Deploying Large Language Model (LLM) services at the edge benefits latency-sensitive and privacy-aware applications. However, the stateless nature of LLMs makes managing user context (e.g., sessions, preferences) across geo-distributed edge nodes challenging. Existing solutions, such as client-side context storage, often introduce network latency and bandwidth overhead, undermining the advantages of edge deployment. We propose DisCEdge, a distributed context management system that stores and replicates user context in tokenized form across edge nodes. By maintaining context as token sequences rather than raw text, our system avoids redundant computation and enables efficient data replication. We implement and evaluate an open-source prototype in a realistic edge environment with commodity hardware. We show DisCEdge improves median response times by up to 14.46% and lowers median inter-node synchronization overhead by up to 15% compared to a raw-text-based system. It also reduces client request sizes by a median of 90% compared to client-side context management, while guaranteeing data consistency.

Recent text-to-image (T2I) models generate semantically coherent images from textual prompts, yet evaluating how well they align with individual user preferences remains an open challenge. Conventional evaluation methods, general reward functions or similarity-based metrics, fail to capture the diversity and complexity of personal visual tastes. In this work, we present PIGReward, a personalized reward model that dynamically generates user-conditioned evaluation dimensions and assesses images through CoT reasoning. T o address the scarcity of user data, PIGReward adopt a self-bootstrapping strategy that reasons over limited reference data to construct rich user contexts, enabling per-sonalization without user-specific training. Beyond evaluation, PIGReward provides personalized feedback that drives user-specific prompt optimization, improving alignment between generated images and individual intent. W e further introduce PIGBench, a per-user preference benchmark capturing diverse visual interpretations of shared prompts. Extensive experiments demonstrate that PIGReward surpasses existing methods in both accuracy and interpretability, establishing a scalable and reasoning-based foundation for personalized T2I evaluation and optimization. T aken together, our findings highlight PIGReward as a robust step toward individually aligned T2I generation.

Generative recommendation (GR) models tokenize each action into a few discrete tokens (called semantic IDs) and autoregressively generate the next tokens as predictions, showing advantages such as memory efficiency, scalability, and the potential to unify retrieval and ranking. Despite these benefits, existing tokenization methods are static and non-personalized. They typically derive semantic IDs solely from item features, assuming a universal item similarity that overlooks user-specific perspectives. However, under the autoregressive paradigm, semantic IDs with the same prefixes always receive similar probabilities, so a single fixed mapping implicitly enforces a universal item similarity standard across all users. In practice, the same item may be interpreted differently depending on user intentions and preferences. To address this issue, we propose a personalized context-aware tokenizer that incorporates a user's historical interactions when generating semantic IDs. This design allows the same item to be tokenized into different semantic IDs under different user contexts, enabling GR models to capture multiple interpretive standards and produce more personalized predictions. Experiments on three public datasets demonstrate up to 11.44% improvement in NDCG@10 over non-personalized action tokenization baselines. Our code is available at https://github.com/YoungZ365/Pctx.

As everyday use cases of large language model (LLM) AI assistants have expanded, it is becoming increasingly important to personalize responses to align to different users' preferences and goals. While reinforcement learning from human feedback (RLHF) is effective at improving LLMs to be generally more helpful and fluent, it does not account for variability across users, as it models the entire user population with a single reward model, meaning it assumes that everyone's preferences are the same. We present a novel framework, Preference Learning Using Summarization (PLUS), that uses reinforcement learning (RL) to learn to produce text-based summaries of each user's preferences, characteristics, and past conversations. These summaries condition the reward model, enabling it to make personalized predictions about the types of responses valued by each user. Both the user-summarization model and reward model are trained simultaneously, creating an online co-adaptation loop. We show that in contrast to the standard Bradley-Terry model, summaries produced by PLUS capture diverse aspects of user preferences, achieving a 11-77% improvement in reward model accuracy. Key strengths of PLUS are: (1) robust performance with new users and conversation topics, achieving a 25% improvement over the best personalized RLHF technique; (2) zero-shot personalization with state-of-the-art proprietary models like GPT -4 (e.g., PLUS-summary-conditioned responses achieved a 72% win rate compared to 28% for default GPT -4o); (3) learning from flexible user contexts beyond preference labels, and (4) interpretable representation of users, enabling greater transparency and user control in pluralistic LLM alignment.

Search-augmented large language models (LLMs) have advanced information-seeking tasks by integrating retrieval into generation, reducing users' cognitive burden compared to traditional search systems. Yet they remain insufficient for fully addressing diverse user needs, which requires recognizing how the same query can reflect different intents across users and delivering information in preferred forms. While recent systems such as ChatGPT and Gemini attempt personalization by leveraging user histories, systematic evaluation of such personalization is under-explored. To address this gap, we propose BESPOKE, the realistic benchmark for evaluating personalization in search-augmented LLMs. BESPOKE is designed to be both realistic, by collecting authentic chat and search histories directly from humans, and diagnostic, by pairing responses with fine-grained preference scores and feedback. The benchmark is constructed through long-term, deeply engaged human annotation, where human annotators contributed their own histories, authored queries with detailed information needs, and evaluated responses with scores and diagnostic feedback. Leveraging BESPOKE, we conduct systematic analyses that reveal key requirements for effective personalization in information-seeking tasks, providing a foundation for fine-grained evaluation of personalized search-augmented LLMs. Our code and data are available at https://augustinlib.github.io/BESPOKE/.

