Search

The ever-happening disruptive events interrupt the operationalisation of manufacturing organisations resulting in stalling the production flow and depleting societies with products. Advancements in cutting-edge technologies, viz. blockchain, artificial intelligence, virtual reality, digital twin, etc. have attracted the practitioners' attention to overcome such saddled conditions. This study attempts to explore the role of artificial intelligence (AI) in building the resilience of production function at manufacturing organisations during a COVID-19 pandemic. In this regard, a decision support system comprising an integrated voting analytical hierarchy process (VAHP) and Bayesian network (BN) method is developed. Initially, through a comprehensive literature review, the critical success factors (CSFs) for implementing AI are determined. Further, using a multi-criteria decision-making (MCDM) based VAHP, CSFs are prioritised to determine the prominent ones. Finally, the machine learning based BN method is adopted to predict and understand the influential CSFs that help achieve the highest production resilience. The present research is one of the early attempts to know the essence of AI and bridge the interplay between AI and production resilience during COVID-19. This study can support academicians, practitioners, and decision-makers in assessing the AI adoption in manufacturing organisations and evaluate the impact of different CSFs of AI on production resilience. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of stochastic jointly scheduling of resumable jobs and preventive maintenance on a single machine, subject to random breakdowns, to minimise the earliness-tardiness cost. The main objective is to investigate using trending machine learning-based methods compared to stochastic optimisation approaches. We propose two different methods from both fields as we solve the same problem firstly with a stochastic dynamic programming model in an approximation way, then with an attention-based deep reinforcement learning model. We conduct a detailed experimental study according to solution quality, run time, and robustness to analyse their performances compared to those of an existing approach in the literature as a baseline. Both algorithms outperform the baseline. Moreover, the machine learning-based algorithm outperforms the stochastic dynamic programming-based heuristic as we report up to 30.5% saving in total cost, a reduction of computational time from 67 min to less than $ 1s $ 1 s on big instances, and a better robustness. These facts highlight clearly its potential for solving such problems. [ABSTRACT FROM AUTHOR]

Disparate impact rules are formally neutral but indirectly discriminate against protected groups (i.e., by targeting a characteristic that is more prevalent in a given group). Because these rules are not obviously malicious, they have been widely enacted to circumvent policies against explicit discrimination. In a series of four experiments, we show that adults and children are sensitive to the moral implications of disparate impact rules. However, we also find that they are more accepting of these rules when strong justification is provided, compared to rules with no justification. Crucially, demographic differences also impact people's judgments of disparate impact rules and their creators. We find that conservatives and those from groups not directly affected by the rule tend to be more accepting of it. By studying people's reasoning about disparate impact rules, this work aims to identify the mechanisms by which these rules may evade detection. Finally, we discuss how these insights may inform the development of interventions that highlight the problematic effects of indirectly discriminatory policies.

Early word learning involves mapping individual words to their meanings and building organized semantic representations among words. Previous corpus-based studies (e.g., using text from websites, newspapers, child-directed speech corpora) demonstrated that linguistic information such as word co-occurrence alone is sufficient to build semantically organized word knowledge. The present study explored two new research directions to advance understanding of how infants acquire semantically organized word knowledge. First, infants in the real world hear words surrounded by contextual information. Going beyond inferring semantic knowledge merely from language input, we examined the role of extra-linguistic contextual information in learning semantic knowledge. Second, previous research relies on large amounts of linguistic data to demonstrate in-principle learning, which is unrealistic compared with the input children receive. Here, we showedthat incorporating extra-linguistic information provides an efficient mechanism through which semantic knowledge can be acquired with a small amount of data infants perceive in everyday learning contexts, such as toy play.

Learning names for novel objects has been shown to be impacted by the context in which they appear. Manipulations of context, therefore, provide a key pathway to explore these learning dynamics. Here we use a neural process model that instantiates the details of ‘context' to generate novel, counterintuitive predictions about how similarity in object properties influence learning. Specifically, we use a dynamic field model, WOLVES, to simulate and predict learning in a cross-situational word learning task in two conditions: one where the two objects presented on each learning trial are metrically similar in a property (‘NEAR') and another condition where the two objects are always dissimilar (‘FAR'). WOLVES predicts—counterintuitively—that participants should learn better in the ‘NEAR' condition (where objects are potentially confusable) than in ‘FAR' condition (where objects are distinctive). We then tested this prediction empirically, finding support for the novel prediction. This study shows the utility of process models which instantiate the details of ‘context' during learning and provides support for WOLVES. We know of no other theory of cross-situational word learning that captures these novel findings.

Our choices often require us to prioritize some features of our rich sensory experience over others. Past work suggests that humans solve this problem by focusing on relevant information while discarding that which is irrelevant. Yet learning which features to prioritize requires extensive experience. Moreover, features that are irrelevant now may become relevant in the future. One way to address these issues is by sampling individual richly encoded experiences from episodic memory. Here we hypothesize that episodic memory is used to guide decisions based on multiple features of past events. We test this hypothesis using an experiment in which participants made choices about the value of features that were present in multiple past experiences. We find evidence suggesting that participants used episodic memories to flexibly access features of past events during decision making. Overall, these results suggest that episodic memory promotes adaptive decisions when knowledge of multiple features is necessary.

Speech perception requires inferring category membership from varied acoustic cues, with listeners adeptly adjusting cue utilization upon encountering novel speech inputs. This adaptivity has been examined through the dimension-based statistical learning (DBSL) paradigm, which reveals that listeners can quickly de-emphasize secondary cues when cue correlations deviate from long-term expectations, resulting in cue-reweighting. Although multiple accounts of cue-reweighting have been proposed, direct comparisons of these accounts against human perceptual data are scarce. This study evaluates three computational models–cue normalization, Bayesian ideal adaptor, and error-driven learning–against classic DBSL findings to elucidate how cue reweighting supports adaptation to new speech patterns. These models differ in how they map cues onto categories for categorization and in how recent exposure to atypical input patterns influences this mapping. Our results show that both the error-driven learning and ideal adaptor models effectively capture the key patterns of cue-reweighting phenomena, whereas prelinguistic cue normalization does not. This comparison not only highlights the models' relative efficacy but also advances our understanding of the dynamic processes underlying speech perception adaptation.

