Surprise-induced enhancements in the associability of Pavlovian cues facilitate learning across behavior systems

Surprising violations of outcome expectancies have long been known to enhance the associability of Pavlovian cues; that is, the rate at which the cue enters into further associations. The adaptive value of such enhancements resides in promoting new learning in the face of uncertainty. However, it is unclear whether associability enhancements reflect increased associative plasticity within a particular behavior system, or whether they can facilitate learning between a cue and any arbitrary outcome, as suggested by attentional models of conditioning. Here, we show evidence consistent with the latter hypothesis. Violating the outcome expectancies generated by a cue in an appetitive setting (feeding behavior system) facilitated subsequent learning about the cue in an aversive setting (defense behavior system). In addition to shedding light on the nature of associability enhancements, our findings offer the neuroscientist a behavioral tool to dissociate their neural substrates from those of other, behavior system- or valence-specific changes. Moreover, our results present an opportunity to utilize associability enhancements to the advantage of counterconditioning procedures in therapeutic contexts.

Fiber-specific white matter reductions in Parkinson hallucinations and visual dysfunction

ObjectiveTo investigate the microstructural and macrostructural white matter changes that accompany visual hallucinations and low visual performance in Parkinson disease, a risk factor for Parkinson dementia.MethodsWe performed fixel-based analysis, a novel technique that provides metrics of specific fiber-bundle populations within a voxel (or fixel). Diffusion MRI data were acquired from patients with Parkinson disease (n = 105, of whom 34 were low visual performers and 19 were hallucinators) and age-matched controls (n = 35). We used whole-brain fixel-based analysis to compare microstructural differences in fiber density (FD), macrostructural differences in fiber bundle cross section (FC), and the combined FD and FC (FDC) metric across all white matter fixels. We then performed a tract-of-interest analysis comparing the most sensitive FDC metric across 11 tracts within the visual system.ResultsPatients with Parkinson disease hallucinations exhibited macrostructural changes (reduced FC) within the splenium of the corpus callosum and the left posterior thalamic radiation compared to patients without hallucinations. While there were no significant changes in FD, we found large reductions in the combined FDC metric in Parkinson hallucinators within the splenium (>50% reduction compared to nonhallucinators). Patients with Parkinson disease and low visual performance showed widespread microstructural and macrostructural changes within the genu and splenium of the corpus callosum, bilateral posterior thalamic radiations, and left inferior fronto-occipital fasciculus.ConclusionsWe demonstrate specific white matter tract degeneration affecting posterior thalamic tracts in patients with Parkinson disease with hallucinations and low visual performance, providing direct mechanistic support for attentional models of visual hallucinations.

Attention-Passing Models for Robust and Data-Efficient End-to-End Speech Translation

Speech translation has traditionally been approached through cascaded models consisting of a speech recognizer trained on a corpus of transcribed speech, and a machine translation system trained on parallel texts. Several recent works have shown the feasibility of collapsing the cascade into a single, direct model that can be trained in an end-to-end fashion on a corpus of translated speech. However, experiments are inconclusive on whether the cascade or the direct model is stronger, and have only been conducted under the unrealistic assumption that both are trained on equal amounts of data, ignoring other available speech recognition and machine translation corpora. In this paper, we demonstrate that direct speech translation models require more data to perform well than cascaded models, and although they allow including auxiliary data through multi-task training, they are poor at exploiting such data, putting them at a severe disadvantage. As a remedy, we propose the use of end- to-end trainable models with two attention mechanisms, the first establishing source speech to source text alignments, the second modeling source to target text alignment. We show that such models naturally decompose into multi-task–trainable recognition and translation tasks and propose an attention-passing technique that alleviates error propagation issues in a previous formulation of a model with two attention stages. Our proposed model outperforms all examined baselines and is able to exploit auxiliary training data much more effectively than direct attentional models.

The blocking effect in associative learning involves learned biases in rapid attentional capture

Blocking refers to the finding that less is learned about the relationship between a stimulus and an outcome if pairings are conducted in the presence of a second stimulus that has previously been established as a reliable predictor of that outcome. Attentional models of associative learning suggest that blocking reflects a reduction in the attention paid to the blocked cue. We tested this idea in three experiments in which participants were trained in an associative learning task using a blocking procedure. Attention to stimuli was measured 250 ms after onset using an adapted version of the dot probe task. This task was presented at the beginning of each learning trial (Experiments 1 and 2) or in independent trials (Experiment 3). Results show evidence of reduced attention to blocked stimuli (i.e. “attentional blocking”). In addition, this attentional bias correlated with the magnitude of blocking in associative learning, as measured by predictive-value judgments. Moreover, Experiments 2 and 3 found evidence of an influence of learning about predictiveness on memory for episodes involving stimuli. These findings are consistent with a central role of learned attentional biases in producing the blocking effect, and in the encoding of new memories.

Discriminative Fear Learners are Resilient to Temporal Distortions during Threat Anticipation

Discriminative fear conditioning requires learning to dissociate between safety cues and cues that predict negative outcomes yet little is known about what processes contribute to discriminative fear learning. According to attentional models of time perception, processes that distract from timing result in temporal underestimation. If discriminative fear learning only requires learning what cues predict what outcomes, and threatening stimuli distract attention from timing, then better discriminative fear learning should predict greater temporal distortion on threat trials. Alternatively, if discriminative fear learning also reflects a more accurate perceptual experience of time in threatening contexts, discriminative fear learning scores would predict less temporal distortion on threat trials, as time is perceived more veridically. Healthy young adults completed discriminative fear conditioning in which they learned to associate one stimulus (CS+) with aversive electrical stimulation and another stimulus (CS−) with non-aversive tactile stimulation and then an ordinal-comparison timing task during which CSs were presented as task-irrelevant distractors. Consistent with predictions, we found an overall temporal underestimation bias on CS+ relative to CS− trials. Differential skin conductance responses to the CS+ versus the CS− during conditioning served as a physiological index of discriminative fear conditioning and this measure predicted the magnitude of the underestimation bias, such that individuals exhibiting greater discriminative fear conditioning showed less underestimation on CS+ versus CS− trials. These results are discussed with respect to the nature of discriminative fear learning and the relationship between temporal distortions and maladaptive threat processing in anxiety.

Reconciling the influence of predictiveness and uncertainty on stimulus salience: a model of attention in associative learning

Theories of selective attention in associative learning posit that the salience of a cue will be high if the cue is the best available predictor of reinforcement (high predictiveness). In contrast, a different class of attentional theory stipulates that the salience of a cue will be high if the cue is an inaccurate predictor of reinforcement (high uncertainty). Evidence in support of these seemingly contradictory propositions has led to: (i) the development of hybrid attentional models that assume the coexistence of separate, predictiveness-driven and uncertainty-driven mechanisms of changes in cue salience; and (ii) a surge of interest in identifying the neural circuits underpinning these mechanisms. Here, we put forward a formal attentional model of learning that reconciles the roles of predictiveness and uncertainty in salience modification. The issues discussed are relevant to psychologists, behavioural neuroscientists and neuroeconomists investigating the roles of predictiveness and uncertainty in behaviour.