Senjie Jin, Lu Chen, Zhiheng Xi, Yuhui Wang, Sirui Song, Yuhao Zhou, Xinbo Zhang, Peng Sun, Hong Lu, Tao Gui, Qi Zhang, Xuanjing Huang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

The stack loading problem has been studied in recent years for its great impact on the container loading and unloading operations. Among different objectives of the problem considered, minimizing the total number of unordered stackings and minimizing the total number of used stacks are the two important ones, which ensure efficient loading and unloading schedules, as well as reduce storage costs, respectively. The load-bearing setting, where each container has its own weight and bearing weight, is frequently considered in box packing operations but rarely in the existing studies on the stack loading problem. However, the load-bearing constraint on containers is very important for stack loading, because safety is of paramount importance. This paper is the first study on the stack loading problem with the load-bearing constraint with an aim to minimize the number of stacks and the number of unordered stackings. We show that this problem is strongly$\mathcal{NP}$-hard even when the number of stacks is given and equals$2$. For the case where the number of stacks is given and jobs on the bottom tiers are fixed, we show that the problem can be solved by dynamic programming in pseudo-polynomial time. For the general problem, based on a two-index integer linear programming formulation and a tabu search heuristic, we develop a binary-search based matheuristic. Our experimental results demonstrate the efficiency and effectiveness of the newly developed matheuristic.Note to Practitioners—This paper is motivated by the stack loading problem and is the first study on the load-bearing limit case. The load-bearing limit is a fundamental constraint but has not been taken into account in studies in the stack loading problem. Based on ISO Standard 1496, the corner posts and corner fittings of ISO Series I containers can bear a certain amount of weight. If the total weight of the containers above exceeds the load-bearing limit of the lower container, it will hazard the load-bearing safety. This paper proposes two problem formulations: three-index formulation and two-index formulation. The three-index formulation adds the load-bearing limit to the existing stack loading problem formulation. It turns out that the traditional three-index formulation of the stack loading problem is not efficient when being used in solving the problem with load-bearing constraints. Therefore, we propose a new two-index formulation. Apart from the theoretical results, this paper proposes a matheuristic solution framework: firstly, using binary search with greedy matheuristic for feasibility checking to minimize the number of stacks, and secondly, using tabu search matheuristic to minimize the number of unordered stackings. In future research, we will apply the matheuristic to different types of container scenarios and the parallel stack loading case.

One way to enhance the reasoning capability of Large Language Models (LLMs) is to conduct Supervised Fine-Tuning (SFT) using Chain-of-Thought (CoT) annotations.This approach does not show sufficiently strong generalization ability, however, because the training only relies on the given CoT data.In math problemsolving, for example, there is usually only one annotated reasoning path for each question in the training data.Intuitively, it would be better for the algorithm to learn from multiple annotated reasoning paths given a question.To address this issue, we propose a simple yet effective approach called Reinforced Fine-Tuning (ReFT) to enhance the generalizability of learning LLMs for reasoning, with math problemsolving as an example.ReFT first warmups the model with SFT, and then employs on-line reinforcement learning, specifically the PPO algorithm in this paper, to further fine-tune the model, where an abundance of reasoning paths are automatically sampled given the question and the rewards are naturally derived from the ground-truth answers.Extensive experiments on GSM8K, MathQA, and SVAMP datasets show that ReFT significantly outperforms SFT, and the performance can be potentially further boosted by combining inference-time strategies such as majority voting and re-ranking.Note that ReFT obtains the improvement by learning from the same training questions as SFT, without relying on extra or augmented training questions.This indicates a superior generalization ability for ReFT 1 .* indicates equal contribution, † indicates corresponding author 1 Code: https://github.com/lqtrung1998/mwp_ReFTSupervised Fine-Tuning Model Question (x): Weng earns $12 an hour for babysitting.Yesterday, she just did 50 minutes of babysitting.How much did she earn?

In this paper, we propose R$^3$: Learning Reasoning through Reverse Curriculum Reinforcement Learning (RL), a novel method that employs only outcome supervision to achieve the benefits of process supervision for large language models. The core challenge in applying RL to complex reasoning is to identify a sequence of actions that result in positive rewards and provide appropriate supervision for optimization. Outcome supervision provides sparse rewards for final results without identifying error locations, whereas process supervision offers step-wise rewards but requires extensive manual annotation. R$^3$ overcomes these limitations by learning from correct demonstrations. Specifically, R$^3$ progressively slides the start state of reasoning from a demonstration’s end to its beginning, facilitating easier model exploration at all stages. Thus, R$^3$ establishes a step-wise curriculum, allowing outcome supervision to offer step-level signals and precisely pinpoint errors. Using Llama2-7B, our method surpasses RL baseline on eight reasoning tasks by $4.1$ points on average. Notably, in program-based reasoning, 7B-scale models perform comparably to larger models or closed-source models with our R$^3$.

Data hiding (DH) is a crucial means to ensure the security of secret data during transmission. It embeds secret data into a carrier to prevent attackers from detecting it. Current research in data hiding focuses on enhancing the covert nature of the embedded secret data and the embedding capacity of the carrier. This article presents a high-capacity and high-security data hiding scheme in encrypted Images based on image filtering and blocking (IFIB-DHEI). Our contributions are as follows: (1) Improves the predictor used in the data hiding process to make it more accurate in predicting pixel points. (2) Using Gaussian filters for preprocessing to improve the smoothness of carrier images greatly increases the embedding capacity of carrier images. (3) Multi-data-hiders technology is used to encrypt images and secret data while transmitting marked encrypted images separately. The proposed scheme achieves an average embedding capacity of 4.49 bpp in the BOSSBase dataset and 4.48 bpp in the BOWS-2 dataset. A larger embedding capacity can be achieved by adjusting the filter parameters.

