Labelled data are limited and self-supervised learning is one of the most important approaches for reducing labelling requirements. While it has been extensively explored in the image domain, it has so far not received the same amount of attention in the acoustic domain. Yet, reducing labelling is a key requirement for many acoustic applications. Specifically in bioacoustic, there are rarely sufficient labels for fully supervised learning available. This has led to the widespread use of acoustic recognisers that have been pre-trained on unrelated data for bioacoustic tasks. We posit that training on the actual task data and combining self-supervised pre-training with few-shot classification is a superior approach that has the ability to deliver high accuracy even when only a few labels are available. To this end, we introduce and evaluate a new architecture that combines CNN-based preprocessing with feature extraction based on state space models (SSMs). This combination is motivated by the fact that CNN-based networks alone struggle to capture temporal information effectively, which is crucial for classifying acoustic signals. SSMs, specifically S4 and Mamba, on the other hand, have been shown to have an excellent ability to capture long-range dependencies in sequence data. We pre-train this architecture using contrastive learning on the actual task data and subsequent fine-tuning with an extremely small amount of labelled data. We evaluate the performance of this proposed architecture for (n-shot, n-way) classification on standard benchmarks as well as real-world data. Our evaluation shows that it outperforms state-of-the-art architectures on the few-shot classification problem.

Age-polyethism-the age-based allocation of tasks in social insect colonies-is a key feature of division of labour. While its hormonal underpinnings have been studied extensively, the behavioural and environmental mechanisms driving age-polyethism remain poorly understood, especially under ecological stress. We present a novel modelling framework that integrates social learning with task-related environmental feedback to explain the emergence and breakdown of age-polyethism. We develop two models: a Social Learning (SL) model, in which individuals adapt task preferences by copying similar peers, and a Stimulus-Response Threshold Social Learning (SRT-SL) model, which extends this framework by incorporating task-related dynamic stimuli and response thresholds that regulate collective task demand. Our models demonstrate that age-polyethism can emerge from simple social imitation processes, without the need for fixed hormonal schedules. We show that under increasing environmental pressure (e.g., resource scarcity), age-polyethism collapses as younger individuals are forced into tasks typically handled by older workers. Importantly, we find that age-polyethism does not necessarily optimize immediate colony efficiency; instead, it appears to reflect a trade-off between environmental constraints and behavioural coordination. These findings provide a mechanistic and ecologically grounded explanation for empirical observations linking environmental stress to dysfunctional division of labour and colony collapse.

The scarcity of labelled data makes training Deep Neural Network (DNN) models in bioacoustic applications challenging. In typical bioacoustics applications, manually labelling the required amount of data can be prohibitively expensive. To effectively identify both new and current classes, DNN models must continue to learn new features from a modest amount of fresh data. Active Learning (AL) is an approach that can help with this learning while requiring little labelling effort. Nevertheless, the use of fixed feature extraction approaches limits feature quality, resulting in underutilization of the benefits of AL. We describe an AL framework that addresses this issue by incorporating feature extraction into the AL loop and refining the feature extractor after each round of manual annotation. In addition, we use raw audio processing rather than spectrograms, which is a novel approach. Experiments reveal that the proposed AL framework requires 14.3%, 66.7%, and 47.4% less labelling effort on benchmark audio datasets ESC-50, UrbanSound8k, and InsectWingBeat, respectively, for a large DNN model and similar savings on a microcontroller-based counterpart. Furthermore, we showcase the practical relevance of our study by incorporating data from conservation biology projects. All codes are publicly available on GitHub.

