9607 research outputs found
Sort by
Don’t Get Too Excited - Eliciting Emotions in LLMs
No full textThis paper investigates the challenges of affect control in large language models (LLMs), focusing on their ability to express appropriate emotional states during extended dialogues. We evaluated state-of-the-art open-weight LLMs to assess their affective expressive range in terms of arousal and valence. Our study employs a novel methodology combining LLM-based sentiment analysis with multiturn dialogue simulations between LLMs.We quantify the models' capacity to express a wide spectrum of emotions and how they fluctuate during interactions. Our findings reveal significant variations among LLMs in their ability to maintain consistent affect, with some models demonstrating more stable emotional trajectories than others. Furthermore, we identify key challenges in affect control, including difficulties in producing and maintaining extreme emotional states and limitations in adapting affect to changing conversational contexts. These findings have important implications for the development of more emotionally intelligent AI systems and highlight the need for improved affect modelling in LLMs
Urban Data Science for Sustainable Mobility
Get PDFTransportation is a major sector of human activity fueling the climate crisis. Therefore, there is an urgent need to make mobility more sustainable. Due to contemporary urbanization, this need is particularly pressing within cities, and can be addressed through the emerging interdisciplinary field of Urban Data Science. This field combines methods from Data Science with domain knowledge of the city, making use of the increasing availability and volume of data on urban environments.In the realm of mobility, sustainability is gaining popularity as a framework for research and policy-making. Nevertheless, available data, tools, and research efforts for sustainable transportation modes like cycling are still dwarfed by those for motorized modes, constituting a considerable research gap. In addition, techno-optimist approaches to sustainability, such as the excessive supportfor car-centric motorized transportation systems, entail serious ethical pitfalls. To address these challenges, this thesis explores how Urban Data Science can ethically support human mobility that is both environmentally and socially sustainable, through the two lenses of Data and Networks. In the realm of Data, we develop an algorithm for multi-purpose spatial network simplification, and a data quality assessment pipeline tailored specifically to bicycle networks. Further, we outline pathways for incorporating data ethics into computational approaches to spatial manifestations of social inequalities. In the realm of Networks, we develop data-driven methods for the planning of bicycle networks and low-traffic neighbourhoods, and showcase their application to various cities.Lastly, we investigate the impact of transportation infrastructure on social connections in cities, quantitatively corroborating that urban highways are barriers to social ties. Stemming from various interdisciplinary collaborations, the results of this thesis cover multiple conceptual levels of Urban Data Science, from open source software development and data quality assessment to transportation network planning and the intersection of social and spatial networks. Through these efforts, this thesis advances the emerging field of Urban Data Science, showcasing the field’s potential to make human mobility more sustainable
Credit Data, Banks, “Packaging Agencies” and the Promise of Digital Lending to Small Businesses in China
No full textMicro, small and medium-sized enterprises (MSMEs) in China often struggle to secure loan financing. In response, the government has required banks to increase credit for MSMEs and incorporate digital technologies into traditional credit evaluation models. In 2018, the state introduced “credit easy loan” digital lending platforms under its social credit system to facilitate collateral-free loans for MSMEs in a bid to enhance financial inclusion and social trust. Meanwhile, the actual implementation of these initiatives remains understudied. Drawing on six months of ethnographic fieldwork, this paper examines how “packaging agencies” act as intermediaries in preparing and “beautifying” bank loan applications. These agencies may manipulate credit data and leverage close relationships (guanxi) to help clients obtain loans, while banks may tacitly approve their practices to fulfil their financial inclusion requirements. Through such processes and in a supposedly digitalized system, a single MSME loan multiplied into ten loans, large companies became small businesses, and one housewife became a creditworthy microentrepreneur
Beyond Isolated Factors: Investigating the Configurations That Shape User Responses to AI Advice
No full textAs artificial intelligence (AI) increasingly supports decision-making across various sectors, understanding how users perceive and respond to AI advice is critical. While previous studies have identified individual factors impacting this interaction between user and AI advice, these factors often interact in complex ways. This study explores how combinations of factors, including performance expectancy, effort expectancy, personal data use, and prediction explainability, influence user decisions to accept or reject AI advice. Using General Systems Theory as a framework and conducting experiments in a medical context with a Symptom Checker application, we aim to investigate these interdependencies through fuzzy-set qualitative comparative analysis. Our findings will contribute to the growing body of research on AI-advised decision-making by identifying key configurations that drive user acceptance or rejection of AI advice, offering insights both for academia and practical applications
SnakModel: Lessons Learned from Training an Open Danish Large Language Model
No full textWe present SnakModel, a Danish large language model (LLM) based on Llama2-7B, which we continuously pre-train on 13.6B Danish words, and further tune on 3.7M Danish instructions. As best practices for creating LLMs for smaller language communities have yet to be established, we examine the effects of early modeling and training decisions on downstream performance throughout the entire training pipeline, including (1) the creation of a strictly curated corpus of Danish text from diverse sources; (2) the language modeling and instruction-tuning training process itself, including the analysis of intermediate training dynamics, and ablations across different hyperparameters; (3) an evaluation on eight language and culturally-specific tasks. Across these experiments SnakModel achieves the highest overall performance, outperforming multiple contemporary Llama2-7B-based models. By making SnakModel, the majority of our pre-training corpus, and the associated code available under open licenses, we hope to foster further research and development in Danish Natural Language Processing, and establish training guidelines for languages with similar resource constraints
