Searcharxiv⌕ Search

arXiv subjects

Estrid He

Publications and source records attributed to Estrid He.

At least 19 recordsLinked to original sources

EvoST: Adapting Traffic Forecasters to Evolving Sensor Graphs

Web-based navigation and traffic information services depend on accurate forecasts as sensor networks evolve. On checkpoints reused years later, updating only shared weights or adding a low-rank node prompt leaves substantial degradation, whereas, in the evaluated architectures, refreshing node identities recovers most of it while the shared spatio-temporal weights are reused. We propose EvoST, a framework that maintains traffic forecasters by refreshing node identities while keeping shared weights fixed. EvoST updates existing identities using recently released labels and initializes new sensors with identities borrowed from behaviorally similar established sensors. Once enough delayed labels arrive, it jointly refits all identities, and a per-sensor recursive least-squares calibrator corrects forecasts as labels become available. Across twelve PeMS region-year settings, identity refresh reduces the relative one-hour MAE gap between five-year-old STAEFormer checkpoints and target-year models from 17.7% to 2.5%, with all forecasters calibrated, and its MAE is 0.7% above that of full-parameter fine-tuning on the same data. In Graph WaveNet, identity refresh recovers 88% of the five-year MAE gap to calibrated target-year models. Similar recovery holds for STAEFormer checkpoints up to fifteen years old, while refreshing STID recovers 67% of its five-year gap. On an annual benchmark covering nine EvoXXLTraffic districts and Transport for NSW, EvoST has the lowest average MAE, RMSE, and MAPE among the reported non-foundation methods on all ten networks, and calibration accounts for most of its Avg-MAE reduction from the matched frozen checkpoints. Code and the extended version: https://github.com/lixus7/EvoST

cs.LG↗

Decouple to Forget: Semantic Mediation for Sensitive Unlearning in Large Language Models

Machine unlearning for large language models seeks to remove specified knowledge while preserving general capability, yet current methods often treat an entire response as the unit of forgetting. This coarse granularity causes unnecessary utility loss when the forget target is only a small set of sensitive facts embedded within otherwise benign explanations. In this paper, we introduce Decouple to Forget (D2F), a targeted unlearning method inspired by semantic mediation that separates sensitive factual content from non-sensitive meaning and surface expression. For each example to be forgotten, D2F extracts a fact-level "mediator" representation from the answer, identifies which parts are sensitive, and removes only those parts while leaving the rest of the structure intact. We then generate counterfactual answers grounded in this mediator that preserve non-sensitive content but exclude sensitive facts, employing preference-based contrastive optimization to reduce the likelihood of information leakage while maintaining helpfulness and fluency. Experiments across various benchmark datasets show that D2F achieves a better balance between forgetting and retention than strong baseline methods. Our code is available at https://github.com/JingyueCong/D2F.

cs.CL↗

LatentRiskBench: Measuring State-Dependent Risks in Enterprise Agents

Enterprise agents face safety risks that cannot be determined from individual actions alone. Existing safety benchmarks largely evaluate action-local risks, overlooking failures whose safety depends on workflow state. In such state-dependent risks, the same action may be safe in one context but violate policy in another. Hence, we introduce LatentRiskBench for evaluating such state-dependent risks in enterprise settings. It pairs risky trajectories with benign counterparts that share the same terminal action but differ in latent policy-relevant states, requiring models to reconstruct workflow context rather than rely on action-level cues. We also introduce new metrics for state-dependent risk evaluation, including a suite of residual risk (RR) measures. Experiments on the benchmark show that action-level evaluation detects only 0.09 of state-dependent violations, whereas state-level evaluation reaches 0.96. Existing guardrail methods largely overlook such risks unless they are explicitly applied to workflow states. Case studies on a commercial deployed scenario further show that the safety metrics embedded in the benchmark provide a more realistic measure of safety for enterprise agents. Our dataset and code are available at https://github.com/JingyueCong/LatentRiskBench.

cs.CL↗

Do We Still Need Gazetteers in the Era of LLMs? Chaining Retrieval with a Spatial Neuro-Symbolic Index

