Statistical Learning Mechanisms Research Articles

Temporal dynamics of uncertainty and prediction error in musical improvisation across different periods

Human improvisational acts contain an innate individuality, derived from one’s experiences based on epochal and cultural backgrounds. Musical improvisation, much like spontaneous speech, reveals intricate facets of the improviser’s state of mind and emotional character. However, the specific musical components that reveal such individuality remain largely unexplored. Within the framework of human statistical learning and predictive processing, this study examined the temporal dynamics of uncertainty and surprise (prediction error) in a piece of musical improvisation. This cognitive process reconciles the raw auditory cues, such as melody and rhythm, with the musical predictive models shaped by its prior experiences. This study employed the Hierarchical Bayesian Statistical Learning (HBSL) model to analyze a corpus of 456 Jazz improvisations, spanning 1905 to 2009, from 78 distinct Jazz musicians. The results indicated distinctive temporal patterns of surprise and uncertainty, especially in pitch and pitch-rhythm sequences, revealing era-specific features from the early 20th to the 21st centuries. Conversely, rhythm sequences exhibited a consistent degree of uncertainty across eras. Further, the acoustic properties remain unchanged across different periods. These findings highlight the importance of how temporal dynamics of surprise and uncertainty in improvisational music change over periods, profoundly influencing the distinctive methodologies artists adopt for improvisation in each era. Further, it is suggested that the development of improvisational music can be attributed to the adaptive statistical learning mechanisms. This study explores the period-specific characteristics in the temporal dynamics of improvisational music, emphasizing how artists adapt their methods to resonate with the cultural and emotional contexts of their times. Such shifts in improvisational ways offer a window into understanding how artists intuitively respond and adapt their craft to resonate with the cultural zeitgeist and the emotional landscapes of their respective times.

Unravelling the interplay of Statistical Learning, Top-Down, and Bottom-Up Mechanisms during target selection: Insights from Behavioural and EEG Experiments

Unravelling the interplay of Statistical Learning, Top-Down, and Bottom-Up Mechanisms during target selection: Insights from Behavioural and EEG Experiments

Impaired Face Feature-to-Location Statistical Learning and Single-Feature Discrimination in Developmental Prosopagnosia.

Individuals with developmental prosopagnosia (DP) experience severe face memory deficits that are often accompanied by impairments in face perception. Images of human facial features are better discriminated between when they are presented in the locations on the visual field that they typically appear in while viewing human faces in daily life, than in locations which they do not typically appear (i.e., better performance for eyes in the upper visual field, and better performance for mouths in the lower visual field). These feature-to-location tuning effects (FLEs) can be explained by a retinotopically organised visual statistical learning mechanism. We had a large group of DP participants (N = 64), a control group (N = 74) and a group of individuals with a mild form of DP (N = 58) complete a single-feature discrimination task to determine whether face perception deficits in DP can be accounted for by an impairment in face feature-to-location tuning. The results showed that individuals with DP did not have significant FLEs, suggesting a marked impairment in the underlying visual statistical learning mechanism. In contrast, the mild DP group showed normal FLE effects which did not differ from the control group. Both DP groups had impaired single-feature processing (SFP) as compared to the control group. We also examined the effects of age on FLEs and SFP.

Selective Neural Entrainment Reveals Hierarchical Tuning to Linguistic Regularities in Reading.

Reading is both a visual and a linguistic task, and as such it relies on both general-purpose, visual mechanisms and more abstract, meaning-oriented processes. Disentangling the roles of these resources is of paramount importance in reading research. The present study capitalizes on the coupling of fast periodic visual stimulation and MEG recordings to address this issue and investigate the role of different kinds of visual and linguistic units in the visual word identification system. We compared strings of pseudo-characters; strings of consonants (e.g., sfcl); readable, but unattested strings (e.g., amsi); frequent, but non-meaningful chunks (e.g., idge); suffixes (e.g., ment); and words (e.g., vibe); and looked for discrimination responses with a particular focus on the ventral, occipito-temporal regions. The results revealed sensitivity to alphabetic, readable, familiar, and lexical stimuli. Interestingly, there was no discrimination between suffixes and equally frequent, but meaningless endings, thus highlighting a lack of sensitivity to semantics. Taken together, the data suggest that the visual word identification system, at least in its early processing stages, is particularly tuned to form-based regularities, most likely reflecting its reliance on general-purpose, statistical learning mechanisms that are a core feature of the visual system as implemented in the ventral stream.