To succeed in social situations, we must learn how social cues predict subsequent events. How do we quickly form associations between a variety of social cues, such as individuals signaling their current emotion state, and social outcomes? To address this question, we developed a task in which participants viewed images of individuals conveying different emotions and searched among these images to gain rewards. Rewards were associated with either individuals' identities or emotion cues. Across four experiments (N=720), individuals learned about rewards more efficiently from individual identity cues versus a wide variety of emotion cues. Participants also generalized cue-outcome associations more easily for individuals versus emotions. Learning was worse if participants experienced a change in the association rule, especially when switching from learning individual-based associations to emotion-based associations. Overall, we show that social cue type influences how associations between cues and rewards are learned, with implications for understanding learning in social contexts.

Humans learn and generate languages, music, games, and seemingly limitless varieties of other structures across domains. Unlike many AI systems, we often do so from little data. How do we learn such large varieties of richly structured representations so efficiently? One possibility is that people ``learn by programming,'' synthesizing data-generating algorithms to explain what we observe. We examine the nature and origins of this learning mechanism in adults, children, and nonhuman primates (macaque monkeys), using a highly unconstrained sequence prediction task. Although adults and children quickly learn many richly structured sequences, monkeys learn only the simplest sequences (e.g. lines). We test multiple learning models, finding that adults are best explained by a ``Language of Thought''-like program-learning model and monkeys by a simpler extrapolation strategy. Children exhibit varied learning strategies but are best fit in aggregate by an intermediately expressive model. Paper available at https://sites.google.com/view/patternlearning.

Although well-intentioned diversity, equity, and inclusion (DEI) initiatives aim to increase minority representation in elite groups, they can sometimes backfire by causing candidates to question whether they were selected for merit. Prior work in social psychology suggests that this effect is driven mainly by stereotype threat. Here, we propose a novel cognitive framework: DEI initiatives backfire due to causal inference. Specifically, when candidates hear that they were selected based on a DEI initiative and/or enter a group where they are a minority, they may hypothesize that their selection was based more on their identity and less on their merit. Across two pre-registered experiments manipulating selection messages (DEI vs. merit) and statistical gender representation (represented or under-represented in the selected group), we find evidence in favor of our hypothesis. DEI messages and under-representation independently caused successful candidates to attribute their selection more to their identity and less to their merit but did not directly impact perceptions of competence. A third pre-registered experiment revealed that women selectively rated themselves as less competent in DEI contexts when selection tasks were more difficult. Taken together, this work shows that people make different causal hypotheses about their selection into elite groups based on DEI messages and group composition in conjunction with selection task difficulty and their social identity. Importantly, this work paves the way for designing DEI-based initiatives that license more helpful causal inferences about success to ensure that minority candidates thrive in their positions.

While the ability of recursion is considered universally available, there are considerable cross- and with-linguistic differences regarding the rules for recursive embedding, which must be learned from language-specific experience. One proposal argues that the recursivity of a structure is learnable as a productive generalization from distributional information in non-embedded input, and adults can indeed use such distributional cues to acquire recursive structures in an artificial language. However, it is not yet known whether children can use distributional information in this way. In this work, we examine children's distributional learning of recursive structures. We exposed children to non-embedded sentences in an artificial grammar, where we manipulated the productivity of the structure across conditions. At test, we found that children exposed to productive input were more likely to accept recursively embedded sentences unattested during the exposure phase. The results suggest that children can make use of distributional information to acquire recursive structures.

High-dimensional deep neural network representations of images and concepts can be aligned to predict human annotations of diverse stimuli. However, such alignment requires the costly collection of behavioral responses, such that, in practice, the deep-feature spaces are only ever sparsely sampled. Here, we propose an active learning approach to adaptively sample experimental stimuli to efficiently learn a Bayesian matrix factorization model with deep side information. We observe a significant efficiency gain over a passive baseline. Furthermore, with a sequential batched sampling strategy, the algorithm is applicable not only to small datasets collected from traditional laboratory experiments but also to settings where large-scale crowdsourced data collection is needed to accurately align the high-dimensional deep feature representations derived from pre-trained networks. This provides cost-effective solutions for collecting behavioral data and generating high-quality predictions in large-scale behavioral and cognitive studies.

Adult language learners have difficulty segmenting words from continuous speech when the phonology is unfamiliar. Since speaker variability is known to improve acquisition of novel language structures, it could be processed in ways that bootstrap phonological patterns and enhance learners' ability to segment words. To test this, the present experiment examined adult participants' learning of a stream of statistically determined tri-syllabic words that were spoken by one or multiple speakers. Syllables were constructed with either English phonology or non-English phonology. Two tasks (target detection and two-alternative forced choice) assessed the extent of listeners' sensitivity to language patterns and word segmentation. Results suggest speaker variability negatively impacted learners' ability to track the underlying statistics. 2AFC word segmentation performance was poor—independent of speaker number; it is hypothesized that attentional demands of the target detection task conflicted with statistical word segmentation mechanisms.

The speech-to-song illusion is a robust psychological phenomenon whereby a spoken sentence sounds increasingly more musical as it is repeated. Despite decades of research, a complete formal account of this transformation is still lacking, and some of its nuanced characteristics, namely, that certain phrases appear to transform while others do not, is not well understood. Here we provide a formal account of this phenomenon, by recasting it as a statistical inference whereby a rational agent attempts to decide whether a sequence of utterances is more likely to have been produced in a song or speech. Using this approach and analyzing song and speech corpora, we further introduce a novel prose-to-lyrics illusion that is purely text-based. In this illusion, simply duplicating written sentences makes them appear more like song lyrics. We provide robust evidence for this new illusion in both human participants and large language models.

Disparate impact rules are formally neutral but indirectly discriminate against protected groups (i.e., by targeting a characteristic that is more prevalent in a given group). Because these rules are not obviously malicious, they have been widely enacted to circumvent policies against explicit discrimination. In a series of four experiments, we show that adults and children are sensitive to the moral implications of disparate impact rules. However, we also find that they are more accepting of these rules when strong justification is provided, compared to rules with no justification. Crucially, demographic differences also impact people's judgments of disparate impact rules and their creators. We find that conservatives and those from groups not directly affected by the rule tend to be more accepting of it. By studying people's reasoning about disparate impact rules, this work aims to identify the mechanisms by which these rules may evade detection. Finally, we discuss how these insights may inform the development of interventions that highlight the problematic effects of indirectly discriminatory policies.

