About
Hi, my name is Tue. I am currently an AI Research Resident at the Quantum AI & Cyber Security Institute (QACI) at FPT Corporation, where I am advised by Dr. Nghi Bui. I received my B.S. from the Hanoi University of Science and Technology (HUST), where I was advised by Dr. Linh Ngo and Prof. Trung Le.
In my free time I draw and play strategy games. I make no claims about being good at either =)).
You can reach me at tue209003@gmail.com. I'd love to hear from you!
Research Interests
My research focuses on reasoning and alignment in large language models. I study how to improve reasoning generalization and efficiency, guide the reasoning process, and align model behavior with intended goals. More broadly, I am interested in learning, optimization, and evaluation. I enjoy exploring related directions and am particularly drawn to intuitive ideas that offer fresh perspectives on familiar problems.
Recently
- Sep '26 I am applying to PhD programs this cycle. If our interests overlap, I would love to hear from you.
- Jan '26 Started a research collaboration with Mila – Quebec AI Institute on secure code generation.
- Jul '25 Graduated from Hanoi University of Science and Technology with a B.S. in Computer Science.
- May '25 Joined the Quantum AI & Cyber Security Institute (QACI), FPT Corporation as an AI Research Resident.
Publications
Preprints
* denotes equal contribution. Click a title to read the abstract.
Publications
* denotes equal contribution. Click a title to read the abstract.
-
Neural topic models have achieved notable advances in generating coherent topics and learning document-topic distributions, yet they commonly rely on multiple loss components that vary widely in scale. These imbalances make joint optimization difficult and often require extensive hyperparameter tuning. Multi-objective optimization (MOO) methods offer a promising direction, but their use has largely been limited to settings with shared-parameter architectures, restricting their applicability to many topic modeling frameworks. Moreover, applying standard MOO techniques directly to neural topic models can lead to degraded performance due to incompatibilities between the objectives. This work introduces a new strategy for integrating MOO into neural topic modeling without relying on hard parameter sharing. The proposed method enables effective coordination of multiple loss functions and facilitates stable optimization. Experiments on widely used benchmark corpora show that the approach yields substantial improvements over baseline neural topic models and consistently surpasses naïve MOO-based variants.
-
Knowledge distillation (KD) is widely recognized as an effective approach for compressing large language models (LLMs). However, standard KD methods often falter when confronted with architectural or tokenization heterogeneity between teacher and student models, which creates a mismatch in their representations. While Optimal Transport (OT) provides a promising solution to align these representations, most OT-based methods rely on a single cost function, which isn't enough to capture the multifaceted discrepancies between models with distinct designs. To address this limitation, we introduce Multi-Cost Wasserstein Knowledge Distillation (MCW-KD), a novel framework that enhances KD by simultaneously optimizing several cost functions within a unified OT formulation. MCW-KD employs specific cost matrices to effectively align both the final hidden states and the output distributions of the models. We also provide a rigorous theoretical foundation for the proposed Multi-Cost Wasserstein Distance, ensuring both mathematical validity and computational ability. Extensive experiments on instruction-following datasets demonstrate that MCW-KD significantly improves student model performance compared to state-of-the-art KD baselines, especially when teacher and student models have different tokenizers.
-
Leveraging the power of Large Language Models (LLMs) through preference optimization is crucial for aligning model outputs with human values. Direct Preference Optimization (DPO) has recently emerged as a simple yet effective method by directly optimizing on preference data without the need for explicit reward models. However, DPO typically relies on human-labeled preference data, which can limit its scalability. Self-Play Fine-Tuning (SPIN) addresses this by allowing models to generate their own rejected samples, reducing the dependence on human annotations. Nevertheless, SPIN uniformly applies learning signals across all tokens, ignoring the fine-grained quality variations within responses. As the model improves, rejected samples increasingly contain high-quality tokens, making the uniform treatment of tokens suboptimal. In this paper, we propose SWIFT (Self-Play Weighted Fine-Tuning), a fine-grained self-refinement method that assigns token-level importance weights estimated from a stronger teacher model. Beyond alignment, we also demonstrate that SWIFT serves as an effective knowledge distillation strategy by using the teacher not for logits matching, but for reward-guided token weighting. Extensive experiments on diverse benchmarks and settings demonstrate that SWIFT consistently surpasses both existing alignment approaches and conventional knowledge distillation methods.
-
Neural topic modeling has substantially improved topic quality and document topic distribution compared to traditional probabilistic methods. These models often incorporate multiple loss functions. However, the disparate magnitudes of these losses can make hyperparameter tuning for these loss functions challenging, potentially creating obstacles for simultaneous optimization. While gradient-based Multi-objective Optimization (MOO) algorithms offer a potential solution, they are typically applied to shared parameters in multi-task learning, hindering their broader adoption, particularly in Neural Topic Models (NTMs). Furthermore, our experiments reveal that naïve MOO applications on NTMs can yield suboptimal results, even underperforming compared to implementations without the MOO mechanism. This paper proposes a novel approach to integrate MOO algorithms, independent of hard-parameter sharing architectures, and effectively optimizes multiple NTMs loss functions. Comprehensive evaluations on widely used benchmark datasets demonstrate that our approach significantly enhances baseline topic model performance and outperforms direct MOO applications on NTMs.
