arXiv subjects

Sebastian Stein

Publications and source records attributed to Sebastian Stein.

At least 19 recordsLinked to original sources

arXiv2026

Instruction-Conditioned Exploration for Reinforcement Learning with Self-Distillation to an Unconditioned Policy

Post-training Large Language Models (LLMs) with Reinforcement Learning (RL) has become an important tool for improving model capabilities, but the LLM action-space structure introduces challenges distinct from classical RL, with implications for inducing exploration. New methods are required that leverage the broad knowledge and flexibility of pre-trained LLMs to deliberately generate diverse experience at training time. We propose Instruction-Conditioned Exploration (ICE), which appends one of a small fixed set of instructions to task prompts during training, using the same set for every problem, increasing the coverage of behaviours attempted. To facilitate ICE, we combine RL on the instruction-conditioned policy with self-distillation of its correct rollouts into the unconditioned test-time policy. ICE with this objective improves Qwen3-1.7B held-out pass@1 performance at 4K response length on mathematical reasoning tasks by $5.0\%$ relative to training with DAPO, with improvement persisting at a longer 8K context. The improvement does not appear for Qwen3-4B at 4K, where the instructions do not expand base-model coverage.

cs.AI↗

arXiv2026

An AI-Based Decision-Support Pipeline for Day-Ahead Photovoltaic Forecasting

Reliable photovoltaic (PV) forecasts can support low-carbon energy systems, but deployed sites may have only short and incomplete records. Physical and hybrid methods can be sensitive to weather inputs, calibration, and timestamp-alignment, while individual machine learning models may capture different parts of the forecasting problem. We study hourly day-ahead PV forecasting at a United Kingdom charging station using one year of inverter measurements, with 9.25% of hours missing. The pipeline checks timestamp-alignment, derives solar and clearness features, adds short-term weather context, and combines five complementary models using non-negative least squares stacking, with the combination fitted only on validation observations. We compare against smart persistence, a weather-scaled baseline that carries the previous day's PV behaviour forward using target-day irradiance. With retrospective weather, the combined model reduces daylight normalised root mean square error (RMSE) by 31.2% under random day-fold evaluation and by 2.9% under rolling-origin evaluation, although the latter improvement is not robust across days. It also improves by 3.0% over the single model selected from validation performance. Replacing retrospective weather with a public product sampled at a constant 24-hour lead increases daylight RMSE by 13.1% and 4.2% under the two protocols, while retaining positive skill over smart persistence.

cs.LG↗

arXiv2026

AdaHome: An Adaptive Smart Home Assistant using Local Small Language Models

Smart home assistants interpret a wide range of user commands, from explicit device control to underspecified and preference dependent requests. While recent systems based on Large Language Models (LLMs) improve this capability, they often rely on heavyweight reasoning pipelines and cloud-based deployment, limiting their efficiency and suitability for resource-constrained environments, and raising privacy concerns. In addition, existing approaches provide limited support for stable long-term personalization. To address these issues, we present AdaHome, an adaptive smart home assistant designed for locally deployed small language models in smart home environments. Rather than applying complex reasoning uniformly, AdaHome introduces an intent-aware planning framework that dynamically routes commands either to straightforward prompt-based or lightweight reasoning-based components. For commands requiring interpretation, we adopt a Chain-of-Draft strategy to enable efficient and stable decision-making. To support personalization, we further propose a preference adaptation mechanism that learns from user feedback over time without requiring prompt augmentation or model retraining. We evaluate AdaHome against representative LLM-based baselines under a unified small model setting. AdaHome achieves substantially higher accuracy on direct commands (86.7%) while reducing latency by up to 3$\times$. Furthermore, it maintains competitive performance on ambiguous inputs with lower computational cost. In multi-turn scenarios, AdaHome achieves 88% preference consistency, compared to 52.5% for a prompt augmentation baseline.