Learning how to make decisions from experience is often studied using probabilistic outcome prediction or choice tasks, as in conditioning, reward learning, or risky gambles (e.g., response A provides reward in 75% of the cases, response B in 25% over repeated trials with feedback). One debated phenomenon in such tasks is that of negative recency, describing that learners expect the rare event after observing a streak of common events (e.g., Gamblers fallacy). Here, we show that this behavior, despite instructing participants to use a visual stimulus, also occurs in probabilistic single-cue conditioning training, where participants predicted whether digging at a specific location on a plane (visual cue) leads to finding a Vase or Nothing (events), when they received reward for correct predictions. We manipulated reward magnitude in three conditions (equal for both common and rare events vs. high for common event vs. high for rare event, between factor). We further manipulated whether the label of the rare event was framed as event (finding a Vase) or non-event (finding Nothing; between factor). The results suggest, that reward magnitude affected the emergence of negative recency, being most prevalent when correctly predicting the rare event yielded a high reward, and least prevalent when the common event yielded a high reward. Interestingly, the event label instead rather affected when the rare event was expected, such that common Vase runs were expected to end earlier than common Nothing runs. We discuss the findings from conditioning and economic perspectives, generally concerning experience-based learning.

The effect of syntactic priming exhibits three well-documented empirical properties: the lexical boost, the inverse frequency effect, and the asymmetrical decay. We aim to show how these three empirical phenomena can be reconciled in a general learning framework, the hierarchical Bayesian model (HBM). The model represents syntactic knowledge in a hierarchical structure of syntactic statistics, where a lower level represents the verb-specific biases of syntactic decisions, and a higher level represents the abstract bias as an aggregation of verb-specific biases. This knowledge is updated in response to experience by Bayesian inference. In simulations, we show that the HBM captures the above-mentioned properties of syntactic priming. The results indicate that some properties of priming which are usually explained by a residual activation account can also be explained by an implicit learning account. We also discuss the model's implications for the lexical basis of syntactic priming.

Distributional models (such as neural network language mod- els) have been successfully used to model a wide range of lin- guistic semantic behaviors. However, they lack a way to dis- tinctly represent different kinds of semantic relations within a single semantic space. Here, we propose that neural network language models can sensibly be interpreted as representing syntagmatic (co-occurrence) relations using their input-output mappings, and as representing paradigmatic (similarity) rela- tions using the similarity of their internal representations. We tested and found support for this hypothesis on four neural net- work architectures (SRNs, LSTMs, Word2Vec and GPT-2) us- ing a carefully constructed artificial language corpus. Using this corpus, we show that the models display interesting but understandable differences in their ability to represent these two kinds of relationships. This work demonstrates distribu- tional models can simultaneously learn multiple kinds of re- lationships, and that systematic investigation of these models can lead to a deeper understanding of how they work.

Age-related memory decline is a multifaceted and heterogeneous process. Previous studies on working memory and episodic memory have demonstrated that older participants' memory for item-context bindings (e.g. the location in which an object appeared) drops dramatically, while memory for individual items is relatively preserved. Here, we extend this research in two ways: first, we study memory for ordered object sequences with spatial context, rather than single objects. Second, we investigate how blocked versus interleaved learning curricula affect independent (or marginal) sequence memory (i.e., which objects appeared, and which spatial locations were seen) versus joint sequence memory (which objects appeared where) for older versus younger adults. Across two behavioral experiments with 108 younger (18-35 years) and 100 older (over 65 years) adults, we found better memory for object sequences than position sequences and worst performance for joint object-position sequence reports in both age-groups. Notably, age differences in memory performance followed the same pattern, being least pronounced for sequential object memory and most for joint object-position sequences. Changing the learning curriculum such that either object or spatial location sequences repeated across times, rather than occurring in an interleaved fashion, improved memory performance in both age groups, but had a stronger effect on older than younger adults, suggesting that blocked learning curricula can help older adults with reallocation of limited cognitive resources.

Searching through memory is mediated by complex interactions between the underlying mental lexicon and the processes that operate on this lexicon. However, these interactions are difficult to study due to the effortless manner in which neurotypical individuals perform cognitive tasks. In this work, we examine these interactions within a sample of prelingually deaf individuals with cochlear implants and normal hearing individuals who were administered the verbal fluency task for the "animals" category. Specifically, we tested how different candidates for underlying mental lexicons and processes account for search behavior within the verbal fluency task across the two groups. The models learned semantic representations from different combinations of textual (word2vec) and speech-based (speech2vec) information. The representations were then combined with process models of memory search based on optimal foraging theory that incorporate different lexical sources for transitions within and between clusters of items produced in the fluency task. Our findings show that semantic, word frequency, and phonological information jointly influence search behavior and highlight the delicate balance of different lexical sources that produces successful search outcomes.

Drawing on the principles of associative learning theory and positing a statistical foundation for language acquisition, this paper investigates the independent contributions of the predictive validities of verbal and constructional cues in English native and non-native speakers' mental representations of the English as-predicative construction. This is examined through two experiments: a sentence completion task targeting constructional outcome retrieval (Experiment 1), and a gap-fill schema task with a focus on verb retrieval (Experiment 2). The results demonstrate that both cues are integrated in parallel when eliciting a constructional outcome (Experiment 1), but only construction cue validity plays a role in eliciting verbal outcomes (Experiment 2). Verb frequency and voice additionally contribute to the retrieval of verbal and constructional information in distinct manners. The present study raises discussions about distributional cue integration in forward versus backward retrieval of linguistic information, in addition to emphasizing the importance of considering cross-linguistic factors in future research.

The present study investigated online neural indices of statistical learning of silent speech. Adult participants were exposed to naturally mouthed, silent syllable streams in an artificial language in two conditions. In one condition, 12 syllables occurred randomly; in the other the syllables were structured into four syllable triplets, i.e. statistical words. In the recorded EEG signal, phase synchronisation in neural oscillations was assessed at the rate of syllables and words occurring in the exposure streams. Largest phase synchronisation was detected for the word rate during exposure to the structured stream. Moreover, the neural synchronisation to word rate increased throughout the exposure within the structured stream. In a behavioural post-test, however, no learning effects were detected. The EEG results demonstrate sensitivity to statistical regularities in viewed silent speech. These findings indicate that statistical learning in speech and language can be effectively measured online even in the absence of auditory cues.

