Tillman Weyde

Background: Compositional symbol-forming and symbol-relating behaviors in deep learning or neuro-symbolic systems have been repeatedly recommended as part of a solution to the shortcomings of current state-of-the-art artificial intelligence. Studying how compositional languages can emerge between tabula rasa deep learning agents may help us understand how to make artificial neural networks represent unstructured, continuous input data in terms of combinations of discrete symbols. Objectives: We aim present a comprehensive overview of recent research into compositional languages emerging between deep learning agents, in a manner that will be accessible to machine learning researchers who are not already aware of emergent communication and emergent languages. Methods: We review roughly ten years of emergent language research, particularly focusing on contributions after 2019 that pertain to measuring or eliciting compositionality in emergent languages. Results: Systematic generalization and topographic similarity (topsim) are the most dominant measures of compositionality in recent literature. “Productivity pressure”, forcing agents to use vocabularies smaller than the number of meanings they need to communicate, is clearly necessary for compositionality to emerge. Regularizing or periodically resetting receiver/listener agents is an effective way of encouraging more compositional languages, perhaps because it creates a pressure for the speaker to create languages that can be learned more efficiently. The relative benefits of various neural network architectures, particularly the Transformer architecture dominant in other areas of deep learning, remains an underexplored topic. As other authors have noted, the field relies heavily on small-scale models and simple, often symbolic environments, which may hinder the generality of current conclusions. Conclusions: Emergent language research provides a testbed for encouraging emergent compositionality in deep learning models, which may in future contribute to the development of safer, more interpretable, and more sample-efficient neuro-symbolic foundational models. We advocate that future research converges on topsim and generalization as the standard approaches to measuring compositionality, but also works to expand topsim into a family of metrics that can detect compositionality in languages displaying forms of linguistic variation such as free word order and synonymy. We also call upon researchers to test promising techniques at larger scales, with a greater range of agent architectures, and with more complex, multi-modal referents.

The aim of this study is to evaluate a machine-learning method in which symbolic representations of folk songs are segmented and classified into tune families with Haar-wavelet filtering. The method is compared with previously proposed Gestalt-based method. Melodies are represented as discrete symbolic pitch-time signals. We apply the continuous wavelet transform (CWT) with the Haar wavelet at specific scales, obtaining filtered versions of melodies emphasizing their information at particular time-scales. We use the filtered signal for representation and segmentation, using the wavelet coefficients' local maxima to indicate local boundaries and classify segments by means of k-nearest neighbours based on standard vector-metrics (Euclidean, cityblock), and compare the results to a Gestalt-based segmentation method and metrics applied directly to the pitch signal. We found that the wavelet based segmentation and wavelet-filtering of the pitch signal lead to better classification accuracy in cross-validated evaluation when the time-scale and other parameters are optimized.

Background: Consuming high amounts of foods or beverages with high levels of saturated fats, salt, or sugar (HFSS) can be harmful for health. Many snacks fall into this category (HFSS snacks). However, the palatability of these snacks means that people can sometimes struggle to reduce their intake. Machine learning algorithms could help in predicting the likely occurrence of HFSS snacking so that just-in-time adaptive interventions can be deployed. However, HFSS snacking data have certain characteristics, such as sparseness and incompleteness, which make snacking prediction a challenge for machine learning approaches. Previous attempts have employed several potential predictor variables and have achieved considerable success. Nevertheless, collecting information from several dimensions requires several potentially burdensome user questionnaires, and thus, this approach may be less acceptable for the general public. Objective: Our aim was to consider the capacity of standard (unmodified in any way; to tailor to the specific learning problem) machine learning algorithms to predict HFSS snacking based on the following minimal data that can be collected in a mostly automated way: day of the week, time of the day (divided into time bins), and location (divided into work, home, and other). Methods: A total of 111 participants in the United Kingdom were asked to record HFSS snacking occurrences and the location category over a period of 28 days, and this was considered the UK dataset. Data collection was facilitated by a purpose-specific app (Snack Tracker). Additionally, a similar dataset from the Netherlands was used (Dutch dataset). Both datasets were analyzed using machine learning methods, including random forest regressor, Extreme Gradient Boosting regressor, feed forward neural network, and long short-term memory. We additionally employed 2 baseline statistical models for prediction. In all cases, the prediction problem was the time to the next HFSS snack from the current one, and the evaluation metric was the mean absolute error. Results: The ability of machine learning methods to predict the time of the next HFSS snack was assessed. The quality of the prediction depended on the dataset, temporal resolution, and machine learning algorithm employed. In some cases, predictions were accurate to as low as 17 minutes on average. In general, machine learning methods outperformed the baseline models, but no machine learning method was clearly better than the others. Feed forward neural network showed a very marginal advantage. Conclusions: The prediction of HFSS snacking using sparse data is possible with reasonable accuracy. Our findings offer a foundation for further exploring how machine learning methods can be used in health psychology and provide directions for further research.