Unraveling the complex interplay between statistical learning and working memory in Chinese children with and without dyslexia across different ages.

The relation between statistical learning and working memory in children with developmental dyslexia (DD) remains unclear. This study employed a distributional and a conditional statistical learning experiment and a working memory task to examine this relation in 651 Chinese 6- to 12-year-olds with and without DD (NDD = 199, 101 females; NwoDD = 452, 227 females; participated 2014-2019). Results showed working memory positively associated with recognizing high-predictable and familiar items in both groups, but negatively associated with recognizing unfamiliar items in the DD group only. These findings uncovered the complex interplay between statistical learning and working memory, demonstrating how different working memory abilities in children with and without DD relate to various statistical learning mechanisms at the item level.

Semantic and Syntactic Predictions in Reading Aloud: Are Good Predictors Good Statistical Learners?

Recent research suggests that becoming a fluent reader may partially rely on a domain-general statistical learning (SL) mechanism that allows a person to automatically extract predictable patterns from the sensory input. The goal of the present study was to investigate a potential link between SL and the ability to make linguistic predictions. All previous studies investigated quite general levels of reading ability rather than the dynamic process of making linguistic predictions. We thus used a recently developed predictive reading task, which consisted of having participants read aloud words that were preceded by either semantically or syntactically predictive contexts. To measure the componential nature of SL, we used a visual and an auditory SL task (VSL, ASL) and the classic serial reaction time task (SRT). General reading ability was assessed with a reading speed/comprehension test. The study was conducted online on a sample of 120 participants to make it possible to explore interindividual differences. The results showed only weak and sometimes even negative correlations between the various SL measures. ASL correlated positively and predicted general reading ability but neither semantic nor syntactic prediction effects. Similarly, one of the SRT measures was significantly associated with reading level and reading speed but not with linguistic prediction effects. In sum, there is little evidence that domain-general SL is a good predictor of people's ability to make domain-specific linguistic predictions. In contrast, SL shows a weak but significant association with general reading ability.

Statistical mechanics of learning via reverberation in bidirectional associative memories

We study bi-directional associative neural networks that, exposed to noisy examples of an extensive number of random archetypes, learn the latter (with or without the presence of a teacher) when the supplied information is enough: in this setting, learning is heteroassociative – involving couples of patterns – and it is achieved by reverberating the information depicted from the examples through the layers of the network. By adapting Guerra’s interpolation technique, we provide a full statistical mechanical picture of supervised and unsupervised learning processes (at the replica symmetric level of description) obtaining analytically phase diagrams, thresholds for learning, a picture of the ground-state in plain agreement with Monte Carlo simulations and signal-to-noise outcomes. In the large dataset limit, the Kosko storage prescription as well as its statistical mechanical picture provided by Kurchan, Peliti, and Saber in the eighties is fully recovered. Computational advantages in dealing with information reverberation, rather than storage, are discussed for natural test cases. In particular, we show how this network admits an integral representation in terms of two coupled restricted Boltzmann machines, whose hidden layers are entirely built of by grand-mother neurons, to prove that by coupling solely these grand-mother neurons we can correlate the patterns they are related to: it is thus possible to recover Pavlov’s Classical Conditioning by adding just one synapse among the correct grand-mother neurons (hence saving an extensive number of these links for further information storage w.r.t. the classical auto-associative setting). Likewise, it is enough to equip the BAM’s synaptic coupling with its transpose matrix (and alternate among these two, e.g., by introducing a clock) to let these networks be able count sequences of patterns (still preserving Detailed Balance, at difference w.r.t. the auto-associative case).