What sorts of graphical formats best convey effect size and degree of certainty of a finding? Confidence intervals are commonly used to show uncertainty, yet lay people and experts fail to correctly interpret their meaning. There has been a recent push to present individual data points rather than only presenting aggregated summary statistics (e.g., means, confidence intervals, lines of best fit). But it is unclear how well people can aggregate raw data presented in a graphical format. Across two studies, we presented participants with hypothetical study outcomes of two independent groups in three graph styles: dot plots, mean with 95% confidence interval (CI) plots, combined plots, and bee plots. We asked participants to make judgments about the effect size using the Common Language Effect Size or Bayes Factors. Participants were more likely to underestimate effect sizes and Bayes Factors for dot plots and bee plots compared to mean + 95% CI plots and combined plots. These findings suggest that people have trouble making statistical inferences when presented with raw data points in graphs.

Understanding abstraction is a stepping stone towards understanding intelligence. We ask the question: How do abstract representations arise when learning sequences? From a normative perspective, we show that abstraction is necessary for an intelligent agent when the perceptual sequence contains objects of similar interaction properties appearing in identical contexts. A rational agent should identify categories of objects of similar properties as an abstract concept, enabling the discovery of higher-order sequential relations that span a longer part of the sequence. We propose a hierarchical variable learning model (HVM) that learns chunks and abstract concepts from sequential data in a cognitively plausible manner. HVM gradually discovers abstraction via a conjunction of variable discovery and chunking, resembling the process of concept discovery during development. In a sequence recall experiment that demands learning and transferring variables, we observe that the model's sequence complexity can explain human behavior in a sequence memorization experiment.

Many human-invented compositional systems (e.g., language, mathematics) embody hierarchical relational structures. How exactly these structures are acquired during learning remains an open question. Here, we examine how the structure of a system engages learners' attention and learning. Participants (N=88) learned an artificial language that describes novel combinations of unknown visual symbols while their eye movements were recorded. Participants were randomly assigned to one of two conditions. The ‘More' condition had three latent rules that connected components in verbal input to visual input. In contrast, the ‘Less' condition had only one latent rule. Despite having more regularities to learn, the ‘More' condition performed as well as the ‘Less' condition. Eye movement data further revealed that participants in the ‘More' condition selectively attended to target symbols more than those in the ‘Less' condition. These results suggest a counterintuitive ‘More is More' principle: the presence of multiple regularities organizes attention and potentiates learning.

Languages across the world organize semantic categories in many ways. Research in semantic typology and efficient communication has shown that languages tend to be shaped by pressures for communicative efficiency. It was recently proposed, in addition to this principle of efficiency, that the cross-linguistic prevalence of a system is explained by considering and formalizing the Typological Prevalence Hypothesis. This recent research found that the interaction between communicative and developmental pressures infers the prevalence of color-naming systems across the world better than phylogenetic relatedness alone. However, it is not yet clear whether the information-theoretic framework developed by the authors can explain the typological prevalence of non-perceptual categories. Therefore, we extend this model to kinship systems to test if this formalization of the Typological Prevalence Hypothesis can generalize to other semantic domains.

Adjusting expectations in response to errors is a cornerstone of several learning theories (Rescorla & Wagner, 1972; Rumelhart et al., 1986). Grimmick (2019) shows that individuals deploy attention during cross-situational word learning based on the strength of the error signal. The current study introduced an equal number of accurate and inaccurate expectations about word-referent pairs. This study manipulated the difficulty of cross-situational word learning trials to examine whether the impact of errors differs depending on task demands. Individuals learned the initially accurate items better than the initially inaccurate ones. Manipulating the demands during word learning did not significantly impact the tendency to benefit from accuracy. This research is part of an ongoing project. This ongoing research explores how individual differences in vocabulary, inhibition, and working memory abilities interact with contextual factors, such as task difficulty, as individuals learn word-referent pairs that violate their expectations.

Consider a classic statistical learning (SL) paradigm, where participants hear an uninterrupted stream of syllables in seemingly random order. In fact, the sequence is generated by repeating 4 word-like patterns, each comprised of 3 syllables. After brief exposure, adults and infants can discriminate ‘words' from the sequence from other syllable sequences (‘nonwords' that did not occur in exposure). If syllables have a fixed duration (e.g., 333.3 ms), syllable rate is fixed (e.g., 3/s or 3hz) and so is word rate (e.g., 1hz). If EEG is acquired during exposure, neural phase-locking is observed, initially to the syllable rate, and gradually to the word rate. This has been interpreted as a neural index of word learning. We tested whether two models that can simulate human SL behavior could simulate neural entrainment (Simple Recurrent Net- works [SRNs] or multi-layer perceptrons [MLPs, feedforward neural networks]). Both models could, although SRNs provided a better fit to correlations observed between entrainment and behavior. We also discovered that raw input sequences (even for a single syllable) have rhythmic properties that generate apparent ‘entrainment' when treated like EEG signals – without learning. We discuss theoretical implications for SL and challenges for interpreting phase-locked entrainment.

Children learn nouns more readily than verbs in early development. Research on candidate explanations for this noun advantage has suggested that, while noun meanings can be easily gleaned from their observational contexts, verb meanings require access to their syntactic constructions, which remain inaccessible until later in development (Gentner, 2006; McDonough et al., 2011; Piccin & Waxman, 2007). This study asks whether previous demonstrations of tenuous verb learning from their observational contexts are partly due to the assessment method. In an adapted version of the Human Simulation Paradigm (HSP; Gillette et al., 1999), we assessed verb learning using multiple tasks. When verb learning was assessed via a free-response task, learning was minimal, replicating the challenge of learning precise verb meanings via observational contexts. The findings from both a categorization and semantic similarity task, however, suggest that learners do acquire partial knowledge of both action and mental verbs via their observational contexts.

