Tu Nguyen

Xuhai Xu, Jun Gong, Carolina Brum et al.

We present a framework for gesture customization requiring minimal examples from users, all without degrading the performance of existing gesture sets. To achieve this, we first deployed a large-scale study (N=500+) to collect data and train an accelerometer-gyroscope recognition model with a cross-user accuracy of 95.7% and a false-positive rate of 0.6 per hour when tested on everyday non-gesture data. Next, we design a few-shot learning framework which derives a lightweight model from our pre-trained model, enabling knowledge transfer without performance degradation. We validate our approach through a user study (N=20) examining on-device customization from 12 new gestures, resulting in an average accuracy of 55.3%, 83.1%, and 87.2% on using one, three, or five shots when adding a new gesture, while maintaining the same recognition accuracy and false-positive rate from the pre-existing gesture set. We further evaluate the usability of our real-time implementation with a user experience study (N=20). Our results highlight the effectiveness, learnability, and usability of our customization framework. Our approach paves the way for a future where users are no longer bound to pre-existing gestures, freeing them to creatively introduce new gestures tailored to their preferences and abilities.

Tu Nguyen, Rasul Tutunov, Xiaotong Ji et al.

A recurring pattern in "reasoning without training" is that base LLMs already assign non-trivial probability mass to correct multi-step solutions; the bottleneck is locating these modes efficiently at inference time. Power sampling provides a principled way to bias decoding toward such modes by targeting p_theta(x)^alpha with alpha > 1, but practical approximations must account for future-dependent correction factors that determine which prefixes remain promising. We introduce Auxiliary Particle Power Sampling (APPS), a blockwise particle algorithm for approximating the sequence-level power target with a bounded population of partial solutions. APPS propagates hypotheses in parallel using proposal-corrected power reweighting and refines their survival through future-value-guided selection at resampling boundaries. This redistributes finite compute across competing prefixes rather than committing to a single unfolding path, while providing a direct scaling knob in the particle count and predictable peak memory. We instantiate the future-value signal with short-horizon rollouts and also study an amortized variant that replaces rollouts with a lightweight learned selection head. Across reasoning benchmarks, APPS improves the accuracy-runtime trade-off of training-free decoding and suggests that part of the gap to post-trained systems can be recovered through more faithful inference-time power approximation.

Bingning Huang, Tu Nguyen, Matthieu Zimmer

Recent advances in reasoning with large language models (LLMs) have shown the effectiveness of Monte Carlo Tree Search (MCTS) for generating high-quality intermediate trajectories, particularly in math and symbolic domains. Inspired by this, we explore how MCTS-derived trajectories-traditionally used for training value or reward models-can be repurposed to improve policy optimization in preference-based reinforcement learning (RL). Specifically, we focus on Group Relative Policy Optimization (GRPO), a recent algorithm that enables preference-consistent policy learning without value networks. We reframe GRPO into a staged training paradigm, leveraging a teacher's MCTS rollouts to construct a tree-structured curriculum of prefixes. This introduces the novel challenge of computing advantages for training samples that originate from different prefixes, each with a distinct expected return. To address this, we propose Staged Advantage Estimation (SAE), a framework for computing low-variance, prefix-aware advantages by projecting rewards onto a constraint set that respects the tree's hierarchy. Our empirical results on mathematical reasoning tasks show that SAE improves final accuracy over standard GRPO. This outcome is grounded in our theoretical analysis, which confirms that SAE reduces gradient variance-a principled path to improved sample efficiency. We demonstrate this through practical SAE implementations, comparing efficient heuristics against a formal quadratic program.

Amirali Sajadi, Tu Nguyen, Kostadin Damevski et al.

Vulnerability detection tools are widely adopted in software projects, yet they often overwhelm maintainers with false positives and non-actionable reports. Automated exploitation systems can help validate these reports; however, existing approaches typically operate in isolation from detection pipelines, failing to leverage readily available metadata such as vulnerability type and source-code location. In this paper, we investigate how reported security vulnerabilities can be assessed in a realistic grey-box exploitation setting that leverages minimal vulnerability metadata, specifically a CWE classification and a vulnerable code location. We introduce Agentic eXploit Engine (AXE), a multi-agent framework for Web application exploitation that maps lightweight detection metadata to concrete exploits through decoupled planning, code exploration, and dynamic execution feedback. Evaluated on the CVE-Bench dataset, AXE achieves a 30% exploitation success rate, a 3x improvement over state-of-the-art black-box baselines. Even in a single-agent configuration, grey-box metadata yields a 1.75x performance gain. Systematic error analysis shows that most failed attempts arise from specific reasoning gaps, including misinterpreted vulnerability semantics and unmet execution preconditions. For successful exploits, AXE produces actionable, reproducible proof-of-concept artifacts, demonstrating its utility in streamlining Web vulnerability triage and remediation. We further evaluate AXE's generalizability through a case study on a recent real-world vulnerability not included in CVE-Bench.