cs.AI↗

arXiv2026

Three Lessons from Citizen-Centric Participatory AI Design

This workshop paper examines challenges in designing agentic AI systems from a citizen-centric perspective. Drawing on three participatory workshops conducted in 2025 with members of the general public and cross-sector stakeholders, we explore how societal values and expectations shape visions of future AI agents. Using constructive design research methods, participants engaged in storytelling and lo-fi prototyping to reflect on potential community impacts. We identify three key challenges: enabling meaningful and sustained public engagement, establishing a shared language between experts and lay participants, and translating speculative participant input into implementable systems. We argue that reflexive, long-term participation is essential for responsible and actionable citizen-centric AI development.

cs.CY↗

arXiv2025

Back to the Communities: A Mixed-Methods and Community-Driven Evaluation of Cultural Sensitivity in Text-to-Image Models

Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.

cs.SI↗

arXiv2025

Analysing Knowledge Construction in Online Learning: Adapting the Interaction Analysis Model for Unstructured Large-Scale Discourse

The rapid expansion of online courses and social media has generated large volumes of unstructured learner-generated text. Understanding how learners construct knowledge in these spaces is crucial for analysing learning processes, informing content design, and providing feedback at scale. However, existing approaches typically rely on manual coding of well-structured discussion forums, which does not scale to the fragmented discourse found in online learning. This study proposes and validates a framework that combines a codebook inspired by the Interaction Analysis Model with an automated classifier to enable large-scale analysis of knowledge construction in unstructured online discourse. We adapt four comment-level categories of knowledge construction: Non-Knowledge Construction, Share, Explore, and Integrate. Three trained annotators coded a balanced sample of 20,000 comments from YouTube education channels. The codebook demonstrated strong reliability, with Cohen's kappa = 0.79 on the main dataset and 0.85--0.93 across four additional educational domains. For automated classification, bag-of-words baselines were compared with transformer-based language models using 10-fold cross-validation. A DeBERTa-v3-large model achieved the highest macro-averaged F1 score (0.841), outperforming all baselines and other transformer models. External validation on four domains yielded macro-F1 above 0.705, with stronger transfer in medicine and programming, where discourse was more structured and task-focused, and weaker transfer in language and music, where comments were more varied and context-dependent. Overall, the study shows that theory-driven, semi-automated analysis of knowledge construction at scale is feasible, enabling the integration of knowledge-construction indicators into learning analytics and the design of online learning environments.

cs.CL↗

arXiv2025

An Active Inference Model of Mouse Point-and-Click Behaviour

We explore the use of Active Inference (AIF) as a computational user model for spatial pointing, a key problem in Human-Computer Interaction (HCI). We present an AIF agent with continuous state, action, and observation spaces, performing one-dimensional mouse pointing and clicking. We use a simple underlying dynamic system to model the mouse cursor dynamics with realistic perceptual delay. In contrast to previous optimal feedback control-based models, the agent's actions are selected by minimizing Expected Free Energy, solely based on preference distributions over percepts, such as observing clicking a button correctly. Our results show that the agent creates plausible pointing movements and clicks when the cursor is over the target, with similar end-point variance to human users. In contrast to other models of pointing, we incorporate fully probabilistic, predictive delay compensation into the agent. The agent shows distinct behaviour for differing target difficulties without the need to retune system parameters, as done in other approaches. We discuss the simulation results and emphasize the challenges in identifying the correct configuration of an AIF agent interacting with continuous systems.

cs.HC↗

arXiv2025

Optimal Network Pricing for Oblivious Users under Projected Decision-Dependent Distributions

Efficient large-scale network allocation requires data-driven pricing mechanisms that internalize stochastic, nonlinear user behavior. We move beyond the classic fully strategic agents to study oblivious users (agents with bounded rationality and imperfect information). Rather than assuming an infinite horizon, our regime acknowledges that real-world flows are too transient to equilibrate among users. We introduce a novel Optimal Network Pricing (ONP) problem for such users, which induces Performativity: a Decision-Dependent environment where pricing decisions endogenously shift the flow distribution. Without a closed-form distribution, the platform must learn optimal prices from sampled responses. This setting introduces a new challenge: capacity boundaries and projection operators make the optimization landscape nonsmooth, invalidating gradient-based methods. We show that a widely adopted optimality concept Performative Stability (PS) fails in ONP, collapsing to a trivial solution. We then define a new optimality concept, the Projected Performative Optimum ({\Pi}PO) for the unique global optimum. Targeting {\Pi}PO is algorithmically hard given the performative nonsmooth Jacobian, so we propose a novel framework combining Sample Average Approximation with Trust-Region Sequential Quadratic Programming, explicitly handling the capacity boundaries, with theoretical guarantees on probabilistic convexity, sample complexity, and computational complexity. Experiments show that our {\Pi}PO solver significantly outperforms PS-seeking heuristics and a proposed baseline (improving social welfare by 81\% on GEANT), highlighting that properly handling capacity boundaries unlocks substantial gains in social welfare. More broadly, this work advances intelligent systems that learn under performative, capacity-constrained feedback, a core challenge in real-world AI applications.

