Loc Nguyen

Vinh Luong, Sang Dinh, Shruti Raghavan et al.

Large Language Models (LLMs) have shown remarkable capabilities, but their inherent probabilistic nature often leads to inconsistency and inaccuracy in complex problem-solving tasks. This paper introduces DANA (Domain-Aware Neurosymbolic Agent), an architecture that addresses these issues by integrating domain-specific knowledge with neurosymbolic approaches. We begin by analyzing current AI architectures, including AutoGPT, LangChain ReAct and OpenAI's ChatGPT, through a neurosymbolic lens, highlighting how their reliance on probabilistic inference contributes to inconsistent outputs. In response, DANA captures and applies domain expertise in both natural-language and symbolic forms, enabling more deterministic and reliable problem-solving behaviors. We implement a variant of DANA using Hierarchical Task Plans (HTPs) in the open-source OpenSSA framework. This implementation achieves over 90\% accuracy on the FinanceBench financial-analysis benchmark, significantly outperforming current LLM-based systems in both consistency and accuracy. Application of DANA in physical industries such as semiconductor shows that its flexible architecture for incorporating knowledge is effective in mitigating the probabilistic limitations of LLMs and has potential in tackling complex, real-world problems that require reliability and precision.

Navaraj Neupane, Loc Nguyen

We study an inverse initial-data problem for a nonlinear Schrödinger equation in which the initial wave field is reconstructed from lateral measurements. Our approach combines a Legendre-polynomial-exponential-time dimensional reduction with a Carleman-based contraction principle. First, we expand the solution in a weighted Legendre basis in time and truncate the expansion to obtain a coupled nonlinear elliptic system for the spatial coefficients. Next, we solve this reduced system by constructing a contraction map on a suitable admissible set. This contraction map admits a unique fixed point, which is the limit of the corresponding Picard iteration. We also establish a stability estimate showing that this fixed point remains close to the exact reduced solution in the noisy-data case. Finally, we present numerical experiments in two space dimensions for several different geometries and nonlinear exponents. The numerical results show that the proposed method accurately reconstructs the main features of the initial wave field and remains stable even when the boundary data contain noise.

Trong-Thang Pham, Loc Nguyen, Anh Nguyen et al.

Generative diffusion models are increasingly used for medical imaging data augmentation, but text prompting cannot produce causal training data. Re-prompting rerolls the entire generation trajectory, altering anatomy, texture, and background. Inversion-based editing methods introduce reconstruction error that causes structural drift. We propose MedSteer, a training-free activation-steering framework for endoscopic synthesis. MedSteer identifies a pathology vector for each contrastive prompt pair in the cross-attention layers of a diffusion transformer. At inference time, it steers image activations along this vector, generating counterfactual pairs from scratch where the only difference is the steered concept. All other structure is preserved by construction. We evaluate MedSteer across three experiments on Kvasir v3 and HyperKvasir. On counterfactual generation across three clinical concept pairs, MedSteer achieves flip rates of 0.800, 0.925, and 0.950, outperforming the best inversion-based baseline in both concept flip rate and structural preservation. On dye disentanglement, MedSteer achieves 75% dye removal against 20% (PnP) and 10% (h-Edit). On downstream polyp detection, augmenting with MedSteer counterfactual pairs achieves ViT AUC of 0.9755 versus 0.9083 for quantity-matched re-prompting, confirming that counterfactual structure drives the gain. Code is at link https://github.com/phamtrongthang123/medsteer

Ali A. Amer, Loc Nguyen

Collaborative filtering (CF) is an important approach for recommendation system which is widely used in a great number of aspects of our life, heavily in the online-based commercial systems. One popular algorithms in CF is the K-nearest neighbors (KNN) algorithm, in which the similarity measures are used to determine nearest neighbors of a user, and thus to quantify the dependency degree between the relative user/item pair. Consequently, CF approach is not just sensitive to the similarity measure, yet it is completely contingent on selection of that measure. While Jaccard - as one of those commonly used similarity measures for CF tasks - concerns the existence of ratings, other numerical measures such as cosine and Pearson concern the magnitude of ratings. Particularly speaking, Jaccard is not a dominant measure, but it is long proven to be an important factor to improve any measure. Therefore, in our continuous efforts to find the most effective similarity measures for CF, this research focuses on proposing new similarity measure via combining Jaccard with several numerical measures. The combined measures would take the advantages of both existence and magnitude. Experimental results on, Movie-lens dataset, showed that the combined measures are preeminent outperforming all single measures over the considered evaluation metrics.