Matthieu Zimmer, Xiaotong Ji, Tu Nguyen et al.

We introduce a novel approach to large language model (LLM) distillation by formulating it as a constrained reinforcement learning problem. While recent work has begun exploring the integration of task-specific rewards into distillation processes, existing methods typically rely on ad-hoc reward weighting. We propose a principled optimization framework that maximizes task-specific rewards while constraining the divergence from the teacher model to remain below a specified threshold. Our approach adapts constrained state augmented reinforcement learning to the distillation setting, introducing a modified reward function that maintains theoretical guarantees of constraint satisfaction without requiring state augmentation or teacher model access during deployment and without the computational overhead of the dual Lagrangian methods. Through extensive experiments on mathematical reasoning tasks, we demonstrate that our method achieves better constraint satisfaction rates and better reasoning compared to the soft Lagrangian relaxation baselines while maintaining competitive task performance. Our framework provides a theoretically grounded and practically efficient solution for reward-aware distillation in resource-constrained settings.

Tu Nguyen, Kevin Du, Alexander Miserlis Hoyle et al.

Two central capabilities of language models (LMs) are: (i) drawing on prior knowledge about entities, which allows them to answer queries such as "What's the official language of Austria?", and (ii) adapting to new information provided in context, e.g., "Pretend the official language of Austria is Tagalog.", that is pre-pended to the question. In this article, we introduce targeted persuasion score (TPS), designed to quantify how persuasive a given context is to an LM where persuasion is operationalized as the ability of the context to alter the LM's answer to the question. In contrast to evaluating persuasiveness only by inspecting the greedily decoded answer under the model, TPS provides a more fine-grained view of model behavior. Based on the Wasserstein distance, TPS measures how much a context shifts a model's original answer distribution toward a target distribution. Empirically, through a series of experiments, we show that TPS captures a more nuanced notion of persuasiveness than previously proposed metrics.

Tu Nguyen, Nedim Šrndić, Alexander Neth

Tactics, Techniques and Procedures (TTPs) represent sophisticated attack patterns in the cybersecurity domain, described encyclopedically in textual knowledge bases. Identifying TTPs in cybersecurity writing, often called TTP mapping, is an important and challenging task. Conventional learning approaches often target the problem in the classical multi-class or multilabel classification setting. This setting hinders the learning ability of the model due to a large number of classes (i.e., TTPs), the inevitable skewness of the label distribution and the complex hierarchical structure of the label space. We formulate the problem in a different learning paradigm, where the assignment of a text to a TTP label is decided by the direct semantic similarity between the two, thus reducing the complexity of competing solely over the large labeling space. To that end, we propose a neural matching architecture with an effective sampling-based learn-to-compare mechanism, facilitating the learning process of the matching model despite constrained resources.

Thi Huyen Nguyen, Tu Nguyen, Tuan-Anh Hoang et al.

For their attractiveness, comprehensiveness and dynamic coverage of relevant topics, community-based question answering sites such as Stack Overflow heavily rely on the engagement of their communities: Questions on new technologies, technology features as well as technology versions come up and have to be answered as technology evolves (and as community members gather experience with it). At the same time, other questions cease in importance over time, finally becoming irrelevant to users. Beyond filtering low-quality questions, "forgetting" questions, which have become redundant, is an important step for keeping the Stack Overflow content concise and useful. In this work, we study this managed forgetting task for Stack Overflow. Our work is based on data from more than a decade (2008 - 2019) - covering 18.1M questions, that are made publicly available by the site itself. For establishing a deeper understanding, we first analyze and characterize the set of questions about to be forgotten, i.e., questions that get a considerable number of views in the current period but become unattractive in the near future. Subsequently, we examine the capability of a wide range of features in predicting such forgotten questions in different categories. We find some categories in which those questions are more predictable. We also discover that the text-based features are surprisingly not helpful in this prediction task, while the meta information is much more predictive.