Driver Maneuver Interaction Identification with Anomaly-Aware Federated Learning on Heterogeneous Feature Representations

Driver maneuver interaction learning (DMIL) refers to the classification task with the goal of identifying different driver-vehicle maneuver interactions (e.g., left/right turns). Existing conventional studies largely focused on the centralized collection of sensor data from the drivers' smartphones (say, inertial measurement units or IMUs, like accelerometer and gyroscope). Such a centralized mechanism might be precluded by data regulatory constraints. Furthermore, how to enable an adaptive and accurate DMIL framework remains challenging due to (i) complexity in heterogeneous driver maneuver patterns, and (ii) impacts of anomalous driver maneuvers due to, for instance, aggressive driving styles and behaviors. To overcome the above challenges, we propose AF-DMIL, an Anomaly-aware Federated Driver Maneuver Interaction Learning system. We focus on the real-world IMU sensor datasets (e.g., collected by smartphones) for our pilot case study. In particular, we have designed three heterogeneous representations for AF-DMIL regarding spectral, time series, and statistical features that are derived from the IMU sensor readings. We have designed a novel heterogeneous representation attention network (HetRANet) based on spectral channel attention, temporal sequence attention, and statistical feature learning mechanisms, jointly capturing and identifying the complex patterns within driver maneuver behaviors. Furthermore, we have designed a densely-connected convolutional neural network in HetRANet to enable the complex feature extraction and enhance the computational efficiency of HetRANet. In addition, we have designed within AF-DMIL a novel anomaly-aware federated learning approach for decentralized DMIL in response to anomalous maneuver data. To ease extraction of the maneuver patterns and evaluation of their mutual differences, we have designed an embedding projection network that projects the high-dimensional driver maneuver features into low-dimensional space, and further derives the exemplars that represent the driver maneuver patterns for mutual comparison. Then, AF-DMIL further leverages the mutual differences of the exemplars to identify those that exhibit anomalous patterns and deviate from others, and mitigates their impacts upon the federated DMIL. We have conducted extensive driver data analytics and experimental studies on three real-world datasets (one is harvested on our own) to evaluate the prototype of AF-DMIL, demonstrating AF-DMIL's accuracy and effectiveness compared to the state-of-the-art DMIL baselines (on average by more than 13% improvement in terms of DMIL accuracy), as well as fewer communication rounds (on average 29.20% fewer than existing distributed learning mechanisms).

The relation between implicit statistical learning and proactivity as revealed by EEG

Environmental events often occur on a probabilistic basis but can sometimes be predicted based on specific cues and thus approached proactively. Incidental statistical learning enables the acquisition of knowledge about probabilistic cue-target contingencies. However, the neural mechanisms of statistical learning about contingencies (SLC), the required conditions for successful learning, and the role of implicit processes in the resultant proactive behavior are still debated. We examined changes in behavior and cortical activity during an SLC task in which subjects responded to visual targets. Unbeknown to them, there were three types of target cues associated with high-, low-, and zero target probabilities. About half of the subjects spontaneously gained explicit knowledge about the contingencies (contingency-aware group), and only they showed evidence of proactivity: shortened response times to predictable targets and enhanced event-related brain responses (cue-evoked P300 and contingent negative variation, CNV) to high probability cues. The behavioral and brain responses were strictly associated on a single-trial basis. Source reconstruction of the brain responses revealed activation of fronto-parietal brain regions associated with cognitive control, particularly the anterior cingulate cortex and precuneus. We also found neural correlates of SLC in the contingency-unaware group, but these were restricted to post-target latencies and visual association areas. Our results document a qualitative difference between explicit and implicit learning processes and suggest that in certain conditions, proactivity may require explicit knowledge about contingencies.

Statistical mechanics of continual learning: Variational principle and mean-field potential.