Geographic information retrieval (GeoIR) tasks require systems to interpret ambiguous toponyms for downstream applications. Traditionally, toponym resolution relies on gazetteers to provide an explicit index of place entities and spatial relationships. Recently, gazetteer-free approaches seek to reduce dependence on handcrafted searches: dense retrieval utilizes text encoders to capture rich context, moving beyond the limitations of lexical search. However, text encoders implicitly assume that learned representations can function as reliable spatial-semantic indexes. In this paper, we evaluate this assumption through a spatial-semantic indexing setup: given a contextualized toponym mention, we retrieve the corresponding gazetteer entity represented by text derived from a gazetteer knowledge graph. We benchmark five frozen text encoders under two retrieval strategies: brute-force nearest-neighbor retrieval over entity representations, and a neuro-symbolic hierarchical beam search that constrains retrieval (i.e. chaining the search with gazetteer hierarchy). Experimental results reveal a distinct coarse-versus-fine trade-off. Unconstrained dense retrieval frequently incurs catastrophic spatial errors. Conversely, hierarchical constraints improve coarse geographic grounding, but still yield limited benefit for fine-grained localization metrics: vanilla text encoders fail to capture the fine-scale spatial fidelity encoded in gazetteers. Our code is publicly available at: https://doi.org/10.25439/rmt.31094269

cs.IR↗

The Commit-Abstain Circuit: Why Language Models Hallucinate Instead of Abstaining

Language models (LMs) often hallucinate by committing to confident answers rather than abstaining, even when they do not have enough information to answer reliably. A large body of existing work mitigates hallucination through detection or abstention mechanisms, but leaves open how models internally arrive at the decision to commit or abstain in the first place. We study this decision through mechanistic analysis, framing hallucination as unsupported commitment: the model commits despite exhibiting signals of unanswerability. Using causal gating, we identify a Commit-Abstain Circuit (CAC), a sparse, causally localised subset of attention heads and MLP sublayers underlying this decision. Across ten LMs (3B-14B) from five families and three benchmarks, the CAC exhibits a recurring accumulate-yet-undercorrect pattern: commitment-promoting components build up commitment in earlier layers, while abstention-promoting components act later as corrective signals that are often insufficient to overturn the accumulated commitment. Building on this finding, a lightweight policy trained on CAC activations improves decision accuracy by 12.2 points over the model's intrinsic commit-abstain margin, reduces false abstentions by 2.5 times, transfers to unseen benchmarks, and extends to larger models (27B-35B). The CAC is both diagnostic, clarifying how models overcommit, and practical, enabling improved abstention decisions.

cs.AI↗

I Am No One: Style-Aware Paraphrasing for Text Anonymization

Authorship attribution models can re-identify users from seemingly anonymized text by exploiting stable stylistic fingerprints, even after explicit identifiers are removed, posing a growing privacy risk for text publishing and analytics. This risk extends to speech-derived text such as ASR transcripts of meetings and call-center conversations, where stylometric leakage can persist even after acoustic anonymization. Differential privacy-based anonymization often severely degrades text quality and utility. We propose a style-aware, prompt-driven anonymization approach that uses pretrained large language models to construct compact stylistic profiles from minimal samples and rewrite text to suppress identifiable style markers while preserving meaning. Across blog and review datasets, our approach reduces authorship attribution F1 by 60-70% while maintaining content quality and readability, substantially outperforming DP-based and non-DP baselines.

cs.CL↗

GraphProfiler: Source-Linked Sensitive Attribute Inference via Personal Knowledge Graphs

Sensitive attributes such as age, income, and occupation can be inferred from user-generated content by aggregating indirect cues across many ordinary posts. LLM-based profilers can perform this aggregation automatically and with high accuracy, which makes large-scale personal attribute inference a major privacy threat. Existing LLM-based profilers, however, offer limited insight into which specific posts, concepts, and relationships made an inference possible, which is key to targeted privacy mitigation, i.e., redacting or rewriting only the few posts that actually leak an attribute, rather than perturbing entire histories. We introduce GraphProfiler, an auditable LLM-based profiler that represents each user's post history as a source-linked personal knowledge graph where nodes and edges trace back to the originating post and resolves attribute predictions to cited graph records and source texts. GraphProfiler reaches 86.7% attack success rate on the eight-attribute SynthPAI benchmark, within two points of strong text-only baselines, and 84.6% on PANDORA, while citing supporting evidence for over 98% of predictions. Our controlled ablation experiments provide evidence that the cited posts contribute to attack success, as removing them reduces the attack success rate substantially more than removing an equal number of random posts.

