Zentrale wissenschaftliche Einrichtungen

Towards Uncertainty-Aware Low-Bit Quantized LLMs for On-Device Inference

2026-03-06T00:00:00Z

Towards Uncertainty-Aware Low-Bit Quantized LLMs for On-Device Inference Sparrenberg, Lorenz; Schneider, Tobias; Deußer, Tobias; Berger, Armin; Sifa, Rafet Quantizing large language models (LLMs) significantly reduces memory usage and computational requirements, enabling efficient on-device inference. However, aggressive quantization can degrade model performance and exacerbate prediction uncertainty. To address this critical issue, we propose a logits-based calibration strategy where the model is restricted to generating a single token from a limited set of predefined decision tokens. By applying a temperature-scaled softmax directly on the logits corresponding to these tokens, we obtain calibrated and interpretable probability distributions, explicitly circumventing stochastic methods such as top-k sampling by directly leveraging deterministic logit values, revealing subtle behavioral shifts caused by quantization. Using Qwen-2.5 models ranging from 7\,B to 72\,B parameters at various quantization levels (2, 4, 6 and 8-bit), we evaluate our method across four recently released benchmarks encompassing regression (README++, CompLex-ZH, GIRAI) and classification (DarkBench) tasks. Thus, minimizing the risk of data leakage into pre-training data. Results indicate moderate quantization (4-bit) as optimal, particularly when combined with minimal few-shot prompting, enabling quantized LLMs to closely match or surpass proprietary models such as GPT-4o and GPT-4.1 in certain tasks. Our open-source toolkit facilitates straightforward deployment of reliable, uncertainty-aware quantized LLMs for privacy-preserving, on-device inference, making them suitable for sensitive settings such as human-subject economic experiments and survey analysis.

Cooperation under Comparison

2026-03-17T00:00:00Z

Cooperation under Comparison Stark, Oded; Kosiorowski, Grzegorz We establish a new approach to the modeling of cooperation, and we formulate a new solution concept for cooperative games. We do this by constructing a game of cooperation between individuals who exhibit distaste for relative deprivation, RD, in the sense that they experience stress when their income is lower than that of their comparators. In such a game, the sharing out of the jointly earned income between these individuals when they cooperate, as prescribed by standard solutions of cooperative games, might not be acceptable to the individuals. The stress from RD may have the upper hand. Measuring stress by RD, we thus model a setting in which two individuals who are concerned with being relatively deprived need to decide whether or not to cooperate. We term this setting an RD cooperative game, and we design a rule, the RD solution, for the distribution of the income yielded in this game. The RD solution prescribes cooperation in spite of cooperation-induced stress and preserves the spirit of standardness (an equal sharing of the gain that accrues from cooperation) for two-player games (a property shared by the main solution concepts for cooperative games).

Women's empowerment and nutrition

2026-02-03T00:00:00Z

Women's empowerment and nutrition Maina, Cecilia Chemeli; Debela, Bethelhem Legesse; Qaim, Matin Women play key roles in food systems, yet continue to face persistent disadvantages in terms of low decision-making power and limited access to goods, services, and markets. Discrimination against women is often deeply ingrained in social norms, policies, and institutions. Widely observed gender gaps are not only unfair; they also undermine broader sustainability objectives. Extensive evidence shows that women's empowerment contributes to productivity, efficiency, and broader social welfare gains. We review and synthesize the literature on links between women's empowerment and nutrition, focusing on rural households in Africa and Asia. We analyze advances in the measurement of women's empowerment, discuss strengths and limitations of existing metrics, and summarize the broad empirical evidence showing that women's empowerment is positively associated with dietary quality and nutrition. Further, we develop a conceptual framework, highlighting key mechanisms of the empowerment-nutrition relationship, including women's bargaining power, control over income, and time allocation. Using this framework and examples from different countries, we show that development initiatives, such as promoting agricultural commercialization and women's off-farm employment, can involve tradeoffs, sometimes resulting in undesirable empowerment and/or nutrition outcomes. Such tradeoffs need to be properly understood and addressed through gender-transformative policies. We conclude by discussing policy and research implications.

Role of agency in envisioning future human-nature relationships in the context of road infrastructure development in the Kavango-Zambezi region, Namibia

2025-10-09T00:00:00Z

Role of agency in envisioning future human-nature relationships in the context of road infrastructure development in the Kavango-Zambezi region, Namibia Musa, Judith K.; Moseti, Vincent; Biber-Freudenberger, Lisa Amid growing concerns over climate change and biodiversity loss, there is increasing recognition of the need to reconcile local communities' economic aspirations with environmental conservation. The Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) introduced the Nature Futures Framework (NFF) to foster nature-positive future scenarios. This study explores the impact of road infrastructure on local communities' agency to envision and achieve desirable futures, focusing on Namibia's Kavango-Zambezi (KAZA) region. Using semi-structured questions and participatory mapping, we assessed how communities near and far from the Trans-Caprivi highway value nature in present, probable, and desirable future scenarios, as defined by the NFF. We also analyzed the impacts of socioeconomic factors such as age, education, occupation, and gender on shaping these visions. Higher education levels were associated with higher overall agency among respondents, both near and far from roads. Additionally, proximity to roads corresponded with higher agency scores for instrumental (Nature for Society) and intrinsic (Nature for Nature) values, while slightly lower scores were observed for relational (Nature as Culture) values. These patterns suggest spatial and educational factors may influence how individuals perceive their ability to shape future human–nature relationships across different value dimensions. These insights underscore the crucial need to foster nature-positive and socially inclusive futures by systematically integrating local knowledge and stakeholder perspectives into infrastructure planning and decision-making processes.