An obstacle to artificial general intelligence is set by continual learning of multiple tasks of a different nature. Recently, various heuristic tricks, both from machine learning and from neuroscience angles, were proposed, but they lack a unified theory foundation. Here, we focus on continual learning in single-layered and multilayered neural networks of binary weights. A variational Bayesian learning setting is thus proposed in which the neural networks are trained in a field-space, rather than a gradient-ill-defined discrete-weight space, and furthermore, weight uncertainty is naturally incorporated, and it modulates synaptic resources among tasks. From a physics perspective, we translate variational continual learning into a Franz-Parisi thermodynamic potential framework, where previous task knowledge serves as a prior probability and a reference as well. We thus interpret the continual learning of the binary perceptron in a teacher-student setting as a Franz-Parisi potential computation. The learning performance can then be analytically studied with mean-field order parameters, whose predictions coincide with numerical experiments using stochastic gradient descent methods. Based on the variational principle and Gaussian field approximation of internal preactivations in hidden layers, we also derive the learning algorithm considering weight uncertainty, which solves the continual learning with binary weights using multilayered neural networks, and performs better than the currently available metaplasticity algorithm in which binary synapses bear hidden continuous states and the synaptic plasticity is modulated by a heuristic regularization function. Our proposed principled frameworks also connect to elastic weight consolidation, weight-uncertainty modulated learning, and neuroscience-inspired metaplasticity, providing a theoretically grounded method for real-world multitask learning with deep networks.

Tone Deafness in Music Does Not Preclude Distributional Learning of Nonnative Tonal Languages in Individuals With Congenital Amusia.

Previous studies have shown that individuals with congenital amusia exhibit deficient pitch processing across music and language domains. This study investigated whether adult Chinese-speaking listeners with amusia were still able to learn Thai lexical tones based on stimulus frequency of statistical distribution via distributional learning, despite their degraded lexical tone perception. Following a pretest-training-posttest design, 21 amusics and 23 typical, musically intact listeners were assigned into bimodal and unimodal distribution conditions. Listeners were asked to discriminate minimal pairs of Thai mid-level tone and falling tone superimposed on variable base syllables and uttered by different speakers. The perceptual accuracy for each test session and improvement from pretest to posttest were collected and analyzed between the two groups using generalized mixed-effects models. When discriminating Thai lexical tones, amusics were less accurate than typical listeners. Nonetheless, similarly to control listeners, perceptual gains from pretest to posttest were observed in bimodally rather than unimodally trained amusics, as evidenced by both trained and nontrained test words. Amusics are able to learn lexical tones in a second or foreign context of speech. This extends previous research by showing that amusics' distributional learning of linguistic pitch remains largely preserved despite their degraded pitch processing. It is thus likely that manifestations of amusia in speech could not result from their abnormal statistical learning mechanism. This study meanwhile provides a heuristic approach for future studies to apply this paradigm into amusics' treatment to mitigate their pitch-processing disorder.

Statistical learning speeds visual search: More efficient selection, or faster response?

Learning statistical regularities of target objects speeds visual search performance. However, we do not yet know whether this statistical learning effect is driven by biasing attentional selection at the early perceptual stage of processing, as theories of attention propose, or by improving the decision-making efficiency at a late response-related stage. Leveraging the high-temporal resolution of the event-related potential (ERP) technique, we had 16 human observers perform a visual search task where we inserted a fine-grained statistical regularity that the target shapes appeared in different colors with six unique probabilities. Observers unintentionally learned these regularities such that they were faster to report targets that appeared in more likely target colors. The observers' ERPs showed that this learning effect resulted in subjects making faster decisions about the target presence, and not by preferentially shifting attention to more rapidly select likely target colors, as is often assumed by the attentional selection account, supporting a post-selection account for statistical learning of the probabilistic regularities of target features. These results provide fundamental insights into the attentional control mechanisms of statistical learning. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Sensitivity to word endings as probabilistic orthographic cues to lexical stress among English as second language learners.

