Self-supervised Learning (SSL) has become a powerful paradigm for representation learning without manual annotations. However, most existing frameworks focus on global alignment and struggle to capture the hierarchical, multi-scale lesion patterns characteristic of plant disease imagery. To address this gap, we propose PSMamba, a progressive self-supervised framework that integrates the efficient sequence modelling of Vision Mamba (VM) with a dual-student hierarchical distillation strategy. Unlike conventional single teacher-student designs, PSMamba employs a shared global teacher and two specialised students: one processes mid-scale views to capture lesion distributions and vein structures, while the other focuses on local views to capture fine-grained cues such as texture irregularities and early-stage lesions. This multi-granular supervision facilitates the joint learning of contextual and detailed representations, with consistency losses ensuring coherent cross-scale alignment. Experiments on three benchmark datasets show that PSMamba consistently outperforms representative CNN-, Transformer-, SSL-, and Mamba-based baselines, delivering superior accuracy and robustness in both domain-shifted and fine-grained scenarios.
Aggregators of consumer energy resources (CERs) like rooftop solar and battery energy storage (BES) face challenges due to their inherent uncertainties. A sensible approach is to use stochastic optimization to handle such uncertainties, which can lead to infeasible problems or loss in revenues if not chosen appropriately. This paper presents three stochastic optimization methods: risk-neutral, robust, and chance-constrained, to address the impact of CER uncertainties for aggregators who participate in energy and regulation services markets in the Australian National Electricity Market. Furthermore, these methods utilize the flexibility of BES, considering precise state-of-charge dynamics and complementarity constraints, aiming for scalable performance while managing uncertainty. The problems are formed as two-stage stochastic mixed-integer linear programs, with relaxations adopted for large scenario sets. The solution approach employs scenario-based methodologies and affine recourse policies to obtain tractable reformulations. These methods are evaluated in terms of profit and constraint violation risk across use cases reflecting diverse operational and market settings, uncertainty characteristics, and decision-making preferences, offering aggregators insight into the selection of appropriate methods. Numerical results indicate that, while stochastic methods outperform traditional deterministic methods in terms of profit and risk, the risk-neutral method performs best when uncertainty is correctly captured, whereas robust and chance-constrained methods are more effective when uncertainty is misspecified.
Livestock systems represent a considerable environmental challenge. In response, various scientists, non-governmental organisations, and policy makers claim that Western populations in particular need to sharply reduce meat consumption. Given people’s attachment to meat, many of these actors favour hard policy interventions based on a range of systemic financial and legal reforms that would go beyond mere nudging and the formulation of recommendations, including the top-down imposition of meat taxes and bans, as well as herd size reductions, which would lead to sharply higher prices. However, arguments in support of such policies tend to oversimplify the issue, ignoring regional variations, mitigation potential, and broader ecological and nutritional contexts. The focus of this article is on dietary greenhouse gas (GHG) emissions as a main target for environmental policymaking, with all livestock production in the West contributing 2.6
We provide an updated 'multiwavelength' version of the G4Jy catalogue that has 127 new host-galaxy identifications, as described in Paper III of this paper series. We also supplement the redshift information (0.0 < z < 3.6), gathered in Paper III, with griz photometry available through DR10 of the DESI Legacy Surveys. Together, this legacy dataset allows us to investigate the multiwavelength properties of these southern radio-bright galaxies, which includes an initial analysis of radio spectral-curvature for this complete sample (S-151 MHz > 4 Jy). For example, we present (for the first time in the literature) the radio-power-size diagram as a function of radio spectral-curvature, [P-D](SCI), noting that the spectral-curvature index (SCI) can act as a proxy for the spectral age of the radio source. This radio-power-size-age diagram shows a predominance of radio galaxies with SCI > 0.15 and D < 200 kpc, which are candidates for both remnant radio-galaxies and young radio-sources, and a vast range of linear sizes for candidate restarted radio-galaxies (having SCI < -0.15). We also show that (i) G4Jy sources populate the entirety of WISE colour-colour space, (ii) optically point-like sources (i.e. candidate quasars) are brighter than the well-studied K-z relation (as expected), and (iii) there is no relation between the SCI of the radio source and its host-galaxy properties.
In this paper, we introduce Rank-R1, a novel LLM-based reranker that performs reasoning over both the user query and candidate documents before performing the ranking task. Existing document reranking methods based on large language models (LLMs) typically rely on prompting or fine-tuning LLMs to order or label candidate documents according to their relevance to a query. For Rank-R1, we use a reinforcement learning algorithm along with only a small set of relevance labels (without any reasoning supervision) to enhance the reasoning ability of LLM-based rerankers. Our hypothesis is that adding reasoning capabilities to the rerankers can improve their relevance assessement and ranking capabilities. Our experiments on the TREC DL and BRIGHT datasets show that Rank-R1 is highly effective, especially for complex queries. In particular, we find that Rank-R1 achieves effectiveness on in-domain datasets at par with that of supervised fine-tuning methods, but utilizing only 18% of the training data used by the fine-tuning methods. We also find that the model largely outperforms zero-shot and supervised fine-tuning when applied to out-of-domain datasets featuring complex queries, especially when a 14B-size model is used. Finally, we qualitatively observe that Rank-R1's reasoning process improves the explainability of the ranking results, opening new opportunities for search engine results presentation and fruition.
