arXiv subjects

Victor

Publications and source records attributed to Victor.

3 recordsLinked to original sources

arXiv2025

Beyond Patents: R&D, Capital, and the Productivity Puzzle in Early-Stage High-Tech Firms

This study investigates the relationship between innovation activities and firm-level productivity among early-stage high-tech startups in China. Using a proprietary dataset encompassing patent records, R&D expenditures, capital valuation, and firm performance from 2020 to 2024, we examine whether and how innovation, measured by patents and R&D input, translates into economic output. Contrary to established literature, we find that patent output does not significantly contribute to either income or profit among the sampled firms. Further investigation reveals that patents may primarily serve a signaling function to external investors and policymakers, rather than reflecting true innovative productivity. In contrast, R&D expenditure shows a consistent and positive association with firm performance. Through mechanism analysis, we explore three channels (organizational environment, employee quality, and policy-driven incentives) to explain the impact of R&D, identifying capital inflow and valuation as key drivers of R&D investment. Finally, heterogeneity analysis indicates that the effects of R&D are more pronounced in sub-industries such as smart terminals and digital creativity, and for firms based in Shenzhen. Our findings challenge the prevailing assumption that patent output is a universal indicator of innovation success and underscore the context-dependent nature of innovation-performance linkages in emerging markets.

econ.GN↗

arXiv2024

From Seconds to Hours: Reviewing MultiModal Large Language Models on Comprehensive Long Video Understanding

The integration of Large Language Models (LLMs) with visual encoders has recently shown promising performance in visual understanding tasks, leveraging their inherent capability to comprehend and generate human-like text for visual reasoning. Given the diverse nature of visual data, MultiModal Large Language Models (MM-LLMs) exhibit variations in model designing and training for understanding images, short videos, and long videos. Our paper focuses on the substantial differences and unique challenges posed by long video understanding compared to static image and short video understanding. Unlike static images, short videos encompass sequential frames with both spatial and within-event temporal information, while long videos consist of multiple events with between-event and long-term temporal information. In this survey, we aim to trace and summarize the advancements of MM-LLMs from image understanding to long video understanding. We review the differences among various visual understanding tasks and highlight the challenges in long video understanding, including more fine-grained spatiotemporal details, dynamic events, and long-term dependencies. We then provide a detailed summary of the advancements in MM-LLMs in terms of model design and training methodologies for understanding long videos. Finally, we compare the performance of existing MM-LLMs on video understanding benchmarks of various lengths and discuss potential future directions for MM-LLMs in long video understanding.

cs.CV↗

arXiv2011

E- Learning: An effective pedagogical tool for learning

In the info-tech age E-Methods of learning are becoming the most important vehicle in disseminating knowledge in higher education institutions. This sector is growing and changing at a rapid speed due to developments in technologies. But teaching is an art. Can there be fun learning with raw and dry technology? How can we make the best use of E- Methods, can we make the required information and data available to the students in a flexible manner, at ease all the time? What are the advantages of traditional methods of teaching and learning? Is E-learning a progressive stage incubating all the benefits of the Manual learning or it is only a window dressing on the face of advancement? Can we convert the boring, tedious subjects into interactive, monotony breaking joyous learning? In this paper the researchers have focused on the modernization of E- Pedagogy vis-a-vis the traditional method of learning. They have highlighted the effectiveness of using the E- learning elements and various E- Methods. This work has used the decision tree algorithms particularly Classifiers.trees.J48 The obtained results show that using online examination attribute plays major role in increasing the average grade of the class in higher education. The novelty of this work is that the researchers have focused on the teaching methodology used by the faculty members and the tools available in the universities. We believe that this work will play a constructive role in building higher education system. Our generated rules/output can be used by the decision makers in the improvement of higher education system processes.

cs.CY↗