cs.CL↗

Twin Worlds: Equivariance-Based Abstention for Evidence-Grounded Reasoning

Knowledge-intensive reasoning requires Large Language Models (LLMs) to ground answers in provided evidence. When evidence is insufficient, it is desirable that models abstain rather than confidently generating unsupported answers. Existing abstention methods rely on uncertainty estimation or evidence sufficiency checks, but neither tests whether the reasoning process for generation, driven by the interaction of provided evidence and the model's internal memory parameters, is actually grounded in the evidence. A key contributing factor is that entity mentions in context activate memorised associations, causing models to generate plausible responses ungrounded in evidence. We propose Twin Worlds (TW), a framework for improving reliability in knowledge-intensive reasoning through equivariance-based abstention: unlike invariance, which requires outputs to remain unchanged, equivariance requires outputs to transform correspondingly under entity substitutions. A model grounded in the evidence should produce answers that shift consistently when entities are substituted while their relations are preserved. TW constructs multiple worlds via typed substitutions of the original input that preserve relational structure while reducing parametric priors, and uses equivariance violations as an abstention signal. Across four benchmarks and three model backbones, TW identifies when answers are not reliably grounded in the provided evidence and outperforms uncertainty- and sufficiency-based baselines.

cs.CL↗

Graph-Guided Selective Unlearning for Language Models: Controlling Support Routes Beyond Forget Seeds

Enterprises fine-tune language models on proprietary data that may later require removal due to privacy, contractual, or compliance obligations. Selective unlearning removes requested knowledge while preserving model utility, offering a practical alternative to full retraining, but existing methods treat the explicitly identified forget examples as the complete deletion scope. This is insufficient when target knowledge remains recoverable through paraphrases, aliases, or neighboring training examples. We propose GRAPHSU, a graph-guided controller that expands the deletion scope beyond forget seeds by constructing a weighted support-route graph, propagating deletion pressure through it, and applying graded forgetting strengths to high-risk neighbors. On the Task of Fictitious Unlearning (TOFU), a synthetic author-profile question-answering benchmark, and PISTOL, a structural-unlearning benchmark built around interconnected factual samples, with GPT-2 Medium and Llama-3.2-3B-Instruct, GRAPHSU achieves the lowest utility-feasible soft leakage across all deletion settings, reducing leakage by up to 49.5 percentage points over a matched seed-only baseline, demonstrating that effective enterprise unlearning requires controlling support routes, not just forget seeds.

cs.AI↗

Deep Contrastive Unlearning for Language Models

The past a few years have witnessed the great success of large language models, demonstrating powerful capabilities in comprehending textual data and generating human-like languages. Large language models achieve success by being trained on vast amounts of textual data, including online sources with copyrighted content and user-generated knowledge. However, this comes at a cost: the potential risk of exposing users' privacy and violating copyright protections. Thus, to safeguard individuals' "right to be forgotten", there has been increasing interests in machine unlearning -- the process of removing information carried by particular training samples from a model while not deteriorating its predictive quality. This is a challenging task due to the black-box nature of language models. Most existing studies focus on mitigating the impact of those forgot samples upon a model's outputs, and do not explicitly consider the geometric distributions of samples in the latent space of a model. To address this issue, we propose a machine unlearning framework, named Deep Contrastive Unlearning for fine-Tuning (DeepCUT) language models. Our proposed model achieves machine unlearning by directly optimizing the latent space of a model. Comprehensive experiments on real-world datasets demonstrate the effectiveness and efficiency of DeepCUT with consistent and significant improvement over baseline methods.

cs.CL↗

Position Bias Undermines Preference Consistency in Listwise LLM-Based Reranking

Large language models (LLMs) have emerged as promising listwise rerankers for recommender systems, but their reliability under equivalent candidate permutations remains unclear. Since recommendation candidates form an unordered set, a reranker should not depend on the arbitrary order used to serialize them. However, decoder-only LLM rerankers can allow input order to affect model scores, pairwise preferences, and rankings. We study how position bias affects the ranking process induced by LLM-based rerankers. Instead of measuring only changes in final ranked lists, we treat rankings produced under equivalent candidate permutations as observations of an induced preference system. We introduce an evaluation framework measuring pairwise preference instability, global preference inconsistency, and listwise output consistency. This framework characterizes candidate-order sensitivity at the pairwise, global, and output levels. Experiments across multiple LLMs, datasets, and list lengths show that these consistency measures are closely aligned, but can diverge from recommendation effectiveness and marginal position-exposure bias. Improving relevance or flattening exposure across positions does not necessarily restore stable pairwise preferences, globally coherent preference structures, or consistent ranked outputs. These results show that reducing marginal exposure skew is insufficient to establish ranking-function validity in LLM-based reranking. Code is available at https://github.com/ejbito/InvariRank .

