Large language models (LLM) agents may offer tools to predict human responses to surveys. A common technique for defining these agents uses only demographics, for example country, age, gender, employment status, income, education and marital status. We compare the predictive accuracy of demographic agents to that of survey agents defined with a larger set of in-domain survey responses. We test both approaches in predicting responses to the multidisciplinary, cross-national Survey of Health, Ageing and Retirement in Europe (SHARE), focusing on five variables from three policy-relevant constructs around personal finance. In these three constructs, we observe that, compared to survey agents trained on broader data, demographics-only agents (1) exhibited a central tendency bias, skewing answers toward population means, and (2) were unrealistically accurate, failing to reproduce the incorrect answers and "don't know" responses typical of human respondents. These performance differences are further substantiated through the replication of a hierarchical regression analysis from prior retirement planning research. Agents based solely on demographic information reproduce the outcome that financial risk tolerance, future time perspective, and knowledge of retirement planning each are predictive of retirement savings. However, only the survey-anchored agents succeed in reproducing the interaction among these three factors. These findings suggest caution in using only demographics to define LLM agents for predicting survey responses.
The rapidly growing computational demands of diffusion models for image generation have raised significant concerns about energy consumption and environmental impact. While existing approaches to energy optimization focus on architectural improvements or hardware acceleration, there is a lack of principled methods to predict energy consumption across different model configurations and hardware setups. We propose an adaptation of Kaplan scaling laws to predict GPU energy consumption for diffusion models based on computational complexity (FLOPs). Our approach decomposes diffusion model inference into text encoding, iterative denoising, and decoding components, with the hypothesis that denoising operations dominate energy consumption due to their repeated execution across multiple inference steps. We conduct comprehensive experiments across four state-of-the-art diffusion models (Stable Diffusion 2, Stable Diffusion 3.5, Flux, and Qwen) on three GPU architectures (NVIDIA A100, A4000, A6000), spanning various inference configurations including resolution (256^2–1024^2), precision (fp16/fp32), step counts (10–50), and classifier-free guidance settings. Our energy scaling law achieves high predictive accuracy within individual architectures (R^2 > 0.9) and exhibits strong cross-architecture generalization, maintaining high rank correlations across models and enabling reliable energy estimation for unseen model–hardware combinations. These results validate the compute-bound nature of diffusion inference and establish energy consumption estimation as a necessary foundation for sustainable AI deployment planning and subsequent carbon footprint assessment.
When used in high-stakes settings, AI systems are expected to produce decisions that are transparent, interpretable, and auditable, a requirement increasingly expected by regulations. Decision trees such as CART provide clear and verifiable rules, but they are restricted to structured tabular data and cannot operate directly on unstructured inputs such as text. In practice, large language models (LLMs) are widely used for such data, yet prompting strategies such as chain-of-thought or prompt optimization still rely on free-form reasoning, limiting their ability to ensure trustworthy behaviors. We present the Agentic Classification Tree (ACT), which extends decision-tree methodology to unstructured inputs by formulating each split as a natural-language question, refined through impurity-based evaluation and LLM feedback via TextGrad. Experiments on text benchmarks show that ACT matches or surpasses prompting-based baselines while producing transparent and interpretable decision paths.
We propose a simple way to use large language models (LLMs) in education. Specifically, our method aims to improve individual comprehension by adding a novel feature to online videos. We combine the low threshold for interactivity in digital experiences with the benefits of rephrased and elaborated explanations typical of face-to-face interactions, thereby supporting to close knowledge gaps at scale. To demonstrate the technical feasibility of our approach, we conducted a proof-of-concept experiment and implemented a prototype which is available for testing online. Through the use case, we also show how caching can be applied in LLM-powered applications to reduce their carbon footprint.
Plural counterfactual examples have been proposed to explain the prediction of a classifier by offering a user several instances of minimal modifications that may be performed to change the prediction. Yet, such explanations may provide too much information, generating potential confusion for the end-users with no specific knowledge, neither on the machine learning, nor on the application domains. In this paper, we investigate the design of explanation user interfaces for plural counterfactual examples offering comparative analysis features to mitigate this potential confusion and improve the intelligibility of such explanations for non-expert users. We propose an implementation of such an enhanced explanation user interface, illustrating it in a financial scenario related to a loan application. We then present the results of a lab user study conducted with 112 participants to evaluate the effectiveness of having plural examples and of offering comparative analysis principles, both on the objective understanding and satisfaction of such explanations. The results demonstrate the effectiveness of the plural condition, both on objective understanding and satisfaction scores, as compared to having a single counterfactual example. Beside the statistical analysis, we perform a thematic analysis of the participants' responses to the open-response questions, that also shows encouraging results for the comparative analysis features on the objective understanding.
