arXiv subjects

Rui Cong

Publications and source records attributed to Rui Cong.

3 recordsLinked to original sources

arXiv2026

Beyond Visual Similarity: Entity-Aligned Retrieval for Knowledge-Based Visual Question Answering

Knowledge-Based Visual Question Answering (KB-VQA) relies on retrieving external information to answer queries involving long-tail entities. However, existing retrieval pipelines predominantly employ CLIP-style dual encoders, which prioritize surface-level visual similarity over entity-level semantic alignment. This paradigm often fails when semantically identical concepts exhibit large visual variations or when distinct entities appear visually similar. To address this, we propose KBMR, the first MLLM-based embedding retriever tailored for KB-VQA. Leveraging the robust autoregressive capabilities of MLLMs, KBMR maps images into a semantic space that better preserves concept identity. To tackle the challenge of noisy supervision in Wikipedia-scale retrieval, we introduce an MLLM-based semantic discriminator that generates continuous entity-consistency weights. These weights guide a novel continuous semantic distillation objective, enabling effective hard negative sampling and soft supervision beyond rigid binary labels. Extensive experiments demonstrate that KBMR significantly outperforms CLIP baselines, yielding up to a 14.7% improvement in retrieval Recall@1 and a 9.4% gain in end-to-end VQA accuracy. Code is available at https://github.com/realHarryX/KBMR.

cs.CV↗

arXiv2016

Stochastic evolutionary dynamics of minimum-effort coordination games

The minimum-effort coordination game, having potentially important implications in both evolutionary biology and sociology, draws recently more attention for the fact that human behavior in this social dilemma is often inconsistent with the predictions of classic game theory. In the framework of classic game theory, any common effort level is a strict and trembling hand perfect Nash equilibrium, so that no desideratum is provided for selecting among them. Behavior experiments, however, show that the effort levels employed by subjects are inversely related to the effort costs. Here, we combine coalescence theory and evolutionary game theory to investigate this game in finite populations. Both analytic results and individual-based simulations show that effort costs play a key role in the evolution of contribution levels, which is in good agreement with those observed experimentally. Besides well-mixed populations, set structured populations, where the population structure itself is a consequence of the evolutionary process, have also been taken into consideration. Therein we find that large number of sets and moderate migration rate greatly promote effort levels, especially for high effort costs. Our results may provide theoretical explanations for coordination behaviors observed in real life from an evolutionary perspective.

physics.soc-ph↗

arXiv2016

The coevolution of overconfidence and bluffing in the resource competition game

Resources are often limited, therefore it is essential how convincingly competitors present their claims for them. Beside a player's natural capacity, here overconfidence and bluffing may also play a decisive role and influence how to share a restricted reward. While bluff provides clear, but risky advantage, overconfidence, as a form of self-deception, could be harmful to its user. Still, it is a long-standing puzzle why these potentially damaging biases are maintained and evolving to a high level in the human society. Within the framework of evolutionary game theory, we present a simple version of resource competition game in which the coevolution of overconfidence and bluffing is fundamental, which is capable to explain their prevalence in structured populations. Interestingly, bluffing seems apt to evolve to higher level than corresponding overconfidence and in general the former is less resistant to punishment than the latter. Moreover, topological feature of the social network plays an intricate role in the spreading of overconfidence and bluffing. While the heterogeneity of interactions facilitates bluffing, it also increases efficiency of adequate punishment against overconfident behavior. Furthermore, increasing the degree of homogeneous networks can trigger similar effect. We also observed that having high real capability may accommodate both bluffing ability and overconfidence simultaneously.

physics.soc-ph↗