Searcharxiv⌕ Search

arXiv subjects

Anh Tuan Bui

Publications and source records attributed to Anh Tuan Bui.

11 recordsLinked to original sources
arXiv2026

Adaptive Subspace Projection for Generative Personalization

Generative personalization often suffers from the semantic collapsing problem (SCP), where a learned personalized concept overpowers the rest of the text prompt, causing the model to ignore important contextual details. To address this, we first analyze the underlying cause, revealing that the semantic drift responsible for SCP is not random but is concentrated within a specific low-dimensional subspace. We also discover that the personalization process perturbs the embedding of the original base concept, making it an unstable reference point. Based on these insights, we introduce Test-time Embedding Adjustment with Adaptive Subspace Projection (AdaptSP), a training-free method that uses the stable, pre-trained embedding as an anchor. AdaptSP isolates the semantic drift and projects it onto the identified subspace, performing a precise adjustment that mitigates SCP while maintaining the subject identity. Our experiments show that this targeted approach significantly improves prompt fidelity and contextual alignment.

cs.CV↗
arXiv2026

Antibody: Strengthening Defense Against Harmful Fine-Tuning for Large Language Models via Attenuating Harmful Gradient Influence

Fine-tuning-as-a-service introduces a threat to Large Language Models' safety when service providers fine-tune their models on poisoned user-submitted datasets, a process known as harmful fine-tuning attacks. In this work, we show that by regularizing the gradient contribution of harmful samples encountered during fine-tuning, we can effectively mitigate the impact of harmful fine-tuning attacks. To this end, we introduce Antibody, a defense strategy that first ensures robust safety alignment for the model before fine-tuning, and then applies a safety-preservation learning algorithm during fine-tuning. Specifically, in the alignment stage before fine-tuning, we propose optimizing the model to be in a flat loss region with respect to harmful samples, which makes the safety alignment more resilient to subsequent harmful fine-tuning. Then, in the fine-tuning stage, we design a fine-tuning algorithm that applies a weighting scheme to all samples in each training batch to inhibit the model from learning from harmful samples while encouraging learning from benign samples. Experimental results demonstrate that Antibody successfully mitigates harmful fine-tuning attacks while boosting fine-tuning performance on the user-submitted dataset.

cs.LG↗
arXiv2025

Conditional Multidimensional Scaling with Incomplete Conditioning Data

Conditional multidimensional scaling seeks for a low-dimensional configuration from pairwise dissimilarities, in the presence of other known features. By taking advantage of available data of the known features, conditional multidimensional scaling improves the estimation quality of the low-dimensional configuration and simplifies knowledge discovery tasks. However, existing conditional multidimensional scaling methods require full data of the known features, which may not be always attainable due to time, cost, and other constraints. This paper proposes a conditional multidimensional scaling method that can learn the low-dimensional configuration when there are missing values in the known features. The method can also impute the missing values, which provides additional insights of the problem. Computer codes of this method are maintained in the cml R package on CRAN.

stat.ML↗
arXiv2024

Connective Viewpoints of Signal-to-Noise Diffusion Models

Diffusion models (DM) have become fundamental components of generative models, excelling across various domains such as image creation, audio generation, and complex data interpolation. Signal-to-Noise diffusion models constitute a diverse family covering most state-of-the-art diffusion models. While there have been several attempts to study Signal-to-Noise (S2N) diffusion models from various perspectives, there remains a need for a comprehensive study connecting different viewpoints and exploring new perspectives. In this study, we offer a comprehensive perspective on noise schedulers, examining their role through the lens of the signal-to-noise ratio (SNR) and its connections to information theory. Building upon this framework, we have developed a generalized backward equation to enhance the performance of the inference process.

cs.CV↗
arXiv2023

Dimension Reduction with Prior Information for Knowledge Discovery

This paper addresses the problem of mapping high-dimensional data to a low-dimensional space, in the presence of other known features. This problem is ubiquitous in science and engineering as there are often controllable/measurable features in most applications. To solve this problem, this paper proposes a broad class of methods, which is referred to as conditional multidimensional scaling (MDS). An algorithm for optimizing the objective function of conditional MDS is also developed. The convergence of this algorithm is proven under mild assumptions. Conditional MDS is illustrated with kinship terms, facial expressions, textile fabrics, car-brand perception, and cylinder machining examples. These examples demonstrate the advantages of conditional MDS over conventional dimension reduction in improving the estimation quality of the reduced-dimension space and simplifying visualization and knowledge discovery tasks. Computer codes for this work are available in the open-source cml R package.