Most works on the fairness of machine learning systems focus on the blind optimization of common fairness metrics, such as Demographic Parity and Equalized Odds. In this paper, we conduct a comparative study of several bias mitigation approaches to investigate their behaviors at a fine grain, the prediction level. Our objective is to characterize the differences between fair models obtained with different approaches. With comparable performances in fairness and accuracy, are the different bias mitigation approaches impacting a similar number of individuals? Do they mitigate bias in a similar way? Do they affect the same individuals when debiasing a model? Our findings show that bias mitigation approaches differ a lot in their strategies, both in the number of impacted individuals and the populations targeted. More surprisingly, we show these results even apply for several runs of the same mitigation approach. These findings raise questions about the limitations of the current group fairness metrics, as well as the arbitrariness, hence unfairness, of the whole debiasing process.
Graph models are fundamental to any kind of application on structured real-world problems. Any comparison between graphs by a graph distance measure requires the solution of the inexact graph matching problem, which constitutes a hard combinatorial optimization problem. An inexact matching problem includes in its formulation robustness to any type of perturbation, such as, for instance, noise, inherently present in real-world environments. In this paper, we introduce the concept of distance-preserving crossover operators for genetic algorithms for this task. For large graphs, our algorithm outperforms any state-of-the-art approximate algorithm-in particular, genetic algorithms with alternative crossover operators, which are to the best of our knowledge currently limited to no more than 50 nodes. We use a two-level local search heuristic to further enhance the results, pushing the limits to up to 300 nodes: a first local search step is directly integrated into the crossover operator; another one is applied independently during offspring generation.
Usual car navigation systems are configured to propose either the shortest or the fastest path between any origin–destination pair, neglecting the particularities of the territory. Some roads are impracticable when raining, some others are to avoid at night for the scarce lighting, or less safe for the presence of criminality and high accident ratio. On the other hand, longer paths can be safer and more pleasant as they pass through less noisy zones, with the presence of beautiful landscapes. In this paper we analyze the faults in current car navigation systems, especially quantifying the trade-off between safety and traveling time or path length. We propose two multi-criteria route planning methods, HVT (Hierarchical with Variable Tolerance) and R2V (Route to Vector), suggesting the best path to drivers also considering safety or multiple drivers’ specific needs. A dataset of 3,170 paths from 600 origin/destination pairs within London is created and shared to the research community. With this dataset, we show that selecting routes with reduced driving risks is indeed possible with a marginal increase in travel times. • Analysis of the impacts of a multi-criteria route selection in route features. • Four different techniques to measure route features. • Two novel route selection methods. • A novel dataset with routes information and GPS coordinates.
The increasing usage of complex Machine Learning models for decision-making has raised interest in explainable artificial intelligence (XAI). In this work, we focus on the effects of providing accessible and useful explanations to non-expert users. More specifically, we propose generic XAI design principles for contextualizing and allowing the exploration of explanations based on local feature importance. To evaluate the effectiveness of these principles for improving users' objective understanding and satisfaction, we conduct a controlled user study with 80 participants using 4 different versions of our XAI system, in the context of an insurance scenario. Our results show that the contextualization principles we propose significantly improve user's satisfaction and is close to have a significant impact on user's objective understanding. They also show that the exploration principles we propose improve user's satisfaction. On the other hand, the interaction of these principles does not appear to bring improvement on both dimensions of users' understanding.
Explainability is becoming an important requirement for organizations that make use of automated decision-making due to regulatory initiatives and a shift in public awareness. Various and significantly different algorithmic methods to provide this explainability have been introduced in the field, but the existing literature in the machine learning community has paid little attention to the stakeholder whose needs are rather studied in the human-computer interface community. Therefore, organizations that want or need to provide this explainability are confronted with the selection of an appropriate method for their use case. In this paper, we argue there is a need for a methodology to bridge the gap between stakeholder needs and explanation methods. We present our ongoing work on creating this methodology to help data scientists in the process of providing explainability to stakeholders. In particular, our contributions include documents used to characterize XAI methods and user requirements (shown in Appendix), which our methodology builds upon.
Veículos autônomos (AVs) usam sistemas de sensoriamento externo para obter informações sobre os diversos objetos presentes ao seu redor e assim tomar decisões de condução de forma automatizada. De fato, as informações obtidas através da telemática veicular servem para monitorar o funcionamento e a forma de condução de quaisquer veículos automotores, não apenas dos AVs. De acordo com a literatura, diferentes métricas são usadas por aplicativos de asseguradoras para avaliar situações de risco potencial e como os motoristas dirigem. Quanto a AVs, porém, a avaliação destas métricas ainda é incipiente, dado que a maioria destes veículos está em desenvolvimento. Assim, com o objetivo de avaliar e quantificar a gravidade de eventos de risco no trânsito envolvendo AVs e outros usuários viários, este trabalho estuda métricas de segurança substitutas baseado em leituras de sensores exteroceptivos e proprioceptivos contidas em conjuntos de dados de AVs. Os resultados indicam que ao menos 10% dos eventos representam algum nível risco para o AV. Estes resultados ajudam a explicar o comportamento e a tomada de decisões no sistema autônomo.
