Machine-learned predictions can speed up offline NP-hard optimization, but asking a predictor what to do amounts to asking it to solve the problem, and committing an unchecked prediction forfeits every worst-case guarantee. CASP (Certificate-Augmented Solution Pruning) instead asks which parts of the search space may be ignored, and accepts each answer only after a sound polynomial-time verifier has checked it, so correctness never depends on prediction quality. We develop the learning theory of this design. The verifier makes the induced loss class uniformly bounded, so certificate parameters are learnable from $\tilde O(\varepsilon^{-2}\log K)$ samples ($K$ the maximum instance size), whereas the unverified commitment class admits no distribution-free rate and, under cost spread $R$, none below $Ω(R/\varepsilon^2)$. Filtering noisy predictions by verifiable confidence dominates the standard min-combiner, with a margin we compute in closed form, and the prediction stays useful even given the LP, because it breaks ties on degenerate optimal faces, where every symmetric LP policy, meaning one whose commitments depend on the instance only through the verifiable confidence values, provably stalls. Experiments on five problems test the theory's quantitative predictions. With trained predictors, unverified pruning loses up to $26%$ of the optimum under distribution shift, while the verified deployment of the same predictions loses nothing.

The aggregation benefit of a committee of artificial intelligence (AI) agents comes from complementary information across members. Classical voting guarantees assume independent errors. Language-model errors often co-occur on the same cases. We combine Sah-Stiglitz screening with error dependence that can differ between good and bad cases. In a homogeneous exchangeable Gaussian-copula model, shared errors create a positive asymptotic error floor for majority voting and can change the approval threshold that minimizes expected loss. We estimate a heterogeneous extension from 174,384 votes cast by 28 language models on four binary-screening benchmarks. Parameters estimated from odd-indexed items predicted committee loss on even-indexed items. For the sampled committee composition, the full-matrix dependence model increased identity-line R^2 from 0.840 under independence to 0.967. In a design-balanced analysis, cost-sensitive threshold selection under independence reduced scaled loss from 60.25 for majority to 52.50. Modeling dependence reduced it further to 50.77, an incremental improvement of 1.73 units (95% bootstrap CI, 0.68-2.33). The overall reduction from majority was 15.73% (95% bootstrap CI, 13.41-16.75%).

Large language models now translate natural-language descriptions of decision problems into solver-ready optimization models, but they fail silently. A generated model often runs and still formulates the wrong problem. This paper develops a theory of falsification-based verification for this setting. Every numeric quantity in the description is a typed slot, and a candidate model is tested only through solver calls on slot-transformed instances; no reference model or label is consulted. From duality, comparative statics, and polyhedral limit arguments we derive a battery of test classes covering directions, curvature, crush probes, prohibitive limits, annihilation, and exchange. Every test is sound, so a violation certifies unfaithfulness and the false-positive rate is zero by design. We characterize what such verification can never see, give conditions under which the canonical error classes are detected with certainty, and prove that no fixed-threshold perturbation tester is simultaneously sound and nontrivial. Experiments on 326 ground-truth models from NL4OPT and four benchmark families confirm the theory. The battery attains a 0.0

Current studies of portfolio mainly use reinforcement learning methods to build models aimed at achieving high investment returns while minimizing risks from market uncertainties. Two main issues will be considered: First, the complexity of financial markets makes it challenging to capture asset price change patterns. Second, current research assumes stock prices accurately show all asset information, and historical prices alone can predict future trends. However, numerous external factors can influence future judgments. We introduce PortfolioZero, a novel model to address these problems. PortfolioZero utilizes three connected deep neural networks combined with a Monte Carlo Tree to discover patterns of financial assets. In the representation network, a Transformer-based model is used to embed financial price data to capture temporal dynamics and potential correlations, providing richer feature representations; the prediction network and Monte Carlo Tree Search are redesigned to handle the continuous action space. Furthermore, we use the StructBERT model to process financial text data, extracting market information into sentiment scores, which are used to reconstruct two reward functions to capture dynamic changes of the financial market. In experiments conducted on the China A-share market, we compared our model with traditional portfolio methods and cutting-edge deep reinforcement learning algorithms. PortfolioZero achieved an average annualized return rate of 21.21% across three portfolio types, outperforming SARL by 20.64% and DDPG by 41.97%, while sentiment-enhanced reward functions improved average annualized return rate by 35% compared to basic reward.

Taking the case driven “Principles and Applications of Big Data Analysis” course ideological and political education as the research object, we analyze the current problems of ideological and political education in science and engineering courses in universities, and propose an overall design framework for course ideological and political education. The principles and criteria for case selection were provided, while typical cases were provided for the four ideological and political elements of science and technology: Serving the country, scientific thinking, engineering ethics, and craftsmanship spirit. The specific implementation methods of case discussions were also introduced in detail. Through a questionnaire survey, it was found that case driven curriculum ideological and political education has a positive impact on students’ learning effectiveness and ideological and political literacy. This study provides new ideas and methods for the ideological and political education of science and engineering courses in universities, and has certain significance for the practice of ideological and political education in science and engineering courses in universities.