Tracking individuals is a vital part of many experiments conducted to understand collective behaviour. Ants are the paradigmatic model system for such experiments but their lack of individually distinguishing visual features and their high colony densities make it extremely difficult to perform reliable racking automatically. Additionally, the wide diversity of their species' appearances makes a generalized approach even harder. In this paper, we propose a data-driven multi-object tracker that, for the first time, employs domain adaptation to achieve the required generalisation. This approach is built upon a joint-detection-and-tracking framework that is extended by a set of domain discriminator modules integrating an adversarial training strategy in addition to the tracking loss. In addition to this novel domain-adaptive tracking framework, we present a new dataset and a benchmark for the ant tracking problem. The dataset contains 57 video sequences with full trajectory annotation, including 30k frames captured from two different ant species moving on different background patterns. It comprises 33 and 24 sequences for source and target domains, respectively. We compare our proposed framework against other domain-adaptive and non-domain-adaptive multi-object tracking baselines using this dataset and show that incorporating domain adaptation at multiple levels of the tracking pipeline yields significant improvements. The code and the dataset are available at https://github.com/chamathabeysinghe/da-tracker.

The adoption of innovative online teaching tools in Computer Science (CS) courses provides opportunities for data-informed instruction as a regular teaching practice in CS classrooms. In this paper, we present a design study for an interactive visual analytics dashboard, called UserFlow, that supports feedback collection from teaching documents and assists instructors in interpreting feedback and acting on it in a timely manner. The design study is conducted with eight domain experts comprising of four teaching instructors, two learning analytics (LA) experts and two instructional designers. UserFlow offers a set of novel visualization designs for presenting the four interleaving aspects of document engagement (i.e., annotations, document traversal path, reading/focus time and student information). We evaluated UserFlow in an undergraduate computer science course with over 700 students. Our results demonstrate the usefulness and need for such a tool for CS educators to inform teaching approaches and courseware improvement.

Learning has always been considered a social and collaborative activity. From the social constructivism perspective, students learn through the process of sharing experiences and build knowledge and understanding through discussion. This pilot study demonstrates the successful integration of an online social annotation tool in a Python programming workshop that allows students to discuss learning material and code segments with their fellow classmates. We investigate how the students leveraged the tool to improve their learning and to log their sentiments while reading the material and attempting the programming activities. We find that those students who had access to the annotation tool spent an above average amount of time on the material. They also had a higher completion rate and performance compared to the control group who did not have the annotation feature available to them. Feedback from the students who used the tool was very positive. Finally, we delve deeper into analysing the annotation patterns and explore how instructors can make use of fine-grained annotation data to inform and adapt the design of CS teaching and learning materials in online or blended contexts.

A new computer science curriculum has been developed for the Victorian Certificate of Education. It gives students direct entry into second year University computer science. The curriculum focuses on data structures and algorithms, with an emphasis on the graph abstract data type and graph algorithms. We taught a pilot course during 2014 involving students from seven schools, and administered an algorithmic thinking quiz on entry and exit, and also tested a first year university reference group. In this paper we present the curriculum and report on the evaluation. We discuss the effectiveness of our approach and make recommendations for improving future versions of the course. Pedagogical issues are discussed in relation to the cognitive education literature on the teaching of algorithmic thinking.

Hybrid methods for solving combinatorial optimization problems have become increasingly popular recently. The present paper is concerned with hybrids of ant colony optimization and constraint programming which are typically useful for problems with hard constraints. However, the original algorithm suffered from large CPU time requirements. It was shown that such an integration can be made efficient via a further hybridization with beam search resulting in CP-Beam-ACO. The original work suggested this in the context of job scheduling. We show here that this algorithm type is also effective on another problem class, namely the car sequencing. We consider an optimization version, where we aim to optimize the utilization rates across the sequence. Car sequencing is a notoriously difficult problem, because it is difficult to obtain good bounds via relaxations. We show that stochastic sampling provides superior results to well known lower bounds for this problem when combined with CP-Beam-ACO.

SUMMARY: We present a massively parallel stochastic simulation algorithm (SSA) for reaction-diffusion systems implemented on Graphics Processing Units (GPUs). These are designated chips optimized to process a high number of floating point operations in parallel, rendering them well-suited for a range of scientific high-performance computations. Newer GPU generations provide a high-level programming interface which turns them into General-Purpose Graphics Processing Units (GPGPUs). Our SSA exploits GPGPU architecture to achieve a performance gain of two orders of magnitude over the fastest existing implementations on conventional hardware. AVAILABILITY: The software is freely available at http://www.csse.monash.edu.au/~berndm/inchman/.

