Anxiao (Andrew) Jiang’s research while affiliated with Texas A&M University and other places

... Transformer-based models, Generative Adversarial Networks (GAN) variants (Goodfellow, Pouget-Abadie, Mirza, Xu, Warde-Farley, Ozair, Courville and Bengio, 2014), and diffusion-based techniques (Yang, Zhang, Song, Hong, Xu, Zhao, Zhang, Cui and Yang, 2023) can further integrate text logs, biometric signals, or mobile data, creating multifaceted synthetic records. Similar methods have been used in healthcare (e.g., blending clinical notes and diagnostic images) (Moor, Banerjee, Abad, Krumholz, Leskovec, Topol and Rajpurkar, 2023;Giuffrè and Shung, 2023) and finance (e.g., simulating trader actions) (Zuo, Jiang and Zhou, 2024), enabling research free of confidentiality breaches. By aligning generation strategies with the unique traits of regional betting cultures, researchers can reveal subtle risk patterns that standardized models might miss, opening opportunities for more focused and relevant interventions. ...

Reference:

Multimodal Generative AI and Foundation Models for Behavioural Health in Online Gambling

... In 2021, Yu et al. [17] proposed a novel method that incorporates template-based QG model with a sequence-to-sequence model for diversity-aware QG. They did not apply stringent templates, instead they used adjustable patterns that can be collected effectively with less cost. ...

Reference:

Using Syntax and Shallow Semantic Analysis for Vietnamese Question Generation

... The finding of an optimal MOI of 1 has important guiding significance for the dose design of future phage preparations. However, it should be noted that the optimal MOI obtained under laboratory conditions may differ from actual application environments, so more factors need to be considered when developing actual treatment plans, such as the physiological state of the host animal and the method of administration [23][24][25]. ...

Reference:

Characterization and Potential Application of Phage vB_PmuM_CFP3 for Phage Therapy Against Avian Pasteurella multocida

... These benchmarks and datasets are often adapted from real-life applications, with many containing domain-specific knowledge that may not generalize effectively to unseen SQL domains. Hence, largescale cross-domain datasets featuring professional SQL queries, such as Squall (Shi et al., 2020), Spider (Yu et al., 2018a), Spider-Syn (Gan et al., 2021), WikiSQL (Zhong et al., 2017), and SparC (Yu et al., 2020), have been introduced to facilitate comprehensive method analyses. In retrospect, we realize two concurrent works which perform systematical benchmarking on text-to-SQL methods. ...

Reference:

DB-GPT-Hub: Towards Open Benchmarking Text-to-SQL Empowered by Large Language Models

... While average-case robustness is more suited for applications such as malware detection, worst-case robustness is relevant in critical applications such as neuromorphic computing. It was recently shown in Raviv et al. (2020) that worst-case robustness is impossible even against one bit erasure (i.e., setting x i = 0 for some i), unless redundancy is added, and a simple methods of adding such redundancy was given. ...

Reference:

Enhancing Robustness of Neural Networks through Fourier Stabilization

... Conventional fault tolerance methods, such as Error Correction Codes (ECCs) and Triple Modular Redundancy (TMR), often impose significant overheads, undermining the advantages of approximate computing [25], [26]. A comprehensive list of such techniques for AccDNN fault detection and mitigation can be found in [27], [28]. ...

Reference:

EPSILON: Adaptive Fault Mitigation in Approximate Deep Neural Network using Statistical Signatures

... The issue of using error-correcting codes to improve the efficiency of RTCSs is considered in detail in [16,17]. The studies report the results of experimental tests, which confirm that the algorithms developed by the authors demonstrate significantly higher performance in the context of functional error correction, compared to conventional approaches. ...

Reference:

Improving the process of control and correction of errors in non-positional code structures

... The work in [54] also extend the analog computing architectures to support dynamic precision with redundant coding by repeating operations and averaging the result. The protection of the weights and bias of neural network from noise using linear and nonlinear analog error correction codes in order to prevent performance degradation has been proposed in [55], [56]. The works also explored the use of unequal error protection method for weights at different layers of a binarized network due to the uneven effect of noise in different layers. ...

Reference:

Effect of Batch Normalization on Noise Resistant Property of Deep Learning Models

... An effective way to solve the above problems is to fully exploit the natural redundancy in the source and protocol stack, so as to improve the forward error correction ability of the received data. Source natural redundancy [11][12][13] refers to the redundancy law that remains when the source is not compressed or compressed incompletely, and protocol natural redundancy refers to the redundancy law that is inevitably introduced into the protocol fields of each layer in the independent design of the network layer. Although it is artificially designed, the purpose is not to improve the communication reliability. ...

Reference:

Separating Noisy TCP/IP Session Flows Based on Natural Redundancy Decoding

... In recent works (including results from the authors of this work), machine learning and algorithmic techniques have been used to exploit NR to correct errors in data [21], [22], [23], [27], [33], [48], [49], [53], [54]. This work studies the Representation-Oblivious scheme for the first time, and also presents new theoretical analysis for the Representation-Aware scheme. ...

Reference:

Machine Learning for Error Correction with Natural Redundancy