Research on infants' observational behavior has predominantly focused on retrospective information gain, leaving the role of prospective evaluation of information gain unclear. We examined 12-month-olds' use of information sources in an eye-tracking study, where participants could use their gaze to 'shake' two out of three boxes to locate a hidden character through auditory cues. Across two pre-registered experiments, we manipulated the probability distributions for character locations to assess forward-looking exploratory strategies. Findings from Experiment 1 with a uniform distribution suggest that while infants learned task contingencies, their choices did not align with maximizing expected information gain, leaning instead towards confirmatory hypothesis testing. Experiment 2 employs a non-uniform probability distribution for character locations to rule out alternative explanations of Experiment 1. In this setup, one box pair provides more information gain, while the other provides confirmatory evidence. Data collection is in progress, results will be presented at the conference.

Artificial Neural Networks can be used to build a general theory of intelligent systems, connecting the computational, algorithmic and implementational levels. I analyze the generalization of learning in simple but challenging problems as a way to build the theory. I report simulations of learning and generalizing sameness, using Simple Recurrent Networks (SRN), Long-Short Term Memories (LSTM) and Transformers. We show that even when minimal requirements to implement sameness in SRNs are met, and a SRN network that can compute sameness theoretically exists, we failed to obtain it by training with backpropagation using all the possible input pairs. LSTMs come close to learn sameness, but the best networks require an inordinate amount of examples and the enrichment of the sample with positive examples. The same happens with Transformers. A similar task applied to ChatGPT revealed related problems. We discuss what this implies for Cognitive and Neural Sciences.

Networks are a useful mathematical tool for capturing the complexity of the world. Using behavioral measures, we showed that human adults were sensitive to the high-level network structure underlying auditory sequences (such as communities) even when presented with incomplete information. Their performance was best explained by a mathematical model following associative learning principles and based on the integration of the transition probabilities between adjacent and non-adjacent elements with memory decay. In a follow up MEG study, we explored the neural correlates of this hypothesis. First, the comparison of the brain responses to tone transitions adhering or not to the community structure revealed an early difference, suggesting an automatic encoding of sequence structure. Second, time-resolved decoding allowed determining the duration and overlap of the representation of each tone. The decoding performance exhibited exponential decay, resulting in a significant overlap between the representations of successive tones, enabling associative learning through Hebbian rule.

Children learn words that name objects (“ball”) and those that name abstract concepts (“story”). One view of learning is that different inputs matter for different words (Snedeker & Gleitman, 2004). That is, many argue that although the observational contexts in which words occur are sufficient for learning object names, they are not for learning abstract “hard words” (Gleitman et al., 2005). This study revisited the contributions of observational contexts to learning one type of hard word: nouns denoting non-basic level object categories (“hard nouns” like “friend”; Kako, 2005). In a new artificial learning paradigm, we reveal that although observational contexts were insufficient for full hard noun learning, they afforded learners partial knowledge that allowed them to succeed in some learning tasks. These data highlight how observational contexts may lay the foundation for learning hard nouns, and underscore how definitions of learning impact our understanding of how the input shapes it.

Second language learners must often learn categories which may not map well with those of their first language. Prepositions often differ between languages, for example, German uses different words for vertically “on” and horizontally “on” which would be novel to an English-speaker. Additionally, learners must contend with varying degrees of noise in the learning environment. A spatial continuum of images was created depicting prepositions such as “above” and “below” (familiar) or horizontal “on” and vertical “on” (novel). We used an artificial preposition learning task in adult English-speakers to explore both the influence of familiarity (familiar or novel) and the degree of statistical regularity in the learning material (noisy or consistent labeling of continuum steps) on learning outcomes. Our results suggest that learners are sensitive to statistics and familiarity and revealed individual differences in the sensitivity to these statistics, suggesting differences in efficiency of learning novel prepositions.

Imagine that a river becomes polluted if two plants generate too much waste. One might be more inclined to say that a plant caused the river to become polluted when it produced more waste than expected. While similar normality effects on causal judgments have been observed in cases with binary variables, little work has focused on cases with continuous variables. To test whether the statistical normality of continuous variables influences causal judgments, we had participants learn statistical norms over repeated iterations of a vignette and make a causal judgment about an instance of that vignette. Following Icard et al. (2017), we manipulated the causal structure and the normality of each cause. By testing whether normality effects on causal judgment generalize to cases with continuous variables, our results help determine whether these effects are central to human cognition, or simply apply to a subset of cases studied thus far.

The swift progression and expansion of machine learning (ML) have not gone unnoticed within the realm of statistical mechanics. In particular, ML techniques have attracted attention by the classical density-functional theory (DFT) community, as they enable automatic discovery of free-energy functionals to determine the equilibrium-density profile of a many-particle system. Within classical DFT, the external potential accounts for the interaction of the many-particle system with an external field, thus, affecting the density distribution. In this context, we introduce a statistical-learning framework to infer the external potential exerted on a classical many-particle system. We combine a Bayesian inference approach with the classical DFT apparatus to reconstruct the external potential, yielding a probabilistic description of the external-potential functional form with inherent uncertainty quantification. Our framework is exemplified with a grand-canonical one-dimensional classical particle ensemble with excluded volume interactions in a confined geometry. The required training dataset is generated using a Monte Carlo (MC) simulation where the external potential is applied to the grand-canonical ensemble. The resulting particle coordinates from the MC simulation are fed into the learning framework to uncover the external potential. This eventually allows us to characterize the equilibrium density profile of the system by using the tools of DFT. Our approach benchmarks the inferred density against the exact one calculated through the DFT formulation with the true external potential. The proposed Bayesian procedure accurately infers the external potential and the density profile. We also highlight the external-potential uncertainty quantification conditioned on the amount of available simulated data. The seemingly simple case study introduced in this work might serve as a prototype for studying a wide variety of applications, including adsorption, wetting, and capillarity, to name a few. [ABSTRACT FROM AUTHOR]