Assigning stress to polysyllabic words is a crucial aspect of reading aloud in English. Previous research demonstrated that native English speakers are sensitive to word endings as probabilistic orthographic cues to stress assignment. However, little is known about if second language (L2) learners of English are sensitive to word endings as cues to lexical stress. The current study investigated whether native Chinese-speakers learning English as a second language (ESL) are sensitive to word endings as probabilistic orthographic cues to lexical stress. Our ESL learners demonstrated sensitivity to word endings as cues in a stress-assignment task and a naming task. With the increase in language proficiency, ESL learners responded more accurately in the stress-assignment task. Moreover, stress position and language proficiency moderated the strength of the sensitivity, with a trochaic bias and better proficiency leading to better sensitivity in the stress-assignment task. However, as language proficiency increased, participants' naming speed became faster in iambic but slower in trochaic patterns, reflecting the learners' fledgling knowledge about the specific stress patterns associated with varying orthographic cues, especially in a demanding naming task. Taken together, the evidence from our ESL learners fits in the proposed statistical learning mechanism, that is, L2 learners are able to implicitly extract statistical regularities from linguistic materials, the orthographic cues to lexical stress in our study. Stress position and language proficiency both play a role in developing this sensitivity.

Computational and neural mechanisms of statistical pain learning

Pain invariably changes over time. These fluctuations contain statistical regularities which, in theory, could be learned by the brain to generate expectations and control responses. We demonstrate that humans learn to extract these regularities and explicitly predict the likelihood of forthcoming pain intensities in a manner consistent with optimal Bayesian inference with dynamic update of beliefs. Healthy participants received probabilistic, volatile sequences of low and high-intensity electrical stimuli to the hand during brain fMRI. The inferred frequency of pain correlated with activity in sensorimotor cortical regions and dorsal striatum, whereas the uncertainty of these inferences was encoded in the right superior parietal cortex. Unexpected changes in stimulus frequencies drove the update of internal models by engaging premotor, prefrontal and posterior parietal regions. This study extends our understanding of sensory processing of pain to include the generation of Bayesian internal models of the temporal statistics of pain.

Remembering the pattern: A longitudinal case study on statistical learning in spatial navigation and memory consolidation

Distinct brain systems are thought to support statistical learning over different timescales. Regularities encountered during online perceptual experience can be acquired rapidly by the hippocampus. Further processing during offline consolidation can establish these regularities gradually in cortical regions, including the medial prefrontal cortex (mPFC). These mechanisms of statistical learning may be critical during spatial navigation, for which knowledge of the structure of an environment can facilitate future behavior. Rapid acquisition and prolonged retention of regularities have been investigated in isolation, but how they interact in the context of spatial navigation is unknown. We had the rare opportunity to study the brain systems underlying both rapid and gradual timescales of statistical learning using intracranial electroencephalography (iEEG) longitudinally in the same patient over a period of three weeks. As hypothesized, spatial patterns were represented in the hippocampus but not mPFC for up to one week after statistical learning and then represented in the mPFC but not hippocampus two and three weeks after statistical learning. Taken together, these findings suggest that the hippocampus may contribute to the initial extraction of regularities prior to cortical consolidation.

Detecting non-adjacent dependencies is the exception rather than the rule.

Statistical learning refers to our sensitivity to the distributional properties of our environment. Humans have been shown to readily detect the dependency relationship of events that occur adjacently in a stream of stimuli but processing non-adjacent dependencies (NADs) appears more challenging. In the present study, we tested the ability of human participants to detect NADs in a new Hebb-naming task that has been proposed recently to study regularity detection in a noisy environment. In three experiments, we found that most participants did not manage to extract NADs. These results suggest that the ability to learn NADs in noise is the exception rather than the rule. They provide new information about the limits of statistical learning mechanisms.

Nonverbal Action Interpretation Guides Novel Word Disambiguation in 12-Month-Olds.

Whether young infants can exploit sociopragmatic information to interpret new words is a matter of debate. Based on findings and theories from the action interpretation literature, we hypothesized that 12-month-olds should distinguish communicative object-directed actions expressing reference from instrumental object-directed actions indicative of one's goals, and selectively use the former to identify referents of novel linguistic expressions. This hypothesis was tested across four eye-tracking experiments. Infants watched pairs of unfamiliar objects, one of which was first targeted by either a communicative action (e.g., pointing) or an instrumental action (e.g., grasping) and then labeled with a novel word. As predicted, infants fast-mapped the novel words onto the targeted objects after pointing (Experiments 1 and 4) but not after grasping (Experiment 2) unless the grasping action was preceded by an ostensive signal (Experiment 3). Moreover, whenever infants mapped a novel word onto the object indicated by a communicative action, they tended to map a different novel word onto the distractor object, displaying a mutual exclusivity effect. This reliance on nonverbal action interpretation in the disambiguation of novel words indicates that sociopragmatic inferences about reference likely supplement associative and statistical learning mechanisms from the outset of word learning.

