evidence supports the notion that this is achieved in part through temporal regulation of signaling mediators' activities. The present challenge is to identify the features of temporal activity profile that represent information about a given stimulus and understand how cells read the temporal codes to control their responses.