The study addresses the long-term effects of promotions in terms of movement in a value-based segmentation (lead, iron, gold, platinum), instead of simply looking at response rates that occur shortly after the promotion. The study develops a framework for planning an optimal promotion strategy via Markov Decision Processes and Machine Learning methods for an online department store. In the first phase, the states are set as the customer profitability segments in order to conduct the MDPs. Then, MDP model is solved, and the optimal decision for each segment is determined. In the second phase, in order to aid the company for making their plans for the next year, the segment that the customer will belong to next year should be predicted. Prediction of the future customer profitability segment is performed by using several machine learning algorithms, and the best performing model is selected. Using this best performing model, the company can predict the future (potential) profitability segment of the customer and make plans which include the optimal promotions that will be directed to the customers depending on their segments (these optimal promotions are the outcomes of the first phase). The proposed framework can be applied by practitioners in e-commerce companies which keep customer data. [ABSTRACT FROM AUTHOR]

This study addresses the potential for improving the accuracy, scope, and value of sales research through the application of data mining and machine learning algorithms. By examining prior research, identifying opportunities for improvement, and assessing gaps that can benefit from machine learning, research and application are made more accessible for sales researchers and managers. Machine learning can address important sales research questions that cannot be answered with the same accuracy or efficiency as traditional research methods. This study demonstrates the benefits of the methods through an example of application to the prediction of salesforce performance based on behavioral, attitudinal, and demographic data. This includes future research ideas, usage cases, and applications where machine learning could advance sales research and management. Machine learning and predictive analytics methods have multiple applications, including in B2C and B2B market contexts and for companies and independent sales teams. [ABSTRACT FROM AUTHOR]

Continuously reinforced concrete pavement (CRCP), crucial for the resilience of transportation infrastructure owing to its continuous steel reinforcement, confronts a critical challenge in the form of spalling—a distress phenomenon posing a threat to pavement durability and overall structural integrity. The detachment or breakage of concrete from the surface compromises CRCP's functionality and raises safety concerns and escalating maintenance costs. To address this pressing issue, our study investigates the multifaceted factors influencing spalling, employing a comprehensive approach that integrates statistical and machine learning techniques for predictive modeling. Descriptive statistics meticulously profile the dataset, emphasizing age, thickness, precipitation, temperature, and traffic parameters. Regression analysis unveils key relationships, emphasizing the significance of age, annual temperature, annual precipitation, maximum humidity, and the initial International Roughness Index (IRI) as influential factors. The correlation matrix heatmap guides feature selection, elucidating intricate interdependencies. Simultaneously, feature importance analysis identifies age, Average Annual Daily Traffic (AADT), and total pavement thickness as crucial contributors to spalling. In machine learning, adopting models, including Gaussian Process Regression and ensemble tree models, is grounded in their diverse capabilities and suitability for the complex task at hand. Their varying predictive accuracies underscore the importance of judicious model selection. This research advances pavement engineering practices by offering nuanced insights into factors influencing spalling in CRCP, refining our understanding of spalling influences. Consequently, the study not only opens avenues for developing improved predictive methodologies but also enhances the durability of CRCP infrastructure, addressing broader implications for informed decision-making in transportation infrastructure management. The selection of Gaussian Process Regression and ensemble tree models stems from their adaptability to capture intricate relationships within the dataset, and their comparative performance provides valuable insights into the diverse predictive capabilities of these models in the context of CRCP spalling. [ABSTRACT FROM AUTHOR]

Zebrafish have become an essential model organism in screening for developmental neurotoxic chemicals and their molecular targets. The success of zebrafish as a screening model is partially due to their physical characteristics including their relatively simple nervous system, rapid development, experimental tractability, and genetic diversity combined with technical advantages that allow for the generation of large amounts of high-dimensional behavioral data. These data are complex and require advanced machine learning and statistical techniques to comprehensively analyze and capture spatiotemporal responses. To accomplish this goal, we have trained semi-supervised deep autoencoders using behavior data from unexposed larval zebrafish to extract quintessential "normal" behavior. Following training, our network was evaluated using data from larvae shown to have significant changes in behavior (using a traditional statistical framework) following exposure to toxicants that include nanomaterials, aromatics, per- and polyfluoroalkyl substances (PFAS), and other environmental contaminants. Further, our model identified new chemicals (Perfluoro-n-octadecanoic acid, 8-Chloroperfluorooctylphosphonic acid, and Nonafluoropentanamide) as capable of inducing abnormal behavior at multiple chemical-concentrations pairs not captured using distance moved alone. Leveraging this deep learning model will allow for better characterization of the different exposure-induced behavioral phenotypes, facilitate improved genetic and neurobehavioral analysis in mechanistic determination studies and provide a robust framework for analyzing complex behaviors found in higher-order model systems. Author summary: We demonstrate that a deep autoencoder using raw behavioral tracking data from zebrafish toxicity screens outperforms conventional statistical methods, resulting in a comprehensive evaluation of behavioral data. Our models can accurately distinguish between normal and abnormal behavior with near-complete overlap with existing statistical approaches, with many chemicals detectable at lower concentrations than with conventional statistical tests; this is a crucial finding for the protection of public health as exposure can lead to a range of neurodevelopmental disorders, including cognitive and other behavioral deficits. Our deep learning models enable the identification of new substances capable of inducing aberrant behavior, and we generated new data to demonstrate the reproducibility of these results. Thus, neurodevelopmentally active chemicals identified by our deep autoencoder models may represent previously undetectable signals of subtle individual response differences. Our method elegantly accounts for the high degree of behavioral variability associated with the genetic diversity found in a highly outbred population, as is typical for zebrafish research, thereby making it applicable to multiple laboratories generating similar data. Utilizing the vast quantities of control data generated during high-throughput screening is one of the most innovative aspects of this study and to our knowledge is the first study to explicitly develop a deep autoencoder model for anomaly detection in large-scale toxicological behavior studies. [ABSTRACT FROM AUTHOR]

The agricultural industry is critical to the Indian GDP, yet it faces several issues, such as energy, unemployment, and water scarcity. Among these challenges, reliable access to electricity stands out as a crucial factor for implementing smart farming techniques tailored to diverse crops and irrigation patterns. This paper addresses the intricate task of load forecasting in the agriculture sector, employing machine learning and ensemble learning methods. By scrutinizing various agroclimatic and weather characteristics, we pinpoint exogenous features influencing load prediction and prioritize their significance. The study introduces a groundbreaking approach, Temporal Adaptive Ensemble Learning (TAEL), designed to augment accuracy by incorporating temporal dynamics. Six distinct cases, varying in input parameters, are meticulously examined, employing ensemble approaches and benchmarking against Multiple Linear Regression (MLR) and K-Nearest Neighbor (KNN) models. The findings emphasize the vital role of integrating agroclimatic and time series features, showcasing notable enhancements in algorithmic performance across all cases. The proposed TAEL model stands out, achieving a remarkable mean accuracy of 87.1% and a 20.4% MAPE error. This underscores its effectiveness compared to traditional statistical and machine learning approaches, addressing limitations observed in existing studies. The research not only identifies critical factors influencing load prediction but also pioneers a novel approach, TAEL, contributing to the advancement of agricultural load forecasting methodologies. [ABSTRACT FROM AUTHOR]