cs.IR↗

StrideDiffusion: Accelerating Diffusion Models for Time-series Generation

Diffusion models have become competitive generators for time series, but their practical use is limited by the large number of sequential denoising steps required at inference time. Existing fast samplers typically use fixed or generic timestep schedules, overlooking a distinctive property of time-series diffusion: different spectral bands evolve at different rates during the reverse process. We introduce StrideDiffusion, a training-free spectral-aware sampler that adaptively selects the denoising stride from band-level activity. At each step, StrideDiffusion monitors relative band energy, log-power drift, and phase velocity to identify whether high- frequency dynamics remain active or whether the trajectory is dominated by stable low-frequency structure. It then takes fine steps when rapidly varying bands are active and larger jumps once only coarse components remain. A bandwise stability analysis shows that inactive frequency bands change only linearly with the jump size under deterministic affine reverse updates, providing a local justification for spectral activity as a step-size indicator. Across six unconditional time-series generation benchmarks, StrideDiffusion uses only 14-66 function evaluations instead of 500/1000 denoising steps, achieving up to 18.9x wall-clock speedup while preserving or improving generation quality. On conditional imputation and forecasting, it further delivers 5-14x average acceleration with comparable predictive accuracy. These results show that spectral evolution provides a practical and principled signal for fast time-series diffusion sampling. Our code is available at https://anonymous.4open.science/r/stridediff-ts.

cs.AI↗

One Pass, Any Order: Position-Invariant Listwise Reranking for LLM-Based Recommendation

Large language models (LLMs) are increasingly used for recommendation reranking, but their listwise predictions can depend on the order in which candidates are presented. This creates a mismatch between the set-based nature of recommendation and the sequence-based computation of decoder-only LLMs, where permuting an otherwise identical candidate set can change item scores and final rankings. Such order sensitivity makes LLM-based rerankers difficult to rely on, since rankings may reflect prompt serialization rather than user preference. We propose InvariRank, a permutation-invariant listwise reranking framework that addresses this dependence at the architectural level. InvariRank blocks cross-candidate attention with a structured attention mask and negates position-induced scoring changes through shared positional framing under Rotary Positional Embeddings (RoPE). Combined with a listwise learning-to-rank objective, the model scores all candidates in a single forward pass, avoiding permutation-based invariance training objectives that require multiple permutations of a candidate set. Experiments on recommendation benchmarks show that InvariRank maintains competitive ranking effectiveness while producing stable rankings across candidate permutations. The results suggest that architectural invariance is a practical route to reliable and efficient LLM-based recommendation reranking. The source code is at https://github.com/ejbito/InvariRank.

cs.IR↗

Hallucinate Less by Thinking More: Aspect-Based Causal Abstention for Large Language Models

Large Language Models (LLMs) often produce fluent but factually incorrect responses, a phenomenon known as hallucination. Abstention, where the model chooses not to answer and instead outputs phrases such as "I don't know", is a common safeguard. However, existing abstention methods typically rely on post-generation signals, such as generation variations or feedback, which limits their ability to prevent unreliable responses in advance. In this paper, we introduce Aspect-Based Causal Abstention (ABCA), a new framework that enables early abstention by analysing the internal diversity of LLM knowledge through causal inference. This diversity reflects the multifaceted nature of parametric knowledge acquired from various sources, representing diverse aspects such as disciplines, legal contexts, or temporal frames. ABCA estimates causal effects conditioned on these aspects to assess the reliability of knowledge relevant to a given query. Based on these estimates, we enable two types of abstention: Type-1, where aspect effects are inconsistent (knowledge conflict), and Type-2, where aspect effects consistently support abstention (knowledge insufficiency). Experiments on standard benchmarks demonstrate that ABCA improves abstention reliability, achieves state-of-the-art performance, and enhances the interpretability of abstention decisions.

cs.CL↗

Identifying Origins of Place Names via Retrieval Augmented Generation

Who is the "Batman" behind "Batman Street" in Melbourne? Understanding the historical, cultural, and societal narratives behind place names can reveal the rich context that has shaped a community. Although place names serve as essential spatial references in gazetteers, they often lack information about place name origins. Enriching these place names in today's gazetteers is a time-consuming, manual process that requires extensive exploration of a vast archive of documents and text sources. Recent advances in natural language processing and language models (LMs) hold the promise of significant automation of identifying place name origins due to their powerful capability to exploit the semantics of the stored documents. This chapter presents a retrieval augmented generation pipeline designed to search for place name origins over a broad knowledge base, DBpedia. Given a spatial query, our approach first extracts sub-graphs that may contain knowledge relevant to the query; then ranks the extracted sub-graphs to generate the final answer to the query using fine-tuned LM-based models (i.e., ColBERTv2 and Llama2). Our results highlight the key challenges facing automated retrieval of place name origins, especially the tendency of language models to under-use the spatial information contained in texts as a discriminating factor. Our approach also frames the wider implications for geographic information retrieval using retrieval augmented generation.

