Retrieval-Augmented Generation (RAG) is becoming increasingly essential for Question Answering (QA) in the financial sector, where accurate and contextually grounded insights from complex public disclosures are crucial. However, existing financial RAG systems face two significant challenges: (1) they struggle to process heterogeneous data formats, such as text, tables, and figures; and (2) they encounter difficulties in balancing general-domain applicability with company-specific adaptation. To overcome these challenges, we present VeritasFi, an innovative hybrid RAG framework that incorporates a multi-modal preprocessing pipeline alongside a cutting-edge two-stage training strategy for its re-ranking component. VeritasFi enhances financial QA through three key innovations: (1) A multi-modal preprocessing pipeline that seamlessly transforms heterogeneous data into a coherent, machine-readable format. (2) A tripartite hybrid retrieval engine that operates in parallel, combining deep multi-path retrieval over a semantically indexed document corpus, real-time data acquisition through tool utilization, and an expert-curated memory bank for high-frequency questions, ensuring comprehensive scope, accuracy, and efficiency. (3) A two-stage training strategy for the document re-ranker, which initially constructs a general, domain-specific model using anonymized data, followed by rapid fine-tuning on company-specific data for targeted applications. By integrating our proposed designs, VeritasFi presents a novel framework that greatly enhances the adaptability and robustness of financial RAG systems, providing a scalable solution for both general-domain and company-specific QA tasks. Code accompanying this work is available at https://github.com/simplew4y/VeritasFi.git.

Conversational generative artificial intelligence (GenAI) has emerged as a promising tool for second language (L2) speaking practice, but the mechanisms behind its effectiveness remain underexplored. This study aims to explore these mechanisms through the lens of the cognitive-motivational model of achievement emotions in the control-value theory. Specifically, we investigate how emotions (enjoyment, boredom, and curiosity), cognitive processing, and the ideal L2 self interact to influence speaking performance. The sample consisted of 158 Chinese L2 majors engaging in GenAI-assisted speaking practice in informal contexts. Using Partial Least Square-Structural Equation Modeling (PLS-SEM) with Smart PLS 4 software, key findings include that gender did not affect the constructs under study, but GenAI competence positively influenced speaking performance. Enjoyment had a direct effect on cognitive processing, which in turn enhanced speaking performance, though it did not influence the ideal L2 self. Curiosity positively influenced the ideal L2 self and speaking performance, but had no effect on cognitive processing. Boredom, however, did not affect either cognitive processing or the ideal L2 self. The study contributes theoretically by advancing understanding of how psychological factors shape L2 speaking performance in GenAI contexts. Pedagogically, it offers insights into optimizing GenAI for language practice.

Blockchain systems encompass many distinct and autonomous entities, each utilizing its own self-contained identity authentication algorithm. Unlike identity authentication within a singular blockchain, cross-chain scenarios demand special attention due to their pivotal role in enabling the acknowledgment of users’ identities across diverse domains. This capability is the foundational prerequisite for the circulation of resources across different chains. Consequently, the central challenge for cross-chain systems lies in establishing mutual recognition and trust in users’ digital identities. This paper proposes a Multi-User Proxy Re-Signature (MU-PRS) algorithm, facilitating the cross-chain conversion of signatures from multiple users. Concurrently, This paper propose the Multi-Notary Signature Conversion (MN-SC) mechanism, designed to address the challenge posed by disparate system mechanisms across blockchains during cross-chain authentication. Leveraging the MU-PRS algorithm and MN-SC mechanism, we present a Multi-User Cross-Chain Authentication Scheme (MU-CCAS) within a heterogeneous blockchain environment. This scheme enables the verification of identities of multiple cross-chain users through a single signature verification. This innovative approach not only addresses the centralization issues inherent in third-party cross-chain authentication but also significantly enhances the efficiency of identity authentication. The evaluation results demonstrate MU-CCAS’s superior security over existing solutions in three dimensions: BAN logic, Scyther verification, and security attribute analysis. Additionally, it establishes that MU-PRS and MU-CCAS have low computational overhead, easy implementation, and excel in algorithm, scheme, and cross-chain performance. Overall, our work provides a robust and efficient framework for cross-chain authentication, addressing centralization challenges and enhancing digital security.