Marine plankton communities form intricate networks of interacting organisms at the base of the food chain, and play a central role in regulating ocean biogeochemical cycles and climate. However, predicting plankton community shifts in response to climate change remains challenging. While species distribution models are valuable tools for predicting changes in species biogeography under climate change scenarios, they generally overlook the key role of biotic interactions, which can significantly shape ecological processes and ecosystem responses. Here, we introduce a novel statistical framework, association distribution modelling (ADM), designed to model and predict ecological associations distribution in space and time. Applied on a Tara Oceans genome-resolved metagenomics dataset, the present-day biogeography of ADM-inferred marine plankton associations revealed four major biogeographic biomes organized along a latitudinal gradient. We predicted the evolution of these biome-specific communities in response to a climate change scenario, highlighting differential responses to environmental change. Finally, we explored the functional potential of impacted plankton communities, focusing on carbon fixation, outlining the predicted evolution of its geographical distribution and implications for ecosystem function. This article is part of the theme issue 'Connected interactions: enriching food web research by spatial and social interactions'. [ABSTRACT FROM AUTHOR]

With the exponential progress in the field of cheminformatics, the conventional modeling approaches have so far been to employ supervised and unsupervised machine learning (ML) and deep learning models, utilizing the standard molecular descriptors, which represent the structural, physicochemical, and electronic properties of a particular compound. Deviating from the conventional approach, in this investigation, we have employed the classification Read-Across Structure-Activity Relationship (c-RASAR), which involves the amalgamation of the concepts of classification-based quantitative structure-activity relationship (QSAR) and Read-Across to incorporate Read-Across-derived similarity and error-based descriptors into a statistical and machine learning modeling framework. ML models developed from these RASAR descriptors use similarity-based information from the close source neighbors of a particular query compound. We have employed different classification modeling algorithms on the selected QSAR and RASAR descriptors to develop predictive models for efficient prediction of query compounds' hepatotoxicity. The predictivity of each of these models was evaluated on a large number of test set compounds. The best-performing model was also used to screen a true external data set. The concepts of explainable AI (XAI) coupled with Read-Across were used to interpret the contributions of the RASAR descriptors in the best c-RASAR model and to explain the chemical diversity in the dataset. The application of various unsupervised dimensionality reduction techniques like t-SNE and UMAP and the supervised ARKA framework showed the usefulness of the RASAR descriptors over the selected QSAR descriptors in their ability to group similar compounds, enhancing the modelability of the dataset and efficiently identifying activity cliffs. Furthermore, the activity cliffs were also identified from Read-Across by observing the nature of compounds constituting the nearest neighbors for a particular query compound. On comparing our simple linear c-RASAR model with the previously reported models developed using the same dataset derived from the US FDA Orange Book (https://www.accessdata.fda.gov/scripts/cder/ob/index.cfm), it was observed that our model is simple, reproducible, transferable, and highly predictive. The performance of the LDA c-RASAR model on the true external set supersedes that of the previously reported work. Therefore, the present simple LDA c-RASAR model can efficiently be used to predict the hepatotoxicity of query chemicals. [ABSTRACT FROM AUTHOR]

The complex jagged trajectory of fractured surfaces of two pieces of forensic evidence is used to recognize a "match" by using comparative microscopy and tactile pattern analysis. The material intrinsic properties and microstructures, as well as the exposure history of external forces on a fragment of forensic evidence have the premise of uniqueness at a relevant microscopic length scale (about 2–3 grains for cleavage fracture), wherein the statistics of the fracture surface become non-self-affine. We utilize these unique features to quantitatively describe the microscopic aspects of fracture surfaces for forensic comparisons, employing spectral analysis of the topography mapped by three-dimensional microscopy. Multivariate statistical learning tools are used to classify articles and result in near-perfect identification of a "match" and "non-match" among candidate forensic specimens. The framework has the potential for forensic application across a broad range of fractured materials and toolmarks, of diverse texture and mechanical properties. The complex jagged trajectory of fractured surfaces of two pieces of forensic evidence is used to recognize a match by using comparative microscopy. Here the authors leveraged statistical analysis of the surface topography to provide a quantitative description of the fracture surface, enabling forensic article comparisons with quantified probabilities. [ABSTRACT FROM AUTHOR]

Research Highlights Previous research has shown that when domain‐general transitional probability (TP) cues to word segmentation are in conflict with language‐specific stress cues, English‐learning 5‐ and 7‐month‐olds rely on TP, whereas 9‐month‐olds rely on stress. In two artificial languages, we evaluated English‐learning infants’ sensitivity to TP cues to word segmentation vis‐a‐vis language‐specific vowel phonotactic (VP) cues—English words do not end in lax vowels. These cues were either consistent or conflicting. When these cues were in conflict, 10‐month‐olds relied on the VP cues, whereas 5‐month‐olds relied on TP. These findings align with statistical bootstrapping accounts, where infants initially use domain‐general distributional information for word segmentation, and subsequently discover language‐specific patterns based on segmented words. Research indicates that when transitional probability (TP) conflicts with stress cues for word segmentation, English‐learning 9‐month‐olds rely on stress, whereas younger infants rely on TP. In two artificial languages, we evaluated English‐learning infants’ sensitivity to TP versus vowel phonotactic (VP) cues for word segmentation. When these cues conflicted, 10‐month‐olds relied on VPs, whereas 5‐month‐olds relied on TP. These findings align with statistical bootstrapping accounts, where infants first utilize domain‐general distributional information for word segmentation, and then identify language‐specific patterns from segmented words. [ABSTRACT FROM AUTHOR]