This paper investigates the use of hybrid meta-heuristics based on ant colony optimization (ACO) for the strip packing problem. Here, a fixed set of rectangular items of fixed sizes have to be placed on a strip of fixed width and infinite height without overlaps and with the objective to minimize the height used. We analyze a commonly used basic placement heuristic (BLF) by itself and in a number of hybrid combinations with ACO. We compare versions that learn item order only, item rotation only, both independently, and rotations conditionally upon placement order. Our analysis shows that integrating a learning meta-heuristic provides a significant performance advantage over using the basic placement heuristic by itself. The experiments confirm that even just learning a placement order alone can provide significant performance improvements. Interestingly, learning item rotations provides at best a marginal advantage. The best hybrid algorithm presented in this paper significantly outperforms previously reported strip packing meta-heuristics.

The web and digital media requires intelligent, adaptive documents whose appearance and content adapts to the viewing context and which support user interaction. While previous research has focussed on textual and multimedia content, this is also true for diagrammatic conte nt. We have designed and implemented an authoring tool which supports the construction of adaptive diagrams. Adaptive layout behaviour is specified by using constraint-based placement tools as well as by allowing the author to specify more radical layout changes using alternate layout configurations. As well as specifying alternate layouts, the author can specify alternate representations for an object, alternate styles and alternate textual content. The resulting space of different versions of the diagram is the cross product of these different alternatives. At display time the version is constructed dynamically, taking into account the author specified preference order on the alternatives, current viewing environment, and user interaction.

Many industrial and academic optimization problems have successfully been tackled with ant colony optimization (ACO), but the ability of ACO to deal with hard constraints has not yet been investigated widely. Most real-world applications, such as fleet routing, rostering and timetabling, however, are subject to hard constraints. Extensions of ACO with hard constraint handling are thus a worthwhile field of study. Constraint handling in meta-heuristics has mostly been studied in the context of evolutionary algorithms. Many of the techniques proposed are penalty-based. A difficulty with such methods is the correct adjustment of penalty factors. Recently a different approach, termed stochastic ranking (SR), has been suggested which does not use a penalty formulation. Benchmarks indicate that SR is currently among the best performing methods for evolutionary constraint optimization. We discuss how the ideas underlying SR can be used in ACO and evaluate two different ways to achieve this integration: one derived from ACS (ACSsr) and one based on ranked ant system (ASRsr). We compare the performance of these to penalty-based ACO as well as to CPACS, an ACO algorithm hybridized with explicit constraint propagation. Initial benchmark results are encouraging.

The Scalable Vector Graphics format SVG is already substantially improving graphics delivery on the web, but some important issues still remain to be addressed. In particular, SVG does not support client-side adaption of documents to different viewing conditions, such as varying screen sizes, style preferences or different device capabilities. Based on our earlier work we show how SVG can be extended with constraint-based specification of document layout to augment it with adaptive capabilities. The core of our proposal is to include one-way constraints into SVG, which offer more expressiveness than the previously suggested class of linear constraints and at the same time require substantially less computational effort.

Abstract Scalable Vector Graphics (SVG) is a language that describestwo-dimensional vector graphics for storage and distribution

We present a general purpose lossless greyscale image compression method, TMW, that is based on the use of linear predictors and implicit segmentation. We then proceed to extend the presented methods to cover near lossless image compression. In order to achieve competitive compression, the compression process is split into an analysis step and a coding step. In the first step, a set of linear predictors and other parameters suitable for the image is calculated, which is included in the compressed file and subsequently used for the coding step. This adaption allows TMW to perform well over a very wide range of image types. Other significant features of TMW are the use of a one-parameter probability distribution, probability calculations based on unquantized prediction values, blending of multiple probability distributions instead of prediction values, and implicit image segmentation. For lossless image compression, the method has been compared to CALIC on a selection of test images, and typically outperforms it by between 2 and 10 percent. For near lossless image compression, the method has been compared to LOCO (Weinberger et al. 1996). Especially for larger allowed deviations from the original image the proposed method can significantly outperform LOCO. In both cases the improvement in compression is achieved at the cost of considerably higher computational complexity.