Visual Feature Learning Research Articles

Unsupervised learning of visual features through Spike Timing Dependent Plasticity

Unsupervised learning of visual features through Spike Timing Dependent Plasticity

Perception of object trajectories during occlusion constrains statistical learning of visual features

Perception of object trajectories during occlusion constrains statistical learning of visual features

Learning to see the trees before the forest: reversible deactivation of the superior colliculus during learning of local and global visual features.

Previous studies have established that deactivation of the superior colliculus severely retards the normally rapid learning of pattern discriminations in the mature cat. The purpose of this study was to test the hypothesis that the midbrain plays an important role in the learning of simple pattern discriminations and that the contribution of this pathway is to the perception of global, rather than local, features of a figure. To answer this question, pattern discrimination learning was studied in three intact cats and in three experimental cats during bilateral reversible deactivation of the superficial layers of the superior colliculus (SC). The animals concurrently learned to discriminate three pairs of compound visual patterns composed of small (local element) Ts or 7s. Congruent and incongruent stimulus pairs were large (global) Ts vs. 7s comprising the same or different local elements, respectively. The random stimulus pair consisted of randomly scattered local elements (Ts vs. 7s). The animals were trained to respond to the large Ts in the congruent and incongruent pairs and the small Ts in the random pair. In the normal cats, learning of the random pair was much slower than the learning of the congruent and incongruent pairs. This finding demonstrated the theory of global precedence, because the animals learned the global features of the congruent and incongruent pairs much more quickly than the local features of the random pair. In contrast, during bilateral deactivation of the superficial layers of the SC, the learning of the incongruent pair was significantly retarded and took longer to learn than the random pair. Congruent and random pair learning rates were unchanged. The specific deficit in learning the incongruent pair indicates that the learning of global, but not local, elements of the visual pattern is impaired during deactivation of the SC. The unimpaired use of local features permitted the animals to learn the congruent and random pairs at normal rates. Therefore, deactivation of the superficial layers of the SC during pattern discrimination learning reverses the precedence for global visual features that is typical of normal learning.

Unsupervised statistical learning of higher-order spatial structures from visual scenes.

Three experiments investigated the ability of human observers to extract the joint and conditional probabilities of shape co-occurrences during passive viewing of complex visual scenes. Results indicated that statistical learning of shape conjunctions was both rapid and automatic, as subjects were not instructed to attend to any particularfeatures of the displays. Moreover, in addition to single-shape frequency, subjects acquired in parallel several different higher-order aspects of the statistical structure of the displays, including absolute shape-position relations in an array, shape-pair arrangements independent of position, and conditional probabilities of shape co-occurrences. Unsupervised learning of these higher-order statistics provides support for Barlow's theory of visual recognition, which posits that detecting "suspicious coincidences" of elements during recognition is a necessary prerequisite for efficient learning of new visual features.

Searching for the memory trace in a mini-brain, the honeybee.

To determine general or species-specific properties in neural systems, it is necessary to use comparative data in evaluating experimental findings. Presented here are data on associative learning and memory formation in honeybees, emphasizing a comparative approach. We focus on four aspects: (1) the role of an identified neuron, VUM(mx1), as a neural substrate of appetitive reinforcement; (2) the sequences of molecular events as they correlate with five forms of memory stages; (3) the localization of the memory traces following appetitive olfactory learning; and (4) the brief description of several forms of complex learning in bees (configuration in olfactory conditioning, categorization in visual feature learning, delayed matching-to-sample learning, and latent learning in navigation). VUM(mx1) activity following the conditioned stimulus odor is sufficient to replace the unconditioned stimulus, and VUM(mx1) changes its response properties during learning similarly to what is known from dopamine neurons in the basal ganglia of the mammalian brain. The transition from short- to mid- and long-term forms of memory can be related to specific activation of second messenger cascades (involving NOS, PKA, PKC, and PKM) resembling general features of neural plasticity at the cellular level. The particular time course of the various memory traces may be adapted to the behavioral context in which they are used; here, the foraging cycle of the bee. Memory traces for even such a simple form of learning as olfactory conditioning are multiple and distributed, involving first- and second-order sensory neuropils (antennal lobe and mushroom bodies), but with distinctly different properties. The wealth of complex forms of learning in the context of foraging indicates basic cognitive capacities based on rule extraction and context-dependent learning. It is believed that bees might be a useful model for studying cognitive faculties at a middle level of complexity.

The role of orientation flights on homing performance in honeybees.

Honeybees have long served as a model organism for investigating insect navigation. Bees, like many other nesting animals, primarily use learned visual features of the environment to guide their movement between the nest and foraging sites. Although much is known about the spatial information encoded in memory by experienced bees, the development of large-scale spatial memory in naive bees is not clearly understood. Past studies suggest that learning occurs during orientation flights taken before the start of foraging. We investigated what honeybees learn during their initial experience in a new landscape by examining the homing of bees displaced after a single orientation flight lasting only 5-10 min. Homing ability was assessed using vanishing bearings and homing speed. At release sites with a view of the landmarks immediately surrounding the hive, 'first-flight' bees, tested after their very first orientation flight, had faster homing rates than 'reorienting foragers', which had previous experience in a different site prior to their orientation flight in the test landscape. First-flight bees also had faster homing rates from these sites than did 'resident' bees with full experience of the terrain. At distant sites, resident bees returned to the hive more rapidly than reorienting or first-flight bees; however, in some cases, the reorienting bees were as successful as the resident bees. Vanishing bearings indicated that all three types of bees were oriented homewards when in the vicinity of landmarks near the hive. When bees were released out of sight of these landmarks, hence forcing them to rely on a route memory, the 'first-flight' bees were confused, the 'reorienting' bees chose the homeward direction except at the most distant site and the 'resident' bees were consistently oriented homewards.