A aquisição cada vez maior de informações pessoais para personalização de serviços coloca em cheque a privacidade dos usuários. No contexto do aprendizado federado, a privacidade pode ser preservada com o compartilhamento apenas de pesos sinápticos de redes neurais entre clientes e servidores. Este trabalho avalia o impacto de diferentes parâmetros de redes de computadores no desempenho de modelos de aprendizado federado em um cenário composto por clientes móveis. Para isso, o desempenho das redes neurais convolucionais para classificação de imagens é considerado com o uso do conjunto de dados CIFAR-10. Os experimentos realizados utilizam o framework Flower para avaliar parâmetros comuns em redes móveis como latência, conectividade, volume de dados e disponibilidade dos clientes. Os resultados indicam que, além do aumento no tempo total de treinamento, um aumento no número de usuários desconectados em uma rodada de treinamento pode até mesmo reduzir o desempenho do modelo federado. Esses resultados reforçam a necessidade de orquestração cliente-servidor para adaptação dinâmica às condições de rede.
This paper analyses the fundamental ingredients behind surrogate explanations to provide a better understanding of their inner workings. We start our exposition by considering global surrogates, describing the trade-off between complexity of the surrogate and fidelity to the black-box being modelled. We show that transitioning from global to local - reducing coverage - allows for more favourable conditions on the Pareto frontier of fidelity-complexity of a surrogate. We discuss the interplay between complexity, fidelity and coverage, and consider how different user needs can lead to problem formulations where these are either constraints or penalties. We also present experiments that demonstrate how the local surrogate interpretability procedure can be made interactive and lead to better explanations.
Most fair regression algorithms mitigate bias towards sensitive sub populations and therefore improve fairness at group level. In this paper, we investigate the impact of such implementation of fair regression on the individual. More precisely, we assess the evolution of continuous predictions from an unconstrained to a fair algorithm by comparing results from baseline algorithms with fair regression algorithms for the same data points. Based on our findings, we propose a set of post-processing algorithms to improve the utility of the existing fair regression approaches.
Machine Learning has provided new business opportunities in the insurance industry, but its adoption is for now limited by the difficulty to explain the rationale behind the prediction provided. In this work, we explore how we can enhance local feature importance explanations for non-expert users. We propose design principles to contextualise these explanations with additional information about the Machine Learning system, the domain and external factors that may influence the prediction. These principles are applied to a car insurance smart pricing interface. We present preliminary observations collected during a pilot study using an online A/B test to measure objective understanding, perceived understanding and perceived usefulness of explanations. The preliminary results are encouraging as they hint that providing contextualisation elements can improve the understanding of ML predictions.
Local surrogate approaches for explaining machine learning model predictions have appealing properties, such as being model-agnostic and flexible in their modelling. Several methods exist that fit this description and share this goal. However, despite their shared overall procedure, they set out different objectives, extract different information from the black-box, and consequently produce diverse explanations, that are -- in general -- incomparable. In this work we review the similarities and differences amongst multiple methods, with a particular focus on what information they extract from the model, as this has large impact on the output: the explanation. We discuss the implications of the lack of agreement, and clarity, amongst the methods' objectives on the research and practice of explainability.
The fourth industrial revolution makes extensive use of IoT, AI, and smart sensors for improved automation, safety, production, and prognostics, and health management. In this paper, we address corrective maintenance based on fault recognition relying on sounds produced by machine components. Different spectral features are extracted from industrial sounds and are used as input of supervised learning algorithms for classification between normal and abnormal operations. Experiments using the MIMII (Malfunctioning Industrial Machine Investigation and Inspection) dataset, which contains sound samples produced by pump, slide rail, valve, and fan components, reveals promising results based on the f1-score. We also evaluate the impact of the different spectral features considered, confirming their incremental impact. Finally, we compare our proposal with a baseline alternative from the literature, which employs unsupervised learning and Mel-spectrogram conversion. Our approach improves the AUC (Area Under the Curve) metric by up to 39.5% compared with the baseline approach.
In recent years, significant work has been done to include fairness constraints in the training objective of machine learning algorithms. Differently from classical prediction retreatment algorithms, we focus on learning fair representations of the inputs. The challenge is to learn representations that capture most relevant information to predict the targeted output Y, while not containing any information about a sensitive attribute S. We leverage recent work which has been done to estimate the Hirschfeld-Gebelein-Renyi (HGR) maximal correlation coefficient by learning deep neural network transformations and use it as a min-max game to penalize the intrinsic bias in a multi dimensional latent representation. Compared to other dependence measures, the HGR coefficient captures more information about the non-linear dependencies, making the algorithm more efficient in mitigating bias. After providing a theoretical analysis of the consistency of the estimator and its desirable properties for bias mitigation, we empirically study its impact at various levels of neural architectures. We show that acting at intermediate levels of neural architectures provides best expressiveness/generalization abilities for bias mitigation, and that using an HGR based loss is more efficient than more classical adversarial approaches from the literature.
Andreas Nürnberger合作论文数Department for Technical & Operational Information Systems, Faculty of Computer Science, Otto-Von-Guericke-University Magdeburg25