cs.IR↗

Evaluating Position Bias in Large Language Model Recommendations

Large Language Models (LLMs) are being increasingly explored as general-purpose tools for recommendation tasks, enabling zero-shot and instruction-following capabilities without the need for task-specific training. While the research community is enthusiastically embracing LLMs, there are important caveats to directly adapting them for recommendation tasks. In this paper, we show that LLM-based recommendation models suffer from position bias, where the order of candidate items in a prompt can disproportionately influence the recommendations produced by LLMs. First, we analyse the position bias of LLM-based recommendations on real-world datasets, where results uncover systemic biases of LLMs with high sensitivity to input orders. Furthermore, we introduce a new prompting strategy to mitigate the position bias of LLM recommendation models called Ranking via Iterative SElection (RISE). We compare our proposed method against various baselines on key benchmark datasets. Experiment results show that our method reduces sensitivity to input ordering and improves stability without requiring model fine-tuning or post-processing.

cs.IR↗

Agent-Based Detection and Resolution of Incompleteness and Ambiguity in Interactions with Large Language Models

Many of us now treat LLMs as modern-day oracles asking it almost any kind of question. However, consulting an LLM does not have to be a single turn activity. But long multi-turn interactions can get tedious if it is simply to clarify contextual information that can be arrived at through reasoning. In this paper, we examine the use of agent-based architecture to bolster LLM-based Question-Answering systems with additional reasoning capabilities. We examine the automatic resolution of potential incompleteness or ambiguities in questions by transducers implemented using LLM-based agents. We focus on several benchmark datasets that are known to contain questions with these deficiencies to varying degrees. We equip different LLMs (GPT-3.5-Turbo and Llama-4-Scout) with agents that act as specialists in detecting and resolving deficiencies of incompleteness and ambiguity. The agents are implemented as zero-shot ReAct agents. Rather than producing an answer in a single step, the model now decides between 3 actions a) classify b) resolve c) answer. Action a) decides if the question is incomplete, ambiguous, or normal. Action b) determines if any deficiencies identified can be resolved. Action c) answers the resolved form of the question. We compare the use of LLMs with and without the use of agents with these components. Our results show benefits of agents with transducer 1) A shortening of the length of interactions with human 2) An improvement in the answer quality and 3) Explainable resolution of deficiencies in the question. On the negative side we find while it may result in additional LLM invocations and in some cases, increased latency. But on tested datasets, the benefits outweigh the costs except when questions already have sufficient context. Suggesting the agent-based approach could be a useful mechanism to harness the power of LLMs to develop more robust QA systems.

cs.AI↗

An Empirical Study of the Role of Incompleteness and Ambiguity in Interactions with Large Language Models

Natural language as a medium for human-computer interaction has long been anticipated, has been undergoing a sea-change with the advent of Large Language Models (LLMs) with startling capacities for processing and generating language. Many of us now treat LLMs as modern-day oracles, asking it almost any kind of question. Unlike its Delphic predecessor, consulting an LLM does not have to be a single-turn activity (ask a question, receive an answer, leave); and -- also unlike the Pythia -- it is widely acknowledged that answers from LLMs can be improved with additional context. In this paper, we aim to study when we need multi-turn interactions with LLMs to successfully get a question answered; or conclude that a question is unanswerable. We present a neural symbolic framework that models the interactions between human and LLM agents. Through the proposed framework, we define incompleteness and ambiguity in the questions as properties deducible from the messages exchanged in the interaction, and provide results from benchmark problems, in which the answer-correctness is shown to depend on whether or not questions demonstrate the presence of incompleteness or ambiguity (according to the properties we identify). Our results show multi-turn interactions are usually required for datasets which have a high proportion of incompleteness or ambiguous questions; and that that increasing interaction length has the effect of reducing incompleteness or ambiguity. The results also suggest that our measures of incompleteness and ambiguity can be useful tools for characterising interactions with an LLM on question-answeringproblems

cs.CL↗