Learning optimal integration of arbitrary features in a perceptual discrimination task

Sensory cue-combination in the context of newly learned categories

A large body of prior research has evaluated how humans combine multiple sources of information pertaining to stimuli drawn from continuous dimensions, such as distance or size. These prior studies have repeatedly demonstrated that in these circumstances humans integrate cues in a near-optimal fashion, weighting cues according to their reliability. However, most of our interactions with sensory information are in the context of categories such as objects and phonemes, thereby requiring a solution to the cue combination problem by mapping sensory estimates from continuous dimensions onto task-relevant categories. Previous studies have examined cue combination with natural categories (e.g., phonemes), providing qualitative evidence that human observers utilize information about the distributional properties of task-relevant categories, in addition to sensory information, in such categorical cue combination tasks. In the present study, we created and taught human participants novel audiovisual categories, thus allowing us to quantitatively evaluate participants’ integration of sensory and categorical information. Comparing participant behavior to the predictions of a statistically optimal observer that ideally combines all available sources of information, we provide the first evidence, to our knowledge, that human observers combine sensory and category information in a statistically optimal manner.

Task-irrelevant emotion facilitates face discrimination learning

We understand poorly how the ability to discriminate faces from one another is shaped by visual experience. The purpose of the present study is to determine whether face discrimination learning can be facilitated by facial emotions. To answer this question, we used a task-irrelevant perceptual learning paradigm because it closely mimics the learning processes that, in daily life, occur without a conscious intention to learn and without an attentional focus on specific facial features. We measured face discrimination thresholds before and after training. During the training phase (4 days), participants performed a contrast discrimination task on face images. They were not informed that we introduced (task-irrelevant) subtle variations in the face images from trial to trial. For the Identity group, the task-irrelevant features were variations along a morphing continuum of facial identity. For the Emotion group, the task-irrelevant features were variations along an emotional expression morphing continuum. The Control group did not undergo contrast discrimination learning and only performed the pre-training and post-training tests, with the same temporal gap between them as the other two groups. Results indicate that face discrimination improved, but only for the Emotion group. Participants in the Emotion group, moreover, showed face discrimination improvements also for stimulus variations along the facial identity dimension, even if these (task-irrelevant) stimulus features had not been presented during training. The present results highlight the importance of emotions for face discrimination learning.

Cue-recruitment for extrinsic signals after training with low information stimuli

Cue-recruitment occurs when a previously ineffective signal comes to affect the perceptual appearance of a target object, in a manner similar to the trusted cues with which the signal was put into correlation during training [1], [2]. Jain, Fuller and Backus [3] reported that extrinsic signals, those not carried by the target object itself, were not recruited even after extensive training. However, recent studies have shown that training using weakened trusted cues can facilitate recruitment of intrinsic signals [4]–[7]. The current study was designed to examine whether extrinsic signals can be recruited by putting them in correlation with weakened trusted cues. Specifically, we tested whether an extrinsic visual signal, the rotary motion direction of an annulus of random dots, and an extrinsic auditory signal, direction of an auditory pitch glide, can be recruited as cues for the rotation direction of a Necker cube. We found learning, albeit weak, for visual but not for auditory signals. These results extend the generality of the cue-recruitment phenomenon to an extrinsic signal and provide further evidence that the visual system learns to use new signals most quickly when other, long-trusted cues are unavailable or unreliable.

Learning to detect and combine the features of an object

To recognize an object, it is widely supposed that we first detect and then combine its features. Familiar objects are recognized effortlessly, but unfamiliar objects--like new faces or foreign-language letters--are hard to distinguish and must be learned through practice. Here, we describe a method that separates detection and combination and reveals how each improves as the observer learns. We dissociate the steps by two independent manipulations: For each step, we do or do not provide a bionic crutch that performs it optimally. Thus, the two steps may be performed solely by the human, solely by the crutches, or cooperatively, when the human takes one step and a crutch takes the other. The crutches reveal a double dissociation between detecting and combining. Relative to the two-step ideal, the human observer's overall efficiency for unconstrained identification equals the product of the efficiencies with which the human performs the steps separately. The two-step strategy is inefficient: Constraining the ideal to take two steps roughly halves its identification efficiency. In contrast, we find that humans constrained to take two steps perform just as well as when unconstrained, which suggests that they normally take two steps. Measuring threshold contrast (the faintness of a barely identifiable letter) as it improves with practice, we find that detection is inefficient and learned slowly. Combining is learned at a rate that is 4× higher and, after 1,000 trials, 7× more efficient. This difference explains much of the diversity of rates reported in perceptual learning studies, including effects of complexity and familiarity.

Perceptual classification in a rapidly changing environment

Humans and monkeys can learn to classify perceptual information in a statistically optimal fashion if the functional groupings remain stable over many hundreds of trials, but little is known about categorization when the environment changes rapidly. Here, we used a combination of computational modeling and functional neuroimaging to understand how humans classify visual stimuli drawn from categories whose mean and variance jumped unpredictably. Models based on optimal learning (Bayesian model) and a cognitive strategy (working memory model) both explained unique variance in choice, reaction time, and brain activity. However, the working memory model was the best predictor of performance in volatile environments, whereas statistically optimal performance emerged in periods of relative stability. Bayesian and working memory models predicted decision-related activity in distinct regions of the prefrontal cortex and midbrain. These findings suggest that perceptual category judgments, like value-guided choices, may be guided by multiple controllers.

Psychophysical reverse correlation with multiple response alternatives

Psychophysical reverse-correlation methods such as the "classification image" technique provide a unique tool to uncover the internal representations and decision strategies of individual participants in perceptual tasks. Over the past 30 years, these techniques have gained increasing popularity among both visual and auditory psychophysicists. However, thus far, principled applications of the psychophysical reverse-correlation approach have been almost exclusively limited to two-alternative decision (detection or discrimination) tasks. Whether and how reverse-correlation methods can be applied to uncover perceptual templates and decision strategies in situations involving more than just two response alternatives remain largely unclear. Here, the authors consider the problem of estimating perceptual templates and decision strategies in stimulus identification tasks with multiple response alternatives. They describe a modified correlational approach, which can be used to solve this problem. The approach is evaluated under a variety of simulated conditions, including different ratios of internal-to-external noise, different degrees of correlations between the sensory observations, and various statistical distributions of stimulus perturbations. The results indicate that the proposed approach is reasonably robust, suggesting that it could be used in future empirical studies.

Linking neurons to behavior in multisensory perception: a computational review

A large body of psychophysical and physiological findings has characterized how information is integrated across multiple senses. This work has focused on two major issues: how do we integrate information, and when do we integrate, i.e., how do we decide if two signals come from the same source or different sources. Recent studies suggest that humans and animals use Bayesian strategies to solve both problems. With regard to how to integrate, computational studies have also started to shed light on the neural basis of this Bayes-optimal computation, suggesting that, if neuronal variability is Poisson-like, a simple linear combination of population activity is all that is required for optimality. We review both sets of developments, which together lay out a path towards a complete neural theory of multisensory perception.