math.NA↗

arXiv2025

Reinforced Language Models for Sequential Decision Making

Large Language Models (LLMs) show potential as sequential decision-making agents, but their application is often limited due to a reliance on large, computationally expensive models. This creates a need to improve smaller models, yet existing post-training methods are designed for single-turn interactions and cannot handle credit assignment in multi-step agentic tasks. To address this, we introduce Multi-Step Group-Relative Policy Optimization (MS-GRPO), a new algorithm for post-training LLM agents, grounded in formal Text-Mediated Stochastic Game (TSMG) and Language-Agent Policy (LAP) frameworks. For credit assignment, MS-GRPO attributes the entire cumulative episode reward to each individual episode step. We supplement this algorithm with a novel absolute-advantage-weighted episode sampling strategy that we show improves training performance. We evaluate our approach by post-training a 3-billion parameter model on Snake and Frozen Lake. Our experiments demonstrate that the method is effective in improving decision-making performance: our post-trained 3B parameter model outperforms a 72B parameter baseline by 50% on the Frozen Lake task. This work demonstrates that targeted post-training is a practical and efficient alternative to relying on model scale for creating sequential decision-making agents using LLMs.

cs.CL↗

arXiv2025

HMCF: A Human-in-the-loop Multi-Robot Collaboration Framework Based on Large Language Models

Rapid advancements in artificial intelligence (AI) have enabled robots to performcomplex tasks autonomously with increasing precision. However, multi-robot systems (MRSs) face challenges in generalization, heterogeneity, and safety, especially when scaling to large-scale deployments like disaster response. Traditional approaches often lack generalization, requiring extensive engineering for new tasks and scenarios, and struggle with managing diverse robots. To overcome these limitations, we propose a Human-in-the-loop Multi-Robot Collaboration Framework (HMCF) powered by large language models (LLMs). LLMs enhance adaptability by reasoning over diverse tasks and robot capabilities, while human oversight ensures safety and reliability, intervening only when necessary. Our framework seamlessly integrates human oversight, LLM agents, and heterogeneous robots to optimize task allocation and execution. Each robot is equipped with an LLM agent capable of understanding its capabilities, converting tasks into executable instructions, and reducing hallucinations through task verification and human supervision. Simulation results show that our framework outperforms state-of-the-art task planning methods, achieving higher task success rates with an improvement of 4.76%. Real-world tests demonstrate its robust zero-shot generalization feature and ability to handle diverse tasks and environments with minimal human intervention.

cs.RO↗

arXiv2025

Improving Controller Generalization with Dimensionless Markov Decision Processes

Controllers trained with Reinforcement Learning tend to be very specialized and thus generalize poorly when their testing environment differs from their training one. We propose a Model-Based approach to increase generalization where both world model and policy are trained in a dimensionless state-action space. To do so, we introduce the Dimensionless Markov Decision Process ($\Pi$-MDP): an extension of Contextual-MDPs in which state and action spaces are non-dimensionalized with the Buckingham-$\Pi$ theorem. This procedure induces policies that are equivariant with respect to changes in the context of the underlying dynamics. We provide a generic framework for this approach and apply it to a model-based policy search algorithm using Gaussian Process models. We demonstrate the applicability of our method on simulated actuated pendulum and cartpole systems, where policies trained on a single environment are robust to shifts in the distribution of the context.