Deep reinforcement learning (DRL) can be used to extract deep features that can be incorporated into reinforcement learning systems to enable improved decision-making; DRL can therefore also be used for managing stock portfolios. Traditional methods cannot fully exploit the advantages of DRL because they are generally based on real-time stock quotes, which do not have sufficient features for making comprehensive decisions. In this study, in addition to stock quotes, we introduced stock financial indices as additional stock features. Moreover, we used Markowitz mean-variance theory for determining stock correlation. A three-agent deep reinforcement learning model called Collaborative Multi-agent reinforcement learning-based stock Portfolio management System (CMPS) was designed and trained based on fused data. In CMPS, each agent was implemented with a deep Q-network to obtain the features of time-series stock data, and a self-attention network was used to combine the output of each agent. We added a risk-free asset strategy to CMPS to prevent risks and referred to this model as CMPS-Risk Free (CMPS-RF). We conducted experiments under different market conditions using the stock data of China Shanghai Stock Exchange 50 and compared our model with the state-of-the-art models. The results showed that CMPS could obtain better profits than the compared benchmark models, and CMPS-RF was able to accurately recognize the market risk and achieved the best Sharpe and Calmar ratios. The study findings are expected to aid in the development of an efficient investment-trading strategy.

Existing portfolio management models based on deep reinforcement learning adjust funds dynamically by constructing an action space with specific trading behaviors. However, this method cannot effectively adapt to complex investment environments. This paper proposes an LSTM-ER-DCPPO portfolio management model based on the improved PPO algorithm. The model provides stable policy learning, enhanced strategy exploration, and trend identification, enabling more precise capital allocation through continuous asset weight output. To evaluate the model, a simulated market trading environment based on real stock data was created. The experimental results show that the LSTM-ER-DCPPO model achieved an annualized return of 58.62%, a maximum drawdown of 8.53%, a Calmar ratio of 6.873, and a Sharpe ratio of 2.434, outperforming other benchmark models.

Due to block Propagation Delay and mining competition in blockchain networks, conflicting blocks are generated and propagated, leading to blockchain forks and the potential for double-spending attacks. Thus, reducing Propagation Delay and Blockchain Fork Rate is a critical research challenge. Traditional methods demonstrate high Propagation Delay and fork rate alongside low Propagation Coverage Rate. We introduce the Seed Node based Propagation (SNP) algorithm to tackle this issue. SNP utilizes a cluster generation method to reorganize nodes, selecting the seed node with the smallest distance to others in each cluster. An update mechanism maintains the clusters when nodes join or leave. Theoretical analysis shows O(n) computational complexity. We use an innovative metric, the Propagation Measurement Index (PMI) to evaluate our algorithms, and extensive experiments demonstrate a two-order-of-magnitude reduction in Propagation Delay and Propagation Coverage Rate. Moreover, SNP outperforms 6 state-of-the-art algorithms in terms of PMI.

The state-of-the-art label propagation algorithm has to randomly updates node and selects label, leading to the monster community and label oscillation problems. To address these, we propose a label comprehensive in-fluence index(CII), which combines the importance of nodes and the similarity between nodes. Based on CII, we propose the similarity-based label propagation algorithm named ISLPA. We performed experiments to evaluate ISLPA on the Lancichinetti-Fortunato-Radicchi artificial benchmark network and a real net-work. Results show that ISLPA improves the community division quality; the normalized mutual information and the modularity of ISLPA were 20% higher than the compared algorithms, respectively. We implemented parallel ISLPA on Spark, and the results show our parallel ISLPA reduced the running time by approximately 50% while guaranteeing community division quality in large networks of more than 100K nodes. Compared with another parallel algorithm, both the NMI and modularity of parallel ISLPA were approximately 20% higher.

