2 August 2022 in Technology Analysis & Strategic Management
Current research on the relationship between digital transformation and financial performance (FP) remains inconclusive. This paper conceptualizes digital transformation as technology-based digital transformation (TDT) and market-based digital transformation (MDT), and explores their heterogeneous impacts on FP. It is further assessed how such distinctive effects are moderated by three types of entrepreneurs’ social capital, i.e. entrepreneurs’ technological social capital (ETSC), entrepreneurs’ business social capital (EBSC), and entrepreneurs’ institutional social capital (EISC). Using a carefully assembled dataset containing the relevant information of 2363 listed Chinese manufacturing firms covering the period of 2008–2018, these relationships are empirically examined, and the following is found: (1) TDT positively affects FP; (2) MDT has a time-lagged positive effect on FP; (3) ETSC positively moderates the relationship between MDT and FP; (4) EBSC positively moderates the impacts of TDT and MDT on FP. These findings enrich the understanding of the concept and application of digital transformation.
24 March 2024 in Proceedings of the AAAI Conference on Artificial Intelligence
Existing neural models are demonstrated to struggle with compositional generalization (CG), i.e., the ability to systematically generalize to unseen compositions of seen components. A key reason for failure on CG is that the syntactic and semantic representations of sequences in both the uppermost layer of the encoder and decoder are entangled. However, previous work concentrates on separating the learning of syntax and semantics instead of exploring the reasons behind the representation entanglement (RE) problem to solve it. We explain why it exists by analyzing the representation evolving mechanism from the bottom to the top of the Transformer layers. We find that the ``shallow'' residual connections within each layer fail to fuse previous layers' information effectively, leading to information forgetting between layers and further the RE problems. Inspired by this, we propose LRF, a novel Layer-wise Representation Fusion framework for CG, which learns to fuse previous layers' information back into the encoding and decoding process effectively through introducing a fuse-attention module at each encoder and decoder layer. LRF achieves promising results on two realistic benchmarks, empirically demonstrating the effectiveness of our proposal. Codes are available at https://github.com/thinkaboutzero/LRF.
21 December 2023 in arXiv
Existing neural models are demonstrated to struggle with compositional generalization (CG), i.e., the ability to systematically generalize to unseen compositions of seen components. A key reason for failure on CG is that the syntactic and semantic representations of sequences in both the uppermost layer of the encoder and decoder are entangled. However, previous work concentrates on separating the learning of syntax and semantics instead of exploring the reasons behind the representation entanglement (RE) problem to solve it. We explain why it exists by analyzing the representation evolving mechanism from the bottom to the top of the Transformer layers. We find that the ``shallow'' residual connections within each layer fail to fuse previous layers' information effectively, leading to information forgetting between layers and further the RE problems. Inspired by this, we propose LRF, a novel \textbf{L}ayer-wise \textbf{R}epresentation \textbf{F}usion framework for CG, which learns to fuse previous layers' information back into the encoding and decoding process effectively through introducing a \emph{fuse-attention module} at each encoder and decoder layer. LRF achieves promising results on two realistic benchmarks, empirically demonstrating the effectiveness of our proposal.