Knowledge graphs (KGs) feature ever more frequently as symbolic components in neurosymbolic research and systems. But even though a central concern of neurosymbolic artificial intelligence is to combine neural learning with symbolic reasoning, relatively little neurosymbolic research focuses on leveraging the logical representation and reasoning capabilities of Web Ontology Language (OWL)-based KGs. The objective of this position article is to inspire more neurosymbolic researchers to embrace the OWL and the Semantic Web by raising awareness of the benefits, capabilities, and applications of OWL-based KGs, particularly with respect to logical reasoning. We describe the ecosystem of open W3C standards-based resources available that support the adoption and use of OWL-based KGs; we describe tools that exist for engineering custom OWL ontologies tailored to particular research needs; we discuss the encoding of background KG knowledge in subsymbolic embedding spaces and various applications of this approach; we discuss and illustrate the reasoning capabilities of OWL-based KGs; and we describe several promising directions for research that focus on leveraging these reasoning capabilities. We also discuss the specialised resources needed to undertake research on OWL-based KGs in neurosymbolic systems. We use the example of NeSy4VRD, an image dataset with a custom-designed companion OWL ontology. The scarcity of this kind of resource should be addressed to accelerate research in this field.

Financial correlation matrices play a vital role in various quantitative finance applications, but generating synthetic correlation matrices that accurately reflect market structures and stylized facts remains challenging. We introduce a novel application of denoising diffusion probabilistic models (DDPMs) for this task, proposing both unconditional (DM) and conditional (CDM) models. Our experimental evaluation demonstrates the superior performance of our models in generating correlation matrices that closely resemble empirical data while capturing differences across market regimes. We also present a case study highlighting the utility of our approach in assessing asset allocation frameworks and enhancing risk modeling by augmenting empirical datasets with synthetic data. Our findings showcase DDPMs’ potential in mitigating limitations of scarce financial data, enabling robust quantitative modeling and analysis.

This paper introduces a novel web platform designed for exploration, analysis, and collaboration in the jazz music domain called JazzDAP. Our platform integrates advanced music information retrieval techniques with user-friendly interfaces, tailored for musicologists, archivists, and jazz enthusiasts. The platform employs a contour based algorithm for pattern recognition, enabling users to search for specific musical motifs, with filters based on metadata, e.g. artist, location and year of recording. Users can listen to audio sections or MIDI excerpts from the matches and delve into detailed metadata, including the years of recordings, prevalence of specific patterns, and information about the artists associated with them. Visualizations aid in uncovering trends, evolution, and connections in the development of jazz. A key innovation of our platform is the introduction of workflow objects, allowing users to save elements of interest accompanied by notes, named workflows, and engage in collaborative discussions. Users can use workflows to annotate, share insights, and communicate with each other, fostering a community-driven exploration of jazz music. This collaborative aspect enhances the platform’s utility for researchers and enthusiasts alike, aiming to create a dynamic environment for the exchange of knowledge and discoveries. This paper outlines the platform’s structure, highlights its key features, and presents preliminary user feedback. We believe that our work opens new avenues for the exploration and understanding of jazz music, offering a valuable resource for researchers, archivists, and enthusiasts passionate about the intricate patterns that shape the genre.