Learning Higher-Order Transitional Probabilities in Nonhuman Primates.

The extraction of cooccurrences between two events, A and B, is a central learning mechanism shared by all species capable of associative learning. Formally, the cooccurrence of events A and B appearing in a sequence is measured by the transitional probability (TP) between these events, and it corresponds to the probability of the second stimulus given the first (i.e., p(B|A)). In the present study, nonhuman primates (Guinea baboons, Papio papio) were exposed to a serial version of the XOR (i.e., exclusive-OR), in which they had to process sequences of three stimuli: A, B, and C. In this manipulation, first-order TPs (i.e., AB and BC) were uninformative due to their transitional probabilities being equal to .5 (i.e., p(B|A) = p(C|B) = .5), while second-order TPs were fully predictive of the upcoming stimulus (i.e., p(C|AB) = 1). In Experiment 1, we found that baboons were able to learn second-order TPs, while no learning occurred on first-order TPs. In Experiment 2, this pattern of results was replicated, and a final test ruled out an alternative interpretation in terms of proximity to the reward. These results indicate that a nonhuman primate species can learn a nonlinearly separable problem such as the XOR. They also provide fine-grained empirical data to test models of statistical learning on the interaction between the learning of different orders of TPs. Recent bioinspired models of associative learning are also introduced as promising alternatives to the modeling of statistical learning mechanisms.

Заключительные этапы проектирования системы сбора и предиктивного анализа данных социальных медиа

This article discusses the design of a system for collecting and predictive analysis of social media data. With the development of the Internet, as well as social media, it has become easier to access and distribute information because the network users themselves are both creators and recipients of varying information. The main type of social media is social networks. Facebook, VK, Instagram, YouTube, Twitter, Odnoklassniki, WhatsApp and Telegram messengers are among the most well-known ones. The most important functions of social media are to influence the perception, attitude and final behavior of consumers. Predictive analytics is based on automatic search for connections, anomalies and patterns between various factors. To form a predictive model, a large set of statistical modeling methods, data mining, machine learning, neural networks and other mechanisms are used. Together with various methods of collecting information from Internet resources, such as parsing and social network APIs, predictive analytics can offer the most interesting sources of information for the user. In order to combine the methods of predictive analysis and data collection methods, it is necessary to take a detailed approach to the system design process. In this paper, special attention is paid to the detailed description of the second of the main parts of the system (the analysis subsystem). In addition, the full architecture and algorithm of operation are highlighted. The results obtained will be used in further development, and it is planned to use them in full. Working on this topic will facilitate the process of subsequent testing and research of the system.

Effects of healthy aging and left hemisphere stroke on statistical language learning

ABSTRACT Spoken sentences are continuous streams of sound, without reliable acoustic cues to word boundaries. We have previously proposed that language learners identify words via an implicit statistical learning mechanism that computes transitional probabilities between syllables. Neuroimaging studies in healthy young adults associate this learning with left inferior frontal gyrus, left arcuate fasciculus, and bilateral striatum. Here, we test the effects of healthy aging and left hemisphere (LH) injury on statistical learning. Following 10-minute exposure to an artificial language, participants rated familiarity of Words, Part-words (sequences spanning word boundaries), and Non-words (unfamiliar sequences). Young controls (N = 14) showed robust learning, rating Words > Part-words > Non-words. Older controls (N = 28) showed this pattern to a weaker degree. Stroke survivors (N = 24) as a group showed no learning. A lesion comparison examining individual differences revealed that “non-learners” are more likely to have anterior lesions. Together, these findings demonstrate that word segmentation is sensitive to healthy aging and LH injury.