Reliability in natural language generation (NLG) is often compromised by semantically plausible but unsupported continuations. We present Two-Stage Distributionally Robust Decoding (TS-DRD), an inference-time framework that combines Wasserstein-robust logit correction with uncertainty-conditioned affine steering. The robust correction is analyzed at the level of the surrogate objective used inside the transport layer. A first-order variational characterization explains why the correction smooths semantically vulnerable token configurations, although it provides no theorem-level guarantee on factual correctness or hallucination rate. Factuality-oriented gains are therefore reported as empirical findings on public benchmarks. We evaluate TS-DRD on TruthfulQA (N=817), HaluEval (N=10,000), CNN/DailyMail (N=11,490), and PopQA. Empirically, TS-DRD improves TruthfulQA MC2 accuracy from 53.8% to 58.2% over static robust baselines and reduces the HaluEval summarization hallucination rate from 28.8% to 24.2%. Disaggregated diagnostics further characterize the source of these gains: a baseline-correct vs. baseline-hallucinating bucket split shows that gains come predominantly from correcting hallucinated cases rather than from uniformly shortening responses; a held-out fidelity check between our single-pass uncertainty predictor and the offline Monte Carlo (MC) Dropout statistics it approximates yields Pearson r=0.86; and per-method generation-length and entity-density measurements rule out trivial conservative shortening. These results support TS-DRD as a practical approach for reliability-oriented decoding in information processing tasks.
Using panel data of 83 Belt and Road Initiative (BRI) countries from 2005 to 2023, we investigate the convergence dynamics of energy poverty to determine whether countries are escaping the energy poverty trap and the role of China's outward foreign direct investment (OFDI) in the convergence process. We find that while traditional convergence tests confirm a catch-up effect, the Phillips-Sul club convergence model identifies five distinct clubs among BRI countries: 1) severe energy poverty trap; 2) high energy poverty with slow convergence; 3) medium energy poverty with sluggish improvement; 4) low energy poverty with steady progress; 5) lowest energy poverty frontier with the fastest convergence. Furthermore, China's OFDI in infrastructure and economic growth sectors within BRI countries promotes the formation of lower energy poverty convergence clubs. Infrastructure-oriented OFDI is particularly effective in the most severe and the least energy poverty clubs, while market-oriented OFDI demonstrates a more consistent and widespread effect across all clubs. Differentiated strategies and strengthened multilateral coordination are essential to maximize poverty reduction outcomes.
Identifying structural dependencies among cryptocurrencies and predicting cross-sectional returns are fundamental to effective cryptocurrency portfolio management. Such structural dependencies involve two interacting layers. The first is the return and volatility spillover effect from BTC/ETH to other cryptocurrencies, driven by their market dominance; this spillover exhibits pronounced time-varying characteristics, causing the effectiveness of predictive signals to drift across market states. The second is the multi-type linkage network formed among cryptocurrencies through shared technological infrastructure, application domains, and investor attention. This paper proposes a unified end-to-end framework that addresses these two layers with dedicated modeling mechanisms. For the spillover layer, the framework employs BTC/ETH return and volume statistics as observable proxies for the prevailing spillover regime, and dynamically reweights predictive features to adapt to spillover-induced shifts in feature effectiveness. For the linkage layer, the framework constructs a multi-relational typed graph attention network with four interpretable relation channels–three derived from an expert-curated three-layer domain taxonomy and one from investor co-attention data–to aggregate cross-asset information. Experiments on 150 cryptocurrencies from Binance spanning 2020-2025 demonstrate that the framework achieves an Information Coefficient (IC) of 0.071 and a long-short portfolio annualized return of 63% with a Sharpe ratio of 2.25. Interpretability analysis reveals economically coherent, state-dependent relation utilization patterns, providing traceable attribution evidence for risk management and systematic asset allocation in cryptocurrency portfolios.
Graph-structured data exhibits universality and widespread applicability across diverse domains, such as social network analysis, biochemistry, financial fraud detection, and network security. Significant strides have been made in leveraging Graph Neural Networks (GNNs) to achieve remarkable success in these areas. However, in real-world scenarios, the training environment for models is often far from ideal, leading to substantial performance degradation of GNN models due to various unfavorable factors, including imbalance in data distribution, the presence of noise in erroneous data, privacy protection of sensitive information, and generalization capability for out-of-distribution (OOD) scenarios. To tackle these issues, substantial efforts have been devoted to improving the performance of GNN models in practical real-world scenarios, as well as enhancing their reliability and robustness. In this paper, we present a comprehensive survey that systematically reviews existing GNN models, focusing on solutions to the four mentioned real-world challenges including imbalance, noise, privacy, and OOD in practical scenarios that many existing reviews have not considered. Specifically, we first highlight the four key challenges faced by existing GNNs, paving the way for our exploration of real-world GNN models. Subsequently, we provide detailed discussions on these four aspects, dissecting how these solutions contribute to enhancing the reliability and robustness of GNN models. Last but not least, we outline promising directions and offer future perspectives in the field.
Geopolitical shocks complicate climate governance and may heighten climate vulnerability, but the magnitude, channels, and the offsetting role of green policy remain unclear. In this context, this study uses panel data for 41 countries from 1995 to 2021 and uses the European Green Deal (EGD) as a policy example to examine how policy intervention shapes the relationship between geopolitical risk and climate vulnerability, employing a two-way fixed-effects model. The main findings are: (1) geopolitical risk exacerbates climate vulnerability, primarily through negative impacts on food, water, health, and ecosystems; (2) the green transition and green investment effectively mitigate the adverse impact of geopolitical risk on climate vulnerability; and (3) by reinforcing these transition and investment channels, the EGD further reduces climate vulnerability associated with geopolitical conflicts. Based on these findings, policy should speed the green transition, mobilize private capital, and strengthen regional coordination to reduce the impact of geopolitical conflicts and enhance climate resilience.