cs.LG↗

arXiv2025

PTFA: An LLM-based Agent that Facilitates Online Consensus Building through Parallel Thinking

Consensus building is inherently challenging due to the diverse opinions held by stakeholders. Effective facilitation is crucial to support the consensus building process and enable efficient group decision making. However, the effectiveness of facilitation is often constrained by human factors such as limited experience and scalability. In this research, we propose a Parallel Thinking-based Facilitation Agent (PTFA) that facilitates online, text-based consensus building processes.The PTFA automatically collects real-time textual input and leverages large language models (LLMs)to perform all six distinct roles of the well-established Six Thinking Hats technique in parallel thinking.To illustrate the potential of the agent, a pilot study was conducted, demonstrating its capabilities in idea generation, emotional probing, and deeper analysis of idea quality. Additionally, future open research challenges such as optimizing scheduling and managing behaviors in divergent phase are identified. Furthermore, a comprehensive dataset that contains not only the conversational content among the participants but also between the participants and the agent is constructed for future study.

cs.HC↗

arXiv2025

When not to target negative ties? Studying competitive influence maximisation in signed networks

We explore the influence maximisation problem in networks with negative ties. Where prior work has focused on unsigned networks, we investigate the need to consider negative ties in networks while trying to maximise spread in a population - particularly under competitive conditions. Given a signed network we optimise the strategies of a focal controller, against competing influence in the network, using two approaches - either the focal controller uses a sign-agnostic approach or they factor in the sign of the edges while optimising their strategy. We compare the difference in vote-shares (or the share of population) obtained by both these methods to determine the need to navigate negative ties in these settings. More specifically, we study the impact of: (a) network topology, (b) resource conditions and (c) competitor strategies on the difference in vote shares obtained across both methodologies. We observe that gains are maximum when resources available to the focal controller are low and the competitor avoids negative edges in their strategy. Conversely, gains are insignificant irrespective of resource conditions when the competitor targets the network indiscriminately. Finally, we study the problem in a game-theoretic setting, where we simultaneously optimise the strategies of both competitors. Interestingly we observe that, strategising with the knowledge of negative ties can occasionally also lead to loss in vote-shares.

cs.SI↗

arXiv2025

TutorLLM: Customizing Learning Recommendations with Knowledge Tracing and Retrieval-Augmented Generation

The integration of AI in education offers significant potential to enhance learning efficiency. Large Language Models (LLMs), such as ChatGPT, Gemini, and Llama, allow students to query a wide range of topics, providing unprecedented flexibility. However, LLMs face challenges, such as handling varying content relevance and lack of personalization. To address these challenges, we propose TutorLLM, a personalized learning recommender LLM system based on Knowledge Tracing (KT) and Retrieval-Augmented Generation (RAG). The novelty of TutorLLM lies in its unique combination of KT and RAG techniques with LLMs, which enables dynamic retrieval of context-specific knowledge and provides personalized learning recommendations based on the student's personal learning state. Specifically, this integration allows TutorLLM to tailor responses based on individual learning states predicted by the Multi-Features with Latent Relations BERT-based KT (MLFBK) model and to enhance response accuracy with a Scraper model. The evaluation includes user assessment questionnaires and performance metrics, demonstrating a 10% improvement in user satisfaction and a 5\% increase in quiz scores compared to using general LLMs alone.

cs.IR↗

arXiv2024

Active Inference and Human--Computer Interaction

Active Inference is a closed-loop computational theoretical basis for understanding behaviour, based on agents with internal probabilistic generative models that encode their beliefs about how hidden states in their environment cause their sensations. We review Active Inference and how it could be applied to model the human-computer interaction loop. Active Inference provides a coherent framework for managing generative models of humans, their environments, sensors and interface components. It informs off-line design and supports real-time, online adaptation. It provides model-based explanations for behaviours observed in HCI, and new tools to measure important concepts such as agency and engagement. We discuss how Active Inference offers a new basis for a theory of interaction in HCI, tools for design of modern, complex sensor-based systems, and integration of artificial intelligence technologies, enabling it to cope with diversity in human users and contexts. We discuss the practical challenges in implementing such Active Inference-based systems.