Autoencoders and their variations provide unsupervised models for learning low-dimensional representations for downstream tasks. Without proper regularization, autoencoder models are susceptible to the overfitting problem and the so-called posterior collapse phenomenon. In this paper, we introduce a quantization-based regularizer in the bottleneck stage of autoencoder models to learn meaningful latent representations. We combine both perspectives of Vector Quantized-Variational AutoEncoders (VQ-VAE) and classical denoising regularization methods of neural networks. We interpret quantizers as regularizers that constrain latent representations while fostering a similarity-preserving mapping at the encoder. Before quantization, we impose noise on the latent codes and use a Bayesian estimator to optimize the quantizer-based representation. The introduced bottleneck Bayesian estimator outputs the posterior mean of the centroids to the decoder, and thus, is performing soft quantization of the noisy latent codes. We show that our proposed regularization method results in improved latent representations for both supervised learning and clustering downstream tasks when compared to autoencoders using other bottleneck structures.

This paper considers the problem of compression for similarity identification. Unlike classic compression problems, the focus is not on reconstructing the original data. Instead, compression is determined by the reliability of answering given queries. The problem is characterized by the identification rate of a source which is the minimum compression rate which allows reliable answers for a given similarity threshold. In this work, we investigate the component-based quadratic similarity identification for multivariate Gaussian sources. The decorrelated original data is processed by a distinct D- admissible system for each component. For a special case, we characterize the component-based identification rate for a correlated Gaussian source. Furthermore, we derived the optimal bit allocation for a given total rate constraint.

This paper considers the problem of compression for similarity queries and discusses tree-structured vector quantizers. Here, the focus is on the trade of between the rate of the compressed data and the reliability of the answers to a given query. This problem is different from classical quantization as there is no need to reconstruct the original data. Instead, compression is determined by the reliability of answering given queries. We consider compression schemes that do not allow false negatives when answering queries. Hence, classical vector quantization needs to be modified. We propose quantizers that hierarchically cluster the data into sphere-shaped quantization cells. The query process will be guided by decision rules that avoid false negatives. In particular, we discuss two classic clustering methods, namely k-means and k-center. We use P{maybe}, a probability that is related to the occurrence of false positives, and the computational cost of queries to assess our scheme. Our experiments show that k-center clustering generally performs better than k-means clustering, while tree-structured clustering reduces the computational cost of queries for both methods.

The scoring function is a central component in mobile visual search. In this paper, we propose an embedded 3D geometry score for mobile 3D visual search (M3DVS). In contrast to conventional mobile visual search, M3DVS uses not only the visual appearance of query objects, but utilizes also the underlying 3D geometry. The proposed scoring function interprets visual search as a process that reduces uncertainty among candidate objects when observing a query. For M3DVS, the uncertainty is reduced by both appearance-based visual similarity and 3D geometric similarity. For the latter, we give an algorithm for estimating the query-dependent threshold for geometric similarity. In contrast to visual similarity, the threshold for geometric similarity is relative due to the constraints of image-based 3D reconstruction. The experimental results show that the embedded 3D geometry score improves the recall-data rate performance when compared to a conventional visual score or 3D geometry-based re-ranking.

This paper proposes multi-view vocabulary trees for mobile 3D visual search. We generate hierarchically structured multi-view features and construct a multi-view vocabulary tree from the multi-view images. As the 3D geometry information is incorporated in the multi-view vocabulary tree, it allows us to design an algorithm for fast 3D geometric verification at low computational complexity. With that, we devise an iterative algorithm that accomplishes jointly matching and geometric verification. The experimental results show that our joint approach to matching and verification improves the recall-data rate performance as well as the subjective ranking results for mobile 3D visual search.

This paper proposes geometry-based ranking for mobile 3D visual search. It utilizes the underlying geometry of the 3D objects as well as the appearance to improve the ranking results. A double hierarchy has been embedded in the data structure, namely the hierarchically structured multi-view features for each object and a tree hierarchy from multi-view vocabulary trees. As the 3D geometry information is incorporated in the multi-view vocabulary tree, it allows us to evaluate the consistency of the 3D geometry at low computational complexity. Thus, a cost function is proposed for object ranking using geometric consistency. With that, we devise an iterative algorithm that accomplishes 3D geometry-based ranking. The experimental results show that our 3D geometry-based ranking improves the recall-datarate performance as well as the subjective ranking results for mobile 3D visual search.