Joint species distribution modelling (JSDM) is a widely used statistical method that analyzes combined patterns of all species in a community, linking empirical data to ecological theory and enhancing community-wide prediction tasks. However, fitting JSDMs to large datasets is often computationally demanding and time-consuming. Recent studies have introduced new statistical and machine learning techniques to provide more scalable fitting algorithms, but extending these to complex JSDM structures that account for spatial dependencies or multi-level sampling designs remains challenging. In this study, we aim to enhance JSDM scalability by leveraging high-performance computing (HPC) resources for an existing fitting method. Our work focuses on the Hmsc R-package, a widely used JSDM framework that supports the integration of various dataset types into a single comprehensive model. We developed a GPU-compatible implementation of its model-fitting algorithm using Python and the TensorFlow library. Despite these changes, our enhanced framework retains the original user interface of the Hmsc R-package. We evaluated the performance of the proposed implementation across various model configurations and dataset sizes. Our results show a significant increase in model fitting speed for most models compared to the baseline Hmsc R-package. For the largest datasets, we achieved speed-ups of over 1000 times, demonstrating the substantial potential of GPU porting for previously CPU-bound JSDM software. This advancement opens promising opportunities for better utilizing the rapidly accumulating new biodiversity data resources for inference and prediction. [ABSTRACT FROM AUTHOR]

Building Performance Simulation extensively uses statistical learning techniques for quicker insights and improved accessibility. These techniques help understand the relationship between input variables and the desired outputs, and they can predict unknown observations. Prediction becomes more informative with uncertainty quantification, which involves computing prediction intervals. Conformal prediction has emerged over the past 25 years as a flexible and rigorous method for estimating uncertainty. This approach can be applied to any pre-trained model, creating statistically rigorous uncertainty sets or intervals for model predictions. This study uses data from simulated buildings to demonstrate the powerful applications of conformal prediction in Building Performance Simulation (BPS) and, consequently, to the broader energy sector. Results show that conformal prediction can be applied when any assumptions about input and output variables are made, enhancing understanding and facilitating informed decision-making in energy system design and operation. [ABSTRACT FROM AUTHOR]

Domain adaptation is a subfield of statistical learning theory that takes into account the shift between the distribution of training and test data, typically known as source and target domains, respectively. In this context, this paper presents an incremental approach to tackle the intricate challenge of unsupervised domain adaptation, where labeled data within the target domain is unavailable. The proposed approach, OTP-DA, endeavors to learn a sequence of joint subspaces from both the source and target domains using Linear Discriminant Analysis (LDA), such that the projected data into these subspaces are domain-invariant and well-separated. Nonetheless, the necessity of labeled data for LDA to derive the projection matrix presents a substantial impediment, given the absence of labels within the target domain in the setting of unsupervised domain adaptation. To circumvent this limitation, we introduce a selective label propagation technique grounded on optimal transport (OTP), to generate pseudo-labels for target data, which serve as surrogates for the unknown labels. We anticipate that the process of inferring labels for target data will be substantially streamlined within the acquired latent subspaces, thereby facilitating a self-training mechanism. Furthermore, our paper provides a rigorous theoretical analysis of OTP-DA, underpinned by the concept of weak domain adaptation learners, thereby elucidating the requisite conditions for the proposed approach to solve the problem of unsupervised domain adaptation efficiently. Experimentation across a spectrum of visual domain adaptation problems suggests that OTP-DA exhibits promising efficacy and robustness, positioning it favorably compared to several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

As precision agriculture (PA) advances, the demand for accurate and high-resolution weather forecasts becomes critical for optimizing agricultural management practices. Despite improvements in Numerical Weather Prediction (NWP) models, they lack the granularity and efficiency needed for PA. Data-driven models offer a promising alternative by integrating predictive capabilities closer to IoT edge data sources, but their efficacy requires evaluation. Here, this paper evaluates six models from three data-driven eras (statistical, machine learning, and deep learning) using agrometeorological data from an Automatic Weather Station (AWS) in Sidi Rahal, East Marrakech, central Morocco, covering 2013–2020 at half-hour intervals, including air temperature, solar radiation, and relative humidity. First, the data is quality-controlled through imputation using ERA5-Land. Then, the dataset was split into training (2013–2019) and evaluation (2020) sets, with validation horizons of 1 day, 3 days, and 1 week. Statistical models generally perform well in air temperature forecasting, occasionally surpassing other models. However, the Temporal Convolutional Neural Network (TCNN) consistently demonstrates superior performance for challenging variables, balancing low RMSE and high R2 across various horizons, with some exceptions. Specifically, for relative humidity, the linear regression model achieves slightly lower RMSE (3,96% and 6,05%) compared to TCNN (4,00% and 6,79%) for 1 day and 3 days, respectively. Additionally, CatBoost outperforms TCNN for 1-week forecasts. In terms of training time, the Transformer requires the longest, followed by AutoARIMA and CatBoost. Uncertainty analysis of stochastic models using solar radiation showed the stable performance of TCNN with 0,80 and 0,01 for the RMSE and R2 standard deviations, respectively. Considering the trade-off between performance, training time, and capturing complex relationships, TCNN emerges as the optimal choice. ANOVA, Tukey's HSD and Mann-Whitney U statistical tests also confirmed TCNN's performance. Finally, a comparison with the Global Forecast System (GFS) reveals TCNN's clear superiority in all metrics, particularly evident for the RMSE of 3 days air temperature forecasts (TCNN: 1,96 °C, GFS: 3,59 °C). [ABSTRACT FROM AUTHOR]

In practice, and especially when training deep neural networks, visual recognition rules are often learned based on various sources of information. On the other hand, the recent deployment of facial recognition systems with uneven performances on different population segments has highlighted the representativeness issues induced by a naive aggregation of the datasets. In this paper, we show how biasing models can remedy these problems. Based on the (approximate) knowledge of the biasing mechanisms at work, our approach consists in reweighting the observations, so as to form a nearly debiased estimator of the target distribution. One key condition is that the supports of the biased distributions must partly overlap, and cover the support of the target distribution. In order to meet this requirement in practice, we propose to use a low dimensional image representation, shared across the image databases. Finally, we provide numerical experiments highlighting the relevance of our approach. [ABSTRACT FROM AUTHOR]