cs.HC↗

arXiv2024

Responsible AI Governance: A Response to UN Interim Report on Governing AI for Humanity

This report presents a comprehensive response to the United Nation's Interim Report on Governing Artificial Intelligence (AI) for Humanity. It emphasizes the transformative potential of AI in achieving the Sustainable Development Goals (SDGs) while acknowledging the need for robust governance to mitigate associated risks. The response highlights opportunities for promoting equitable, secure, and inclusive AI ecosystems, which should be supported by investments in infrastructure and multi-stakeholder collaborations across jurisdictions. It also underscores challenges, including societal inequalities exacerbated by AI, ethical concerns, and environmental impacts. Recommendations advocate for legally binding norms, transparency, and multi-layered data governance models, alongside fostering AI literacy and capacity-building initiatives. Internationally, the report calls for harmonising AI governance frameworks with established laws, human rights standards, and regulatory approaches. The report concludes with actionable principles for fostering responsible AI governance through collaboration among governments, industry, academia, and civil society, ensuring the development of AI aligns with universal human values and the public good.

cs.CY↗

arXiv2024

Integrating LSTM and BERT for Long-Sequence Data Analysis in Intelligent Tutoring Systems

The field of Knowledge Tracing aims to understand how students learn and master knowledge over time by analyzing their historical behaviour data. To achieve this goal, many researchers have proposed Knowledge Tracing models that use data from Intelligent Tutoring Systems to predict students' subsequent actions. However, with the development of Intelligent Tutoring Systems, large-scale datasets containing long-sequence data began to emerge. Recent deep learning based Knowledge Tracing models face obstacles such as low efficiency, low accuracy, and low interpretability when dealing with large-scale datasets containing long-sequence data. To address these issues and promote the sustainable development of Intelligent Tutoring Systems, we propose a LSTM BERT-based Knowledge Tracing model for long sequence data processing, namely LBKT, which uses a BERT-based architecture with a Rasch model-based embeddings block to deal with different difficulty levels information and an LSTM block to process the sequential characteristic in students' actions. LBKT achieves the best performance on most benchmark datasets on the metrics of ACC and AUC. Additionally, an ablation study is conducted to analyse the impact of each component of LBKT's overall performance. Moreover, we used t-SNE as the visualisation tool to demonstrate the model's embedding strategy. The results indicate that LBKT is faster, more interpretable, and has a lower memory cost than the traditional deep learning based Knowledge Tracing methods.

cs.CY↗

arXiv2024

Combinatorial Client-Master Multiagent Deep Reinforcement Learning for Task Offloading in Mobile Edge Computing

Recently, there has been an explosion of mobile applications that perform computationally intensive tasks such as video streaming, data mining, virtual reality, augmented reality, image processing, video processing, face recognition, and online gaming. However, user devices (UDs), such as tablets and smartphones, have a limited ability to perform the computation needs of the tasks. Mobile edge computing (MEC) has emerged as a promising technology to meet the increasing computing demands of UDs. Task offloading in MEC is a strategy that meets the demands of UDs by distributing tasks between UDs and MEC servers. Deep reinforcement learning (DRL) is gaining attention in task-offloading problems because it can adapt to dynamic changes and minimize online computational complexity. However, the various types of continuous and discrete resource constraints on UDs and MEC servers pose challenges to the design of an efficient DRL-based task-offloading strategy. Existing DRL-based task-offloading algorithms focus on the constraints of the UDs, assuming the availability of enough storage resources on the server. Moreover, existing multiagent DRL (MADRL)--based task-offloading algorithms are homogeneous agents and consider homogeneous constraints as a penalty in their reward function. We proposed a novel combinatorial client-master MADRL (CCM\_MADRL) algorithm for task offloading in MEC (CCM\_MADRL\_MEC) that enables UDs to decide their resource requirements and the server to make a combinatorial decision based on the requirements of the UDs. CCM\_MADRL\_MEC is the first MADRL in task offloading to consider server storage capacity in addition to the constraints in the UDs. By taking advantage of the combinatorial action selection, CCM\_MADRL\_MEC has shown superior convergence over existing MADDPG and heuristic algorithms.

cs.AI↗