stat.ML↗
arXiv2023

Optimal Transport Model Distributional Robustness

Distributional robustness is a promising framework for training deep learning models that are less vulnerable to adversarial examples and data distribution shifts. Previous works have mainly focused on exploiting distributional robustness in the data space. In this work, we explore an optimal transport-based distributional robustness framework in model spaces. Specifically, we examine a model distribution within a Wasserstein ball centered on a given model distribution that maximizes the loss. We have developed theories that enable us to learn the optimal robust center model distribution. Interestingly, our developed theories allow us to flexibly incorporate the concept of sharpness awareness into training, whether it's a single model, ensemble models, or Bayesian Neural Networks, by considering specific forms of the center model distribution. These forms include a Dirac delta distribution over a single model, a uniform distribution over several models, and a general Bayesian Neural Network. Furthermore, we demonstrate that Sharpness-Aware Minimization (SAM) is a specific case of our framework when using a Dirac delta distribution over a single model, while our framework can be seen as a probabilistic extension of SAM. To validate the effectiveness of our framework in the aforementioned settings, we conducted extensive experiments, and the results reveal remarkable improvements compared to the baselines.

cs.LG↗
arXiv2022

Multiple Perturbation Attack: Attack Pixelwise Under Different $\ell_p$-norms For Better Adversarial Performance

Adversarial machine learning has been both a major concern and a hot topic recently, especially with the ubiquitous use of deep neural networks in the current landscape. Adversarial attacks and defenses are usually likened to a cat-and-mouse game in which defenders and attackers evolve over the time. On one hand, the goal is to develop strong and robust deep networks that are resistant to malicious actors. On the other hand, in order to achieve that, we need to devise even stronger adversarial attacks to challenge these defense models. Most of existing attacks employs a single $\ell_p$ distance (commonly, $p\in\{1,2,\infty\}$) to define the concept of closeness and performs steepest gradient ascent w.r.t. this $p$-norm to update all pixels in an adversarial example in the same way. These $\ell_p$ attacks each has its own pros and cons; and there is no single attack that can successfully break through defense models that are robust against multiple $\ell_p$ norms simultaneously. Motivated by these observations, we come up with a natural approach: combining various $\ell_p$ gradient projections on a pixel level to achieve a joint adversarial perturbation. Specifically, we learn how to perturb each pixel to maximize the attack performance, while maintaining the overall visual imperceptibility of adversarial examples. Finally, through various experiments with standardized benchmarks, we show that our method outperforms most current strong attacks across state-of-the-art defense mechanisms, while retaining its ability to remain clean visually.

cs.CV↗
arXiv2019

The mod 2 cohomology rings of congruence subgroups in the Bianchi groups

We provide new tools for the calculation of the torsion in the cohomology of congruence subgroups in the Bianchi groups : An algorithm for finding particularly useful fundamental domains, and an analysis of the equivariant spectral sequence combined with torsion subcomplex reduction.\_\_\_\_\_\_\_\_\_\_With an appendix by Bui Anh Tuan and Sebastian Sch{ö}nnenbeck

math.KT↗
arXiv2018

The Farrell--Tate and Bredon homology for PSL\_4(Z) via cell subdivisions

We provide some new computations of Farrell--Tate and Bredon (co)homology for arithmetic groups. For calculations of Farrell--Tate or Bredon homology, one needs cell complexes wherecell stabilizers fix their cells pointwise. We provide two algorithms computing an efficient subdivision of a complex to achieve this rigidity property. Applying these algorithms to available cell complexes for PSL4(Z) provides computations of Farrell--Tate cohomology for small primes as well as the Bredon homology for the classifying spaces of proper actions with coefficients in the complex representation ring.

math.KT↗
arXiv2017

A monitoring and diagnostic approach for stochastic textured surfaces

We develop a supervised-learning-based approach for monitoring and diagnosing texture-related defects in manufactured products characterized by stochastic textured surfaces that satisfy the locality and stationarity properties of Markov random fields. Examples of stochastic textured surface data include images of woven textiles; image or surface metrology data for machined, cast, or formed metal parts; microscopy images of material microstructure samples; etc. To characterize the complex spatial statistical dependencies of in-control samples of the stochastic textured surface, we use rather generic supervised learning methods, which provide an implicit characterization of the joint distribution of the surface texture. We propose two spatial moving statistics, which are computed from residual errors of the fitted supervised learning model, for monitoring and diagnosing local aberrations in the general spatial statistical behavior of newly manufactured stochastic textured surface samples in a statistical process control context. We illustrate the approach using images of textile fabric samples and simulated 2-D stochastic processes, for which the algorithm successfully detects local defects of various natures. Supplemental discussions, results, data and computer codes are available online.

stat.AP↗