arXiv subjects

Viet Nguyen

Publications and source records attributed to Viet Nguyen.

25 records · Page 2Linked to original sources

arXiv2024

On Inference Stability for Diffusion Models

Denoising Probabilistic Models (DPMs) represent an emerging domain of generative models that excel in generating diverse and high-quality images. However, most current training methods for DPMs often neglect the correlation between timesteps, limiting the model's performance in generating images effectively. Notably, we theoretically point out that this issue can be caused by the cumulative estimation gap between the predicted and the actual trajectory. To minimize that gap, we propose a novel \textit{sequence-aware} loss that aims to reduce the estimation gap to enhance the sampling quality. Furthermore, we theoretically show that our proposed loss function is a tighter upper bound of the estimation loss in comparison with the conventional loss in DPMs. Experimental results on several benchmark datasets including CIFAR10, CelebA, and CelebA-HQ consistently show a remarkable improvement of our proposed method regarding the image generalization quality measured by FID and Inception Score compared to several DPM baselines. Our code and pre-trained checkpoints are available at \url{https://github.com/VinAIResearch/SA-DPM}.

cs.CV↗

arXiv2023

A dual-space classification scheme of spatial heterogeneities in 2D monatomic supercooled liquids

Understanding the physics of supercooled liquids near glassy transition remains one of the major challenges in condensed matter science. There has been long recognized that supercooled liquids have spatially dynamical heterogeneity whose dynamics in some regions of the sample could potentially be orders of magnitude faster than the dynamics in other regions only a few nanometers away. However, to identify such domain structures both structurally and configurationally in a consistent fashion and the connection between structures and dynamics remains elusive. A new approach to classify these spatial heterogeneities of supercooled liquids is developed. Using average weighted coordination numbers (WCNs) of particles as features for the Principle Component Analysis (PCA) and Gaussian Mixture (GM) clustering, a representation of the feature space to perform GM clustering is constructed after applying PCA to WCNs. Nano-domains or aggregated clusters are found in configurational (real) space by direct mapping the identities of clusters in the feature (structural) space with some classification uncertainties. These uncertainties can be improved by a co-learning strategy to transmit information between the structural space and configurational space iteratively until convergence. The domains are indeed found to be remarkably consistent in both spaces over long times as well as measured to have heterogeneous dynamics. Using such a classification scheme the order parameter time correlation function follows the non-conserved order parameter scaling law, which indicates that such domains are the result of liquid-liquid phase separation after quenching.

cond-mat.stat-mech↗

arXiv2023

Automated characterization of spatial and dynamical heterogeneity in supercooled liquids via implementation of Machine Learning

A computational approach by an implementation of the Principle Component Analysis (PCA) with K-means and Gaussian Mixture (GM) clustering methods from Machine Learning (ML) algorithms to identify structural and dynamical heterogeneities of supercooled liquids is developed. In this method, a collection of the average weighted coordination numbers ($\overline{WCNs}$) of particles calculated from particles' positions are used as an order parameter to build a low-dimensional representation of feature (structural) space for K-means clustering to sort the particles in the system into few meso-states using PCA. Nano-domains or aggregated clusters are also formed in configurational (real) space from a direct mapping using associated meso-states' particle identities with some misclassified interfacial particles. These classification uncertainties can be improved by a co-learning strategy which utilizes the probabilistic GM clustering and the information transfer between the structural space and configurational space iteratively until convergence. A final classification of meso-states in structural space and domains in configurational space are stable over long times and measured to have dynamical heterogeneities. Armed with such a classification protocol, various studies over the thermodynamic and dynamical properties of these domains indicate that the observed heterogeneity is the result of liquid-liquid phase separation after quenching to a supercooled state.

cond-mat.stat-mech↗

arXiv2022

A general classification scheme of detecting spatial and dynamical heterogeneities in super-cooled liquids

A computational approach via implementation of the Principle Component Analysis (PCA) and Gaussian Mixture (GM) clustering methods from Machine Learning (ML) algorithms to identify domain structures of supercooled liquids is developed. Raw features data are collected from the coordination numbers of particles smoothed using its radial distribution function and are used as an order-parameter of disordered structures for GM clustering after dimensionality reduction from the PCA. To transfer the knowledge from features(structural) space to configurational space, another GM clustering is performed using the Cartesian coordinates as an order-parameter with the particles' identity from GM in the feature space. Both GM clustering are performed iteratively until convergence. Results show the appearance of aggregated clusters of nano-domains over sufficient long timescale both in structural and configurational spaces with heterogeneous dynamics. More importantly, consistent nano-domains tilling up the whole space regardless of the system size are observed and our approach can be applied to any disordered systems.

cond-mat.stat-mech↗

arXiv2021

Improvement in Machine Translation with Generative Adversarial Networks

In this paper, we explore machine translation improvement via Generative Adversarial Network (GAN) architecture. We take inspiration from RelGAN, a model for text generation, and NMT-GAN, an adversarial machine translation model, to implement a model that learns to transform awkward, non-fluent English sentences to fluent ones, while only being trained on monolingual corpora. We utilize a parameter $λ$ to control the amount of deviation from the input sentence, i.e. a trade-off between keeping the original tokens and modifying it to be more fluent. Our results improved upon phrase-based machine translation in some cases. Especially, GAN with a transformer generator shows some promising results. We suggests some directions for future works to build upon this proof-of-concept.

cs.CL↗

arXiv2021

Human Imperceptible Attacks and Applications to Improve Fairness

Modern neural networks are able to perform at least as well as humans in numerous tasks involving object classification and image generation. However, small perturbations which are imperceptible to humans may significantly degrade the performance of well-trained deep neural networks. We provide a Distributionally Robust Optimization (DRO) framework which integrates human-based image quality assessment methods to design optimal attacks that are imperceptible to humans but significantly damaging to deep neural networks. Through extensive experiments, we show that our attack algorithm generates better-quality (less perceptible to humans) attacks than other state-of-the-art human imperceptible attack methods. Moreover, we demonstrate that DRO training using our optimally designed human imperceptible attacks can improve group fairness in image classification. Towards the end, we provide an algorithmic implementation to speed up DRO training significantly, which could be of independent interest.

cs.CV↗

arXiv2021

Randomized Exploration for Reinforcement Learning with General Value Function Approximation

We propose a model-free reinforcement learning algorithm inspired by the popular randomized least squares value iteration (RLSVI) algorithm as well as the optimism principle. Unlike existing upper-confidence-bound (UCB) based approaches, which are often computationally intractable, our algorithm drives exploration by simply perturbing the training data with judiciously chosen i.i.d. scalar noises. To attain optimistic value function estimation without resorting to a UCB-style bonus, we introduce an optimistic reward sampling procedure. When the value functions can be represented by a function class $\mathcal{F}$, our algorithm achieves a worst-case regret bound of $\widetilde{O}(\mathrm{poly}(d_EH)\sqrt{T})$ where $T$ is the time elapsed, $H$ is the planning horizon and $d_E$ is the $\textit{eluder dimension}$ of $\mathcal{F}$. In the linear setting, our algorithm reduces to LSVI-PHE, a variant of RLSVI, that enjoys an $\widetilde{\mathcal{O}}(\sqrt{d^3H^3T})$ regret. We complement the theory with an empirical evaluation across known difficult exploration tasks.

cs.LG↗