A key challenge in current stock market analysis research is how to extract meaningful and actionable insights from the vast and continuously updated amount of data available. Common data mining and machine learning algorithms primarily focus on uncovering correlations between variables but struggle to determine causality, which hinders investors' ability to make informed decisions. The widely used Granger causality test in stock market analysis has limitations when applied to high-dimensional data, given issues of mis-fitting and inefficiency. As a result, it falls short in coping with the scale of stock data available today. To address these challenges, we employ ANM (Additive Noise Model), a causal inference method based on a nonlinear causal function model, to identify the factors influencing stock price changes. Through experiments, we find that ANM demonstrates a certain level of reliability in factor selection compared to Granger analysis when discovering causal relationships among low-dimensional macroeconomic factors. Additionally, we compare the effects of ANM-selected factors with full factor models and models constructed using PCA factors, combined with machine learning algorithms. The results indicate that the factors identified by ANM outperform the other two methods. Our findings suggest that ANM serves as a valuable approach for factor selection in stock market analysis. By leveraging ANM and considering its results in conjunction with machine learning algorithms, we can more effectively identify influential factors for stock price changes compared to other methodologies. It is important to note that the field of stock market analysis and causal inference is constantly evolving, and different methodologies have their own strengths and weaknesses. Ongoing research and experimentation are essential to further refine and validate these approaches.

Ranker and retriever are two important components in dense passage retrieval. The retriever typically adopts a dual-encoder model, where queries and documents are separately input into two pre-trained models, and the vectors generated by the models are used for similarity calculation. The ranker often uses a cross-encoder model, where the concatenated query-document pairs are input into a pre-trained model to obtain word similarities. However, the dual-encoder model lacks interaction between queries and documents due to its independent encoding, while the cross-encoder model requires substantial computational cost for attention calculation, making it difficult to obtain real-time retrieval results. In this paper, we propose a dense retrieval model called MD2PR based on multi-level distillation. In this model, we distill the knowledge learned from the cross-encoder to the dual-encoder at both the sentence level and word level. Sentence-level distillation enhances the dual-encoder on capturing the themes and emotions of sentences. Word-level distillation improves the dual-encoder in analysis of word semantics and relationships. As a result, the dual-encoder can be used independently for subsequent encoding and retrieval, avoiding the significant computational cost associated with the participation of the cross-encoder. Furthermore, we propose a simple dynamic filtering method, which updates the threshold during multiple training iterations to ensure the effective identification of false negatives and thus obtains a more comprehensive semantic representation space. The experimental results over two standard datasets show our MD2PR outperforms 11 baseline models in terms of MRR and Recall metrics.

Most existing reinforcement learning (RL)-based portfolio management model do not consider the financial indicators. They more focus on the changes in prices and interest rates. In this paper, we propose a portfolio management strategy based on the framework of Deep Deterministic Policy Gradient (DDPG). To assess the effectiveness of our strategy, we compare DDPG using financial indicators with DDPG using only price as a feature, which simulate the best performance in the financial market. Experimentally, we select a set of stocks with low correlation and compare the accumulative portfolio value of our strategy with other strategies, including Uniform Buy and Hold, Exponential Gradient, and Universal Portfolios. Through experimentation, we find that our strategy outperforms most of the other strategies, with an accumulative portfolio value of 5.782. Furthermore, our strategy exhibits a Sharp Ratio of 2.064, which is only 5.2% lower than the best-performing strategy.

Disclosure is the soul of supervision, the bridge between companies and investors, and can help public fully understand companies’ business operations. The disclosure quality of listed companies’ annual reports affects security market efficiency and investor rights protection. Currently, there are issues including but not limited to contradictory information and avoidance of important problems. Moreover, the qualities of disclosure reports from different companies vary greatly. This paper is to focus on annual reports of listed companies, dividing them into two categories based on the assessment results of disclosure by the Shenzhen Stock Exchange, and conducting the research on textual characteristics. Firstly, three characteristic indicators, tone, readability, and file size of each annual report are analyzed and compared, with validation. To explore the impact of characteristic indicators on text classification, these three indicators are introduced into text models, constructing comprehensive models. To achieve dimension reduction for training models, feature selection is performed by using Chi-square statistics. Different lengths keywords dictionaries are constructed. In conclusion, prediction performance of classifier models can be improved or maintained after introducing indicators, with the random forest model having the best improvement after introduction of indicators.

Aiming at the problems of the current methods of accelerating block propagation by optimizing the blockchain network topology: (1) these methods are based on a fixed network topology, however, the network topology of the blockchain network dynamically changes; (2) these methods take use of the local centralized characteristics, which contradict the decentralized characteristics of the blockchain network, we propose a new blockchain network routing algorithm which changes the blockchain network routing problem to a reinforcement learning problem. The algorithm reacts to the change of network environment and optimizes the route based on the feedback. Therefore, a better propagation path can be obtained by reacting to network change timely, which can ultimately reduce the total propagation delay. The experimental results show that: (1) compared with the Gossip algorithm of the bitcoin network, our proposed algorithm can reduce the total propagation delay by 38% when the number of nodes exceeds 10,000 and the degree of network variation is 10%; (2) compared with Gossip, when the number of nodes reaches 10,000, and the degree of network variation increases from 1% to 10%, our proposed algorithm can reduce the variation range of the total propagation delay from 73% to 14%.