DaKultur: Evaluating the Cultural Awareness of Language Models for Danish with Native Speakers
No full textLarge Language Models (LLMs) have seen widespread societal adoption. However, while they are able to interact with users in languages beyond English, they have been shown to lack cultural awareness, providing anglocentric or inappropriate responses for underrepresented language communities. To investigate this gap and disentangle linguistic versus cultural proficiency, we conduct the first cultural evaluation study for the mid-resource language of Danish, in which native speakers prompt different models to solve tasks requiring cultural awareness. Our analysis of the resulting 1,038 interactions from 63 demographically diverse participants highlights open challenges to cultural adaptation: Particularly, how currently employed automatically translated data are insufficient to train or measure cultural adaptation, and how training on native-speaker data can more than double response acceptance rates. We release our study data as DaKultur - the first native Danish cultural awareness dataset
Improving Reasoning Performance in Large Language Models via Representation Engineering
No full textRecent advancements in large language models (LLMs) have resulted in increasingly anthropomorphic language concerning the ability of LLMs to reason. Whether reasoning in LLMs should be understood to be inherently different is, however, widely debated. We propose utilizing a representation engineering approach wherein model activations are read from the residual stream of an LLM when processing a reasoning task. The activations are used to derive a control vector that is applied to the model as an inference-time intervention, modulating the representational space of the model, to improve performance on the specified task. We publish the code for deriving control vectors and analyzing model representations. The method allows us to improve performance on reasoning benchmarks and assess how control vectors influence the final logit distribution of a model via metrics such as KL divergence and entropy. We apply control vectors to Mistral-7B-Instruct and a range of Pythia models on an inductive, a deductive and mathematical reasoning task. We show that an LLM can, to a certain degree, be controlled to improve its perceived reasoning ability by modulating activations. The intervention is dependent upon the ability to reliably extract the model's typical state when correctly solving a task. Our results suggest that reasoning performance can be modulated in the same manner as other information-processing tasks performed by LLMs and demonstrate that we are capable of improving performance on specific tasks via a simple intervention on the residual stream with no additional training
Deep Reinforcement Learning for Revenue Management under Uncertainty in Master Stowage Planning on Container Vessels
Get PDFAdvanced planning policies obtained by machine learning have shown promisingresults in solving well-known combinatorial optimization problems in transportation and logistics. However, a significant challenge arises when dealing with complex action spaces in realistic planning, where it is less straightforward for machine learning models to generate feasible actions. A relevant and complex example is master stowage planning on container vessels, which plays a crucial role in global trade and the green transition. This planning problem aims to maximize cargo revenue and minimize operational costs while addressing strict constraints and demand uncertainty. To tackle this challenge, our paper introduces a deep reinforcement learning framework with a general feasibility layer to solve a novel Markov decision process of master stowage planning under demand uncertainty. The experimental evaluation shows that our architecture efficiently finds feasible solutions for a multistage stochastic optimization problem, which is intractable using traditional benchmark methods from combinatorial optimization. Our approach demonstrates the potential of advanced planning policies to tackle complex, real-world problems, with implications for global trade and sustainability
Deep Reinforcement Learning for Revenue Management under Uncertainty in Master Stowage Planning on Container Vessels
Get PDFAdvanced planning policies obtained by machine learning have shown promisingresults in solving well-known combinatorial optimization problems in transportation and logistics. However, a significant challenge arises when dealing with complex action spaces in realistic planning, where it is less straightforward for machine learning models to generate feasible actions. A relevant and complex example is master stowage planning on container vessels, which plays a crucial role in global trade and the green transition. This planning problem aims to maximize cargo revenue and minimize operational costs while addressing strict constraints and demand uncertainty. To tackle this challenge, our paper introduces a deep reinforcement learning framework with a general feasibility layer to solve a novel Markov decision process of master stowage planning under demand uncertainty. The experimental evaluation shows that our architecture efficiently finds feasible solutions for a multistage stochastic optimization problem, which is intractable using traditional benchmark methods from combinatorial optimization. Our approach demonstrates the potential of advanced planning policies to tackle complex, real-world problems, with implications for global trade and sustainability
Critical Perspectives on Predictive Policing:Anticipating Proof?
No full textTaking a critical approach, this book advances understanding of the social, legal and ethical aspects of digitalisation in law enforcement and the reliance on data-driven tools to predict and prevent crime. It shows how the proliferation of data analytics challenges citizens’ rights, at a time when what counts as ‘safety’ or ‘policing’ is being fundamentally transformed