In the field of Human-Robot Interaction (HRI), a fundamental challenge is to facilitate human understanding of robots. The emerging domain of eXplainable HRI (XHRI) investigates methods to generate explanations and evaluate their impact on human-robot interactions. Previous works have highlighted the need to personalise the level of detail of these explanations to enhance usability and comprehension. Our paper presents a framework designed to update and retrieve user knowledge-memory models, allowing for adapting the explanations' level of detail while referencing previously acquired concepts. Three architectures based on our proposed framework that use Large Language Models (LLMs) are evaluated in two distinct scenarios: a hospital patrolling robot and a kitchen assistant robot. Experimental results demonstrate that a two-stage architecture, which first generates an explanation and then personalises it, is the framework architecture that effectively reduces the level of detail only when there is related user knowledge.

Sequential recommendation (SR), which encodes user activity to predict the next action, has emerged as a widely adopted strategy in developing commercial personalized recommendation systems. Although Transformer-based models have proven effective for sequential recommendation, the complexity of the self-attention module in Transformers scales quadratically with the sequence length. Controlling model complexity is essential for large-scale recommendation systems, as these systems may need to handle billion-scale vocabularies that evolve continuously, as well as user behavior sequences that can exceed tens of thousands in length. In this paper, we propose a novel multi-head latent Mamba architecture, which employs multiple low-dimensional Mamba layers and fully connected layers coupled with positional encoding to simultaneously capture historical and item information within each latent subspace. Our proposed method not only enables scaling up to large-scale parameters but also extends to multi-domain recommendation by integrating and fine-tuning LLMs. Through extensive experiments on public datasets, we demonstrate how Hydra effectively addresses the effectiveness-efficiency dilemma, outperforming state-of-the-art sequential recommendation baselines with significantly fewer parameters and reduced training time.

This paper introduces a neuro-symbolic approach for relational exploration in cultural heritage knowledge graphs, leveraging Large Language Models (LLMs) for explanation generation and a novel mathematical framework to quantify the interestingness of relationships. We demonstrate the importance of interestingness measure using a quantitative analysis, by highlighting its impact on the overall performance of our proposed system, particularly in terms of precision, recall, and F1-score. Using the Wikidata Cultural Heritage Linked Open Data (WCH-LOD) dataset, our approach yields a precision of 0.70, recall of 0.68, and an F1-score of 0.69, representing an improvement compared to graph-based (precision: 0.28, recall: 0.25, F1-score: 0.26) and knowledge-based baselines (precision: 0.45, recall: 0.42, F1-score: 0.43). Furthermore, our LLM-powered explanations exhibit better quality, reflected in BLEU (0.52), ROUGE-L (0.58), and METEOR (0.63) scores, all higher than the baseline approaches. We show a strong correlation (0.65) between interestingness measure and the quality of generated explanations, validating its effectiveness. The findings highlight the importance of LLMs and a mathematical formalization for interestingness in enhancing the effectiveness of relational exploration in cultural heritage knowledge graphs, with results that are measurable and testable. We further show that the system enables more effective exploration compared to purely knowledge-based and graph-based methods. Keywords Knowledge Graphs, Large Language Models (LLMs), Explainable AI (XAI), Cultural Heritage, Neuro-Symbolic AI, Interestingness Score, Contextual Relevance, Relational Search 1. Introduction The digitization of cultural heritage artifacts and historical records has generated a vast amount of knowledge encoded in the form of interconnected knowledge graphs (KGs) [1, 2]. Unlocking meaningful insights from these KGs requires more than simple keyword searches [3].

We study the predictability of online speech on social media, and whether predictability improves with information outside a user's own posts. Recent work suggests that the predictive information contained in posts written by a user's peers can surpass that of the user's own posts. Motivated by the success of large language models, we empirically test this hypothesis. We define unpredictability as a measure of the model's uncertainty, i.e., its negative log-likelihood on future tokens given context. As the basis of our study, we collect a corpus of 6.25M posts from more than five thousand X (previously Twitter) users and their peers. Across three large language models ranging in size from 1 billion to 70 billion parameters, we find that predicting a user's posts from their peers' posts performs poorly. Moreover, the value of the user's own posts for prediction is consistently higher than that of their peers'. Across the board, we find that the predictability of social media posts remains low, comparable to predicting financial news without context. We extend our investigation with a detailed analysis about the causes of unpredictability and the robustness of our findings. Specifically, we observe that a significant amount of predictive uncertainty comes from hashtags and @-mentions. Moreover, our results replicate if instead of prompting the model with additional context, we finetune on additional context.

Large Language Model or LLM inference has two phases, the prompt (or prefill) phase to output the first token and the extension (or decoding) phase to the generate subsequent tokens. In this work, we propose an efficient parallelization scheme, KV-Runahead to accelerate the prompt phase. The key observation is that the extension phase generates tokens faster than the prompt phase because of key-value cache (KV-cache). Hence, KV-Runahead parallelizes the prompt phase by orchestrating multiple processes to populate the KV-cache and minimizes the time-to-first-token (TTFT). Dual-purposing the KV-cache scheme has two main benefits. First, since KV-cache is designed to leverage the causal attention map, we minimize computation and computation automatically. Second, since it already exists for the extension phase, KV-Runahead is easy to implement. We further propose context-level load-balancing to handle uneven KV-cache generation (due to the causal attention) and to optimize TTFT. Compared with an existing parallelization scheme such as tensor or sequential parallelization where keys and values are locally generated and exchanged via all-gather collectives, our experimental results demonstrate that KV-Runahead can offer over 1.4x and 1.6x speedups for Llama 7B and Falcon 7B respectively.