Hanoi Open University

Large Language Models (LLMs) fine-tuning techniques not only improve the adaptability to diverse downstream tasks, but also mitigate adverse effects of model quantization. Despite this, conventional quantization suffers from its structural limitation that hinders flexibility during the fine-tuning and deployment stages. Practical on-device tasks demand different quantization precisions (i.e. different bit-widths), e.g., understanding tasks tend to exhibit higher tolerance to reduced precision compared to generation tasks. Conventional quantization, typically relying on scaling factors that are incompatible across bit-widths, fails to support the on-device switching of precisions when confronted with complex real‑world scenarios. To overcome the dilemma, we propose OTARo, a novel method that enables on-device LLMs to flexibly switch quantization precisions while maintaining performance robustness through once fine-tuning. OTARo introduces Shared Exponent Floating Point (SEFP), a distinct quantization mechanism, to produce different bit-widths through simple mantissa truncations of a single model. Moreover, to achieve bit-width robustness in downstream applications, OTARo performs a learning process toward losses induced by different bit-widths. The method involves two critical strategies: (1) Exploitation-Exploration Bit-Width Path Search (BPS), which iteratively updates the search path via a designed scoring mechanism; (2) Low-Precision Asynchronous Accumulation (LAA), which performs asynchronous gradient accumulations and delayed updates under low bit-widths. Experiments on popular LLMs, e.g., LLaMA3.2-1B, LLaMA3-8B, demonstrate that OTARo achieves consistently strong and robust performance for all precisions.

This study delves into the experiences of older workers involved in the hospitality industry, investigating the significance of previous, current and future knowledge in equipping their hospitality careers. The conceptual notions of the knowledge-based view of the firm were appraised. Semi-structured open-ended interviews conducted among 56 older workers involved in both managerial and non-managerial positions highlighted the magnitude of 11 conceptual dimensions. Among these, the organisational, human resource, and self-development uplift dimensions help explain how previous knowledge has equipped participants and their firms. Similarly, the organisational momentum or career-enlightening dimensions elucidate current knowledge in fortifying earlier accumulated knowledge, while future required knowledge can be understood by the individual-organisational and generational-team bridging dimensions. The analysis also reveals alignments between the crafted dimensions and the knowledge-based model. A developed framework resulting from the compendium of dimensions, together with the identified alignments, underscores implications for hospitality employees, their firms, industry and policymakers. • Examines past, current, and future knowledge among older hospitality employees. • Gathers the viewpoints of experienced hospitality managers and non-managers. • Uses qualitative content analysis and data structure as analysis tools. • Uncovers 11 conceptual dimensions as the basis for a proposed framework. • Proposes theoretical and practical implications for practitioners and researchers.

The affordable housing crisis and its impact on a growing population is one of the most significant political and policy issues in the US and globally. The Financialization of Human Shelter examines the factors contributing to the sharp rise in rental housing costs in the US, the financialization of housing, and the consequences for tenants. Based on a public sociology research project conducted in the Sunbelt city of Jacksonville, Florida, authors Jaffee and Renzi delve into the financialization of housing in the rental housing market involving the acquisition by corporate entities of single-family and multifamily properties that are then bundled into portfolios as asset class investment opportunities for wealthy investors. Analyzing how this business model driving this trend is organized to extract the maximum amount of income from the properties and provide a healthy return to investors through the raising of rents, adding of superfluous fees, and minimizing of maintenance costs, this study shows how tenants face rising cost burdens, unsafe conditions, housing insecurity, displacement, and homelessness. Informed by a conceptual depth and rich quantitative and qualitative data, the authors assess how policy could address the affordable rental housing crisis and the need for investment in public, non-profit, housing provision protected from market forces. Highlighting the role of community-engaged, public sociology as a driver for addressing a pressing social issue, this is refreshing reading for those interested in alternative and critical analyzes of the rental housing system in the US, urban case studies, and the struggles of renters/tenants as an exploited class.

Visual foundation models, characterized by their robust generalization and adaptability, serve as the basis for a wide array of downstream tasks. When fine-tuned for specific tasks, these models encapsulate confidential and valuable task-specific knowledge, making them prime targets for model stealing (MS) attacks. While recent efforts have exposed MS threats in practical scenarios such as data-free and hard-label contexts, these attacks predominantly target traditional victim models trained from scratch. Fine-tuned visual foundation models, pre-trained on vast and diverse datasets and then fine-tuned on downstream tasks, present significant challenges for traditional MS attacks to extract task-specific knowledge. In this paper, we introduce an innovative MS attack, named SIAMESE, to steal fine-tuned visual foundation models under black-box, data-free, and hard-label settings. The core approach of SIAMESE involves constructing a stolen model using a foundation model that is efficiently and concurrently fine-tuned with multiple diversified soft prompts. To integrate the knowledge derived from these prompts, we propose a novel and tractable loss function that analyzes the output distributions while enforcing orthogonality among the prompts to minimize interference. Additionally, a unique alignment module enhances SIAMESE by synchronizing interpretations between the victim and stolen models. Extensive experiments validate that SIAMESE outperforms state-of-the-art baseline attacks over 10% in accuracy, exposing the heightened vulnerability of fine-tuned visual foundation models to MS threats.

Bioactive peptides are short amino acid sequences involved in regulating key physiological processes, including cellular signaling and metabolic functions. Despite their small size, these peptides demonstrate high specificity and significant biological activity through receptor interactions. Silkworm pupae are a rich source of proteins, lipids, carbohydrates, minerals, vitamins, and polyphenolic compounds, making them a promising raw material for peptide extraction. However, effective extraction requires an initial pretreatment step to enhance raw material preservation and facilitate lipid removal, which otherwise impairs protease activity during enzymatic hydrolysis. In this study, lipase-assisted hydrolysis was employed to remove lipids from silkworm pupae powder. Using lipase (CAS 9001-62-1), optimal conditions were established at 50℃, 1.5% enzyme concentration, and 2 hours of reaction time. This pretreatment significantly improved protease hydrolysis efficiency, resulting in a 44% increase in amino nitrogen yield compared to untreated samples.

With the advancement of artificial intelligence (AI) technologies, object detection, data processing, and lidar navigation have become increasingly prevalent in autonomous driving, offering convenient transportation solutions. However, challenges such as low obstacle detection accuracy, high error rates, and frequent failures in emergency avoidance persist, necessitating further research into AI applications in autonomous driving. This paper explores the application scenarios of object detection, data processing, and navigation technologies in autonomous driving, compares the advantages and disadvantages of autonomous and human driving, and proposes a solution based on the fusion of data processing, visual navigation, and lidar navigation. The aim is to enhance the intelligence level of autonomous driving and provide theoretical support and practical references for the technology.

The objective of this study is to clarify the impact of social responsibility on the profitability of insurance companies operating in Vietnam. As a service business, the vessel absorbs risk for the economy. The database was collected online by the authors. The survey period was from May 2024 to November 2024. Secondary data were collected from 2013 to 2023. The research results were based on 712 responses from insurance company employees, insurance consumers, and domestic insurance experts. Using quantitative research methods, PLS-SEM linear structural modeling was performed using SPSS and AMOS 20 software. The results indicate that CSR is positively correlated with the profitability of insurance businesses, primarily through the responsibility of insurance companies to consumers. Recommended solutions include: 1) Commitment to the quality of CSR reporting; 2) Protecting consumer rights through model transaction contracts; 3) Implementing a code of business ethics; 4) Addressing issues in the insurance compensation settlement process. The study supplements basic theories for insurance businesses and recommends that they actively invest in CSR programs to enhance profitability.