Patients with central vision loss depend on peripheral vision for everyday functions. A preferred retinal locus (PRL) on the intact retina is commonly trained as a new “fovea” to help. However, reprogramming the fovea-centered oculomotor control is difficult, so saccades often bring the defunct fovea to block the target. Aligning PRL with distant targets also requires multiple saccades and sometimes head movements. To overcome these problems, we attempted to train normal-sighted observers to form a preferred retinal annulus (PRA) around a simulated scotoma, so that they could rely on the same fovea-centered oculomotor system and make short saccades to align PRA with the target. Observers with an invisible simulated central scotoma (5° radius) practiced making saccades to see a tumbling-E target at 10° eccentricity. The otherwise blurred E target became clear when saccades brought a scotoma-abutting clear window (2° radius) to it. The location of the clear window was either fixed for PRL training, or changing among 12 locations for PRA training. Various cues aided the saccades through training. Practice quickly established a PRL or PRA. Comparing to PRL-trained observers whose first saccades persistently blocked the target with scotoma, PRA-trained observers produced more accurate first saccades. The benefits of more accurate PRA-based saccades also outweighed the costs of slower latency. PRA training may provide a very efficient strategy to cope with central vision loss, especially for aging patients who have major difficulties adapting to a PRL.

One interesting observation of perceptual learning is the asymmetric transfer between stimuli at different external noise levels: learning at zero/low noise can transfer significantly to the same stimulus at high noise, but not vice versa. The mechanisms underlying this asymmetric transfer have been investigated by psychophysical, neurophysiological, brain imaging, and computational modeling studies. One study (PNAS 113 (2016) 5724-5729) reported that rTMS stimulations of dorsal and ventral areas impair motion direction discrimination of moving dot stimuli at 40% coherent (“noisy”) and 100% coherent (zero-noise) levels, respectively. However, after direction training at 100% coherence, only rTMS stimulation of the ventral cortex is effective, disturbing direction discrimination at both coherence levels. These results were interpreted as learning-induced changes of functional specializations of visual areas. We have concerns with the behavioral data of this study. First, contrary to the report of highly location-specific motion direction learning, our replicating experiment showed substantial learning transfer (e.g., transfer/learning ratio = 81.9% vs. 14.8% at 100% coherence). Second and more importantly, we found complete transfer of direction learning from 40% to 100% coherence, a critical baseline that is missing in this study. The transfer effect suggests that similar brain mechanisms underlie motion direction processing at two coherence levels. Therefore, this study’s conclusions regarding the roles of dorsal and ventral areas in motion direction processing at two coherence levels, as well as the effects of perceptual learning, are not supported by proper experimental evidence. It remains unexplained why distinct impacts of dorsal and ventral rTMS stimulations on motion direction discrimination were observed.

Neuronal responses to one-dimensional orientations are combined to represent two-dimensional composite patterns, which plays a key role in intermediate-level vision such as texture segmentation. However, where and how the visual cortex starts to represent composite patterns, such as a plaid consisting of two superimposing gratings of different orientations, remains neurophysiologically elusive. Psychophysical and modeling evidence has suggested the existence of early neural mechanisms specialized in plaid detection [1-6], but the responses of V1 neurons to an optimally orientated grating are actually suppressed by a superimposing grating of different orientation (i.e., cross-orientation inhibition) [7, 8]. Would some other V1 neurons be plaid detectors? Here we used two-photon calcium imaging [9] to compare the responses of V1 superficial-layer neurons to gratings and plaids in awake macaques. We found that many non-orientation-tuned neurons responded weakly to gratings, but strongly to plaids, often with plaid orientation selectivity and cross-angle selectivity. In comparison, most (~94%) orientation-tuned neurons showed more or less cross-orientation inhibition, regardless of the relative stimulus contrasts. Only a small portion (~8%) of them showed plaid facilitation at off-peak orientations. These results suggest separate subpopulations of plaid and grating responding neurons. Because most plaid neurons (~95%) were insensitive to motion direction, they were plaid pattern detectors, not plaid motion detectors.

Perceptual learning, which improves stimulus discrimination, typically results from training with a single stimulus condition. Two major learning mechanisms, early cortical neural plasticity and response reweighting, have been proposed. Here we report a new format of perceptual learning that by design may have bypassed these mechanisms. Instead it is more likely based on abstracted stimulus evidence from multiple stimulus conditions. Specifically, we had observers practice orientation discrimination with Gabors or symmetric dot patterns at up to 47 random or rotating location´orientation conditions. Although each condition received sparse trials (16 trials/session), the practice produced significant orientation learning. Learning also transferred to a Gabor at a single untrained condition with 2-3 time lower orientation thresholds. Moreover, practicing a single stimulus condition with matched trial frequency (16 trials/session) failed to produce significant learning. These results suggested that learning with multiple stimulus conditions may not come from early cortical plasticity or response reweighting with each particular condition. Rather, it may materialize through a new format of perceptual learning, in which orientation evidence invariant to particular orientations and locations is first abstracted from multiple stimulus conditions, and then reweighted by later learning mechanisms. The coarse-to-fine transfer of orientation learning from multiple Gabors or symmetric-dot-patterns to a single Gabor also suggested the involvement of orientation concept learning by the learning mechanisms.

A person’s ability to discriminate fine differences in tone frequency is vital for everyday hearing such as listening to speech and music. This ability can be improved through training (i.e., tone frequency learning). Depending on stimulus configurations and training procedures, tone frequency learning can either transfer to new frequencies, which would suggest learning of a general task structure, or show significant frequency specificity, which would suggest either changes in neural representations of trained frequencies, or reweighting of frequency-specific neural responses. Here we tested the hypothesis that frequency specificity in tone frequency learning can be abolished with a double-training procedure. Specifically, participants practiced tone frequency discrimination at 1 or 6 kHz, presumably encoded by different temporal or place coding mechanisms, respectively. The stimuli were brief tone pips known to produce significant specificity. Tone frequency learning was indeed initially highly frequency specific (Experiment 1). However, with additional exposure to the other untrained frequency via an irrelevant temporal interval discrimination task, or even background play during a visual task, learning transferred completely (1-to-6 kHz or 6-to-1 kHz) (Experiments 2-4). These results support general task structure learning, or concept learning in our term, in tone frequency learning despite initial frequency specificity. They also suggest strategies to design efficient auditory training in practical settings.