Abstract

Abstract
The recent developments of modern probabilistic programming languages have enabled the combination of pattern recognition engines implemented by neural networks to guide inference over explanatory factors written as symbols in probabilistic programs. We argue that learning to invert fixed generative programs, instead of learned ones, places stronger restrictions on the representations learned by feature extraction networks, which reduces the space of latent hypotheses and enhances training efficiency. To empirically demonstrate this, we investigate a neurosymbolic object-centric representation learning approach that combines a slot-based neural module optimized via inference compilation to invert a prior generative program of scene generation. By amortizing the search over posterior hypotheses, we demonstrate that approximate inference using data-driven sequential Monte Carlo methods achieves competitive results when compared to state-of-the-art fully neural baselines while requiring several times fewer training steps. © 2026, Transactions on Machine Learning Research. All rights reserved.

Abstract

Abstract
Evaluating diffusion properties of novel optical clearing (OC) agents is critical for advancing medical imaging. Tartrazine (TTZ), a strong absorbing dye, has shown promise in enhancing tissue transparency, yet its diffusion properties remain uncharacterized. In this work, OC treatments with TTZ-water solutions with varying osmolarities were performed, and the diffusion times (tau) that characterize the tissue dehydration and the RI matching mechanisms were estimated. From kinetic T-c measurements during treatment, tau values of water and TTZ were estimated in muscles as 60.0 s and 416.0 s, respectively. Corresponding diffusion coefficients (D) were derived from sample thickness data measured during treatments where the unique fluxes of TTZ and water occur. The respective D values were then calculated as 1.9 x 10(-6) cm(2)/s for water and 3.6 x 10(-7) cm(2)/s for TTZ. These findings provide key insights into TTZ diffusion in skeletal muscle and support its potential as an effective OC agent.

Abstract
Deep-learning beat-tracking algorithms have achieved remarkable accuracy in recent years. However, despite these advancements, challenges persist with musical examples featuring complex rhythmic structures, especially given their under-representation in training corpora. Expanding on our prior work, this paper demonstrates how our user-centred beat-tracking methodology effectively handles increasingly demanding musical scenarios. We evaluate its adaptability and robustness through musical pieces that exhibit rhythmic dissonance, while maintaining ease of integration with leading methods through minimal user annotations. The selected musical works-Uruguayan Candombe, Colombian Bambuco, and Steve Reich's Piano Phase-present escalating levels of rhythmic complexity through their respective polyrhythm, polymetre, and polytempo characteristics. These examples not only validate our method's effectiveness but also demonstrate its capability across increasingly challenging scenarios, culminating in the novel application of beat tracking to polytempo contexts. The results show notable improvements in terms of the F-measure, ranging from 2 to 5 times the state-of-the-art performance. The beat annotations used in fine-tuning reduce the correction edit operations from 1.4 to 2.8 times, while reducing the global annotation effort to between 16% and 37% of the baseline approach. Our experiments demonstrate the broad applicability of our human-in-theloop strategy in the domain of Computational Ethnomusicology, confronting the prevalent Music Information Retrieval (MIR) constraints found in non-Western musical scenarios. Beyond beat tracking and computational rhythm analysis, this user-driven adaptation framework suggests wider implications for various MIR technologies, particularly in scenarios where musical signal ambiguity and human subjectivity challenge conventional algorithms.

Abstract