Abstract Over the past 70 years, the semiconductor industry has undergone transformative changes, largely driven by the miniaturization of devices and the integration of innovative structures and materials. Two-dimensional (2D) materials like transition metal dichalcogenides (TMDs) and graphene are pivotal in overcoming the limitations of silicon-based technologies, offering innovative approaches in transistor design and functionality, enabling atomic-thin channel transistors and monolithic 3D integration. We review the important progress in the application of 2D materials in future information technology, focusing in particular on microelectronics and optoelectronics. We comprehensively summarize the key advancements across material production, characterization metrology, electronic devices, optoelectronic devices, and heterogeneous integration on silicon. A strategic roadmap and key challenges for the transition of 2D materials from basic research to industrial development are outlined. To facilitate such a transition, key technologies and tools dedicated to 2D materials must be developed to meet industrial standards, and the employment of AI in material growth, characterizations, and circuit design will be essential. It is time for academia to actively engage with industry to drive the next 10 years of 2D material research.

In recent years, DNA computing model has gradually attracted attention due to its low energy consumption, high capacity of storing information and good parallelism. DNA computational model is calculated by DNA molecule as the medium, so its core is to design a high quality DNA sequence conforming to various constraints. Designing DNA sequences that meet a series of constraints, such as temperature, H-measure, and continuity, is a typical multi-objective optimization problem. In traditional multi-objective optimization problems, various fitness functions are usually only related to their own solutions, and have no correlation with other redundant candidate solutions. Based on the DNA coding problem's characteristics, we propose a two-stage constrained multi-objective evolutionary algorithm. Our algorithm overcomes shortcomings of traditional algorithms in solving DNA coding problems which are easy to fall into local optimal solutions. Experimental results demonstrate that our algorithm is effective and reliable in solving DNA coding problems when compared to other mainstream algorithms from recent years.

Given a natural language statement, how to verify its veracity against a large-scale textual knowledge source like Wikipedia? Most existing neural models make predictions without giving clues about which part of a false claim goes wrong. In this paper, we propose LOREN, an approach for interpretable fact verification. We decompose the verification of the whole claim at phrase-level, where the veracity of the phrases serves as explanations and can be aggregated into the final verdict according to logical rules. The key insight of LOREN is to represent claim phrase veracity as three-valued latent variables, which are regularized by aggregation logical rules. The final claim verification is based on all latent variables. Thus, LOREN enjoys the additional benefit of interpretability --- it is easy to explain how it reaches certain results with claim phrase veracity. Experiments on a public fact verification benchmark show that LOREN is competitive against previous approaches while enjoying the merit of faithful and accurate interpretability. The resources of LOREN are available at: https://github.com/jiangjiechen/LOREN.

The main task of noninvasive load monitoring is to disaggregate the power consumption of a single household appliance from an electricity meter that detects the power consumption of all household appliances. The deep neural network method has achieved leading results in this field. In this article, a multichannel spatio-temporal feature fusion method is proposed, where the spatial features extracted by convolution neural network and the temporal features extracted by the recurrent neural network are fused. And the attention module is introduced to further improve the performance of the model. Finally, the effectiveness and superiority of the proposed method are verified on three public datasets.

Semantic parsing-based RDF question answering (QA) systems are to interpret users' natural language questions as query graphs and return answers over RDF repository. However, due to the complexity of linking natural phrases with specific RDF items (e.g., entities and predicates), it remains difficult to understand users' question sentences precisely, hence QA systems may not meet users' expectation, offering wrong answers and dismissing some correct answers. In this paper, we design an I nteractive M echanism aiming for PRO motion V ia users' fe edback to Q A systems (IMPROVE-QA), a whole framework to not only make existing QA systems return more precise answers based on a few feedbacks over the original answers given by RDF QA systems, but also enhance paraphrasing dictionaries to ensure a continuous-learning capability in improving RDF QA systems. To provide better interactivity and online performance, we design a holistic graph mining algorithm (HWspan) to automatically refine the query graph. Extensive experiments on both Freebase and DBpedia confirm the effectiveness and superiority of our approach.

Although natural language question answering over knowledge graphs have been studied in the literature, existing methods have some limitations in answering complex questions.To address that, in this paper, we propose a State Transition-based approach to translate a complex natural language question N to a semantic query graph (SQG) Q S , which is used to match the underlying knowledge graph to find the answers to question N .In order to generate Q S , we propose four primitive operations (expand, fold, connect and merge) and a learning-based state transition approach.Extensive experiments on several benchmarks (such as QALD, WebQuestions and ComplexQuestions) with two knowledge bases (DBpedia and Freebase) confirm the superiority of our approach compared with stateof-the-arts.

RDF Question/Answering(Q/A) systems can interpret user's question N as SPARQL query Q and return answer set $Q(D)$ over RDF repository D to the user. However, due to the complexity of linking natural phrases with specific RDF items (e.g., entities and predicates), it remains difficult to understand users' questions precisely, hence $Q(D)$ may not meet users' expectation, offering wrong answers and dismissing some correct answers. In this demo, we design an I Interactive Mechanism aiming for PRO motion V ia feedback to Q/A systems (IMPROVE-QA), a whole platform to make existing Q/A systems return more precise answers (denoted as $\mathcal Q^\prime (D)$) to users. Based on user's feedback over $Q(D)$, IMPROVE-QA automatically refines the original query Q into a new query graph $\mathcal Q^\prime $ with minimum modifications, where $\mathcal Q^\prime (D)$ provides more precise answers. We will also demonstrate how IMPROVE-QA can apply the "lesson'' learned from the user in each query to improve the precision of Q/A systems on subsequent natural language questions.