Causal inference has become an important research direction in the field of computing. Traditional methods have mainly used Bayesian networks to discover the causal effects between variables. These methods have limitations, namely, on the one hand, the computing cost is expensive if one wants to achieve accurate results, i.e., exponential growth along with the number of variables. On the other hand, the accuracy is not good enough if one tries to reduce the computing cost. In this study, we use prior knowledge iteration or time series trend fitting between causal variables to resolve the limitations and discover bidirectional causal edges between the variables. Subsequently, we obtain real causal graphs, thus establishing a more accurate causal model for the evaluation and calculation of causal effects. We present two new algorithms, namely, the PC+ algorithm and the DCM algorithm. The PC+ algorithm is used to address the problem of the traditional PC algorithm, which needs to enumerate all Markov equivalence classes at a high computational cost or with immediate output of non-directional causal edges. In the PC+ algorithm, the causal tendency among some variables was analyzed via partial exhaustive analysis. By fixing the relatively certain causality as prior knowledge, a causal graph of higher accuracy is the final output at a low running cost. The DCM algorithm uses the d-separation strategy to improve the traditional CCM algorithm, which can only handle the pairwise fitting of variables, and thus identify the indirect causality as the direct one. By using the d-separation strategy, our DCM algorithm achieves higher accuracy while following the basic criteria of Bayesian networks. In this study, we evaluate the proposed algorithms based on the COVID-19 pandemic with experimental and theoretical analysis. The experimental results show that our improved algorithms are effective and efficient. Compared to the exponential cost of the PC algorithm, the time complexity of the PC+ algorithm is reduced to a linear level. Moreover, the accuracies of the PC+ algorithm and DCM algorithm are improved to different degrees; specifically, the accuracy of the PC+ algorithm reaches 91%, much higher than the 33% of the PC algorithm.

Mobility, group awareness, and temperature are considered as the important factors that may impact the increase in confirmed cases of the COVID-19[1]. This paper aims to verify the above factors on the COVID-19 and show the possible confounding factors of each research variable in reality. Based on this, we collected data about the epidemic from January 20, 2020 to February 24, 2021, including the relevant data of 31 provinces and regions in China. Plus, we use the directed acyclic graph (DAG)[2] to show the causal relationship between the above influencing factors and the confirmed daily epidemic cases, and the confounding is estimated based on DAG. The effective adjustment set of factors are used to perform the regression of the total causal effect among the explanatory variables and the confirmed cases of the epidemic using negative binomial regression. Through the comprehensive causal analysis of the decisive factors for the COVID-19, we provide strong evidence for population mobility, group awareness and the impact of weather on the epidemic, and estimates the possible confounding factors in all aspects of society. Incorporating the above factors, we provide suggestions for future decisions on the prevention of large-scale epidemics.

Deep reinforcement learning is a method to address the problem of achieving maximum return or getting specific goals through learning strategies when interact with environment, which can be used in portfolio investments to achieve more benefits. In this paper, we use two deep reinforcement learning methods, that is, the policy gradient (PG) and the deep deterministic policy gradient (DDPG), to develop models for stock portfolio investments. To determine the weights for a group of stocks, we first process the data using the wavelet transform method; then, we improve the methods by constructing a neural network framework with a convolution neural network (CNN) along with a Squeeze-and-Excitation Block (SE Block); in the model, we employ the average logarithmic cumulative returns and the cumulative Sharpe Ratio as the reward functions. Our experimental results show the improved model is robustness, and perform significantly better over the actual Chinese stocks in risk prevention and returns increment.

Aiming at the problems existing in the traditional classroom teaching mode adopted in the teaching of big data courses in financial universities: fail to reflect "student-centered", unable to cultivate students' autonomous learning ability, and lack of students' learning time and thinking time, we study a blended teaching method for big data course in financial universities, and design the implementation scheme of blended teaching from three stages: before class, in class and after class. At the same time, a specific teaching example is given, and the evaluation method of blended teaching is further put forward.

The traditional blockchain structure cannot adapt to large-scale and real-time-application scenarios because of its inherently slow response. To solve this problem, a theoretical framework of DAG (directed acyclic graph) blockchain is proposed, transforming the chain processing of a traditional blockchain into parallel processing. On this basis, the non-independent task-scheduling problem in the DAG blockchain environment is studied, and a fusion-partitioning genetic task-scheduling algorithm based on deterministic annealing technology in DAG blockchains is proposed. The experimental results show that the algorithm can adapt to the heterogeneity, dynamism, and wide area of DAG blockchain nodes, and its scheduling performance is better than that of the traditional scheduling algorithm. While optimizing the task-completion time, the algorithm takes account of the load-balancing problem and effectively improves the response speed. It is a feasible method for solving the non-independent task-scheduling problem in the DAG blockchain environment.