-
Recent advances in neural topic models (NTMs) have improved topic quality but still face challenges: weak document-topic alignment, high inference costs due to large pretrained language models (PLMs), and limited modeling of hierarchical topic structures. To address these issues, we introduce HiCOT (Hierarchical Clustering and Contrastive Learning with Optimal Transport for Neural Topic Modeling), a novel framework that enhances topic coherence and efficiency. HiCOT integrates Optimal Transport to refine document-topic relationships using compact PLM-based embeddings, captures semantic structure of the documents. Additionally, it employs hierarchical clustering combine with contrastive learning to disentangle topic-word and topic-topic relationships, ensuring clearer structure and better coherence. Experimental results on multiple benchmark datasets demonstrate HiCOT's superior effectiveness over existing NTMs in topic coherence, topic performance, representation quality, and computational efficiency.
-
Recent advanced frameworks in topic models have significantly enhanced the performance compared to conventional probabilistic approaches. Such models, mostly constructed from neural network architecture together with other advanced techniques such as contextual embedding, optimal transport distance and pre-trained language model, etc. have effectively improved the topic quality and document topic distribution. Despite the improvements, these methods lack considerations of effective optimization for complex objective functions that contain log-likelihood and additional regularization terms. In this study, we propose to apply an efficient optimization method to improve the generalization and performance of topic models. Our approach explicitly considers the sharpness of the loss landscape during optimization, which forces the optimizer to choose directions in the parameter space that lead to flatter minima, in which the models are typically more stable and robust to small perturbations in the data. Additionally, we propose an effective strategy to select the flatness region for parameter optimization by leveraging the optimal transport distance between doc-topic distributions and doc-cluster proportions, which can effectively enhance document representation. Experimental results on popular benchmark datasets demonstrate that our method effectively improves the performance of baseline topic models.
Preprints
-
Large language models (LLMs) have achieved strong performance on mathematical, logical, and scientific reasoning tasks. However, this progress is commonly measured on benchmarks that remain close to the distributions encountered during training. Small changes in notation or problem structure, as well as deeper reasoning or new domains, can lead to substantial performance degradation, raising a central question: when and how can LLMs generalize their reasoning beyond their training distributions? This paper presents a comprehensive survey of reasoning generalization in LLMs. We organize existing work into four complementary pillars: (1) training methods that encourage transferable reasoning, (2) inference strategies that adapt computation at test time, (3) architectural designs that support reusable operations, and (4) analyses that characterize when and why generalization succeeds or fails. Building on this taxonomy, we compare the assumptions, evaluation settings, and generalization claims of representative approaches. We further summarize the main empirical findings and theoretical results, and discuss the evaluation practices and open challenges that shape this emerging research direction.
-
As coding models and agent harnesses become more capable, reliable security verification is becoming a central bottleneck. Improving secure code generation requires a signal whose reported progress remains aligned with actual program behavior under optimization; otherwise, an optimizer can improve the signal's verdict without removing the underlying vulnerability. Existing model-based approaches require models to either infer program behavior from source code or specify correct outcomes for adversarial tests, placing substantial reasoning demands on the security signal. We introduce FALCON, a counterexample-grounded verification framework that separates attack exploration from behavioral verification and connects them through execution traces of the program under test. In test-time repair, FALCON yields substantially stronger held-out functional-security gains than alternative security signals across secure code generation benchmarks, while preserving or improving functionality. As an RL reward, FALCON also outperforms static-analysis and learned rewards on held-out functional-security, even when all policies successfully optimize their respective training rewards. Combining FALCON with either proxy reward further improves held-out performance, suggesting that execution-grounded feedback can mitigate reward overoptimization while preserving useful guidance from existing security signals.
-
Existing benchmarks for AI coding agents focus on isolated, single-issue tasks such as fixing a bug or adding a small feature. However, real-world software engineering is a long-horizon endeavor: developers interpret high-level requirements, coordinate changes across many files, and evolve codebases over multiple iterations while preserving functionality. We introduce SWE-EVO, a benchmark for this long-horizon software evolution challenge. Constructed from release notes of seven mature open-source Python projects, SWE-EVO comprises 48 tasks requiring multi-step modifications spanning an average of 21 files, validated against test suites averaging 874 tests per instance. Experiments reveal a striking capability gap: GPT-5.4 with OpenHands achieves only 25% on SWE-EVO versus 72.80% achieved by GPT-5.2 on SWE-Bench Verified, showing that current agents struggle with sustained, multi-file reasoning. We also propose Fix Rate, a metric capturing partial progress on these complex, long-horizon tasks.
Experience
- 2026 — now Research Collaborator, Mila – Quebec AI Institute · Remote Training and inference methods for the security, safety, and reliability of coding agents.
- 2025 — now AI Research Resident, Quantum AI & Cyber Security Institute (QACI), FPT Corporation · Hanoi Training and evaluation pipelines for reasoning and coding models: post-training recipes, benchmark design, and the measurement work needed to tell a real capability gain from a benchmark artifact.
- 2024 – 25 AI Engineer, VinBigdata Institute · Hanoi Multimodal models for speech–language understanding and generation, focused on audio–text alignment.
- 2023 – 25 Research Student, Data Science Lab, BKAI – HUST · Hanoi Optimization algorithms for training stability and generalization across optimization tasks.
Education
- 2021 – 25 B.S. Computer Science, Hanoi University of Science and Technology Talented Program · GPA 3.4 / 4.0
- 2018 – 21 Ha Tinh High School for Gifted Students Mathematics Honors Class · 9.4 / 10
Honors
- 2021 Vietnam IMO Team Selection Test Selected among the top 50 mathematics students nationally through the Vietnamese Mathematical Olympiad, and competed for a place on the national IMO team.
- 2020 – 21 Vietnamese Mathematical Olympiad Second Prize, 2021 · Third Prize, 2020
- 2020 Odon Vallet Scholarship Awarded by the Odon Vallet Fund to the best students of the year.