Tu Nguyen

This extended abstract describes our solution for the Traffic4Cast Challenge 2019. The task requires modeling both fine-grained (pixel-level) and coarse (region-level) spatial structure while preserving temporal relationships across long sequences. Building on Conv-LSTM ideas, we introduce a tile-aware, cascaded-memory Conv-LSTM augmented with cross-frame additive attention and a memory-flexible training scheme: frames are sampled per spatial tile so the model learns tile-local dynamics and per-tile memory cells can be updated sparsely, paged, or compressed to scale to large maps. We provide a compact theoretical analysis (tight softmax/attention Lipschitz bound and a tiling error lower bound) explaining stability and the memory-accuracy tradeoffs, and empirically demonstrate improved scalability and competitive forecasting performance on large-scale traffic heatmaps.

Tu Nguyen, Tuan Tran, Wolfgang Nejdl

Measuring entity relatedness is a fundamental task for many natural language processing and information retrieval applications. Prior work often studies entity relatedness in static settings and an unsupervised manner. However, entities in real-world are often involved in many different relationships, consequently entity-relations are very dynamic over time. In this work, we propose a neural networkbased approach for dynamic entity relatedness, leveraging the collective attention as supervision. Our model is capable of learning rich and different entity representations in a joint framework. Through extensive experiments on large-scale datasets, we demonstrate that our method achieves better results than competitive baselines.

Tu Nguyen, Markus Rokicki

With continuous glucose monitoring (CGM), data-driven models on blood glucose prediction have been shown to be effective in related work. However, such (CGM) systems are not always available, e.g., for a patient at home. In this work, we conduct a study on 9 patients and examine the online predictability of data-driven (aka. machine learning) based models on patient-level blood glucose prediction; with measurements are taken only periodically (i.e., after several hours). To this end, we propose several post-prediction methods to account for the noise nature of these data, that marginally improves the performance of the end system.

Tu Nguyen, Nattiya Kanhabua, Wolfgang Nejdl

Entity aspect recommendation is an emerging task in semantic search that helps users discover serendipitous and prominent information with respect to an entity, of which salience (e.g., popularity) is the most important factor in previous work. However, entity aspects are temporally dynamic and often driven by events happening over time. For such cases, aspect suggestion based solely on salience features can give unsatisfactory results, for two reasons. First, salience is often accumulated over a long time period and does not account for recency. Second, many aspects related to an event entity are strongly time-dependent. In this paper, we study the task of temporal aspect recommendation for a given entity, which aims at recommending the most relevant aspects and takes into account time in order to improve search experience. We propose a novel event-centric ensemble ranking method that learns from multiple time and type-dependent models and dynamically trades off salience and recency characteristics. Through extensive experiments on real-world query logs, we demonstrate that our method is robust and achieves better effectiveness than competitive baselines.

Tu Nguyen

Recent work have done a good job in modeling rumors and detecting them over microblog streams. However, the performance of their automatic approaches are not relatively high when looking early in the diffusion. A first intuition is that, at early stage, most of the aggregated rumor features (e.g., propagation features) are not mature and distinctive enough. The objective of rumor debunking in microblogs, however, are to detect these misinformation as early as possible. In this work, we leverage neural models in learning the hidden representations of individual rumor-related tweets at the very beginning of a rumor. Our extensive experiments show that the resulting signal improves our classification performance over time, significantly within the first 10 hours. To deepen the understanding of these low and high-level features in contributing to the model performance over time, we conduct an extensive study on a wide range of high impact rumor features for the 48 hours range. The end model that engages these features are shown to be competitive, reaches over 90% accuracy and out-performs strong baselines in our carefully cured dataset.

Tu Nguyen, Cheng Li, Claudia Niederée

Recently a lot of progress has been made in rumor modeling and rumor detection for micro-blogging streams. However, existing automated methods do not perform very well for early rumor detection, which is crucial in many settings, e.g., in crisis situations. One reason for this is that aggregated rumor features such as propagation features, which work well on the long run, are - due to their accumulating characteristic - not very helpful in the early phase of a rumor. In this work, we present an approach for early rumor detection, which leverages Convolutional Neural Networks for learning the hidden representations of individual rumor-related tweets to gain insights on the credibility of each tweets. We then aggregate the predictions from the very beginning of a rumor to obtain the overall event credits (so-called wisdom), and finally combine it with a time series based rumor classification model. Our extensive experiments show a clearly improved classification performance within the critical very first hours of a rumor. For a better understanding, we also conduct an extensive feature evaluation that emphasized on the early stage and shows that the low-level credibility has best predictability at all phases of the rumor lifetime.