Physics-informed neural networks (PINNs) are sensitive to how collocation points are placed and refreshed during training. Existing adaptive sampling, loss-balancing, and decomposition methods improve PINNs from different angles, but most of them still assume that one collocation mechanism can remain effective across different tasks. This premise is often difficult to sustain when PDE instances exhibit markedly different source geometries. We therefore cast adaptive collocation as a task-level algorithm selection problem. We propose a geometry-aware hierarchical compositional routing framework. Before training, a deterministic top-level router extracts low-cost descriptors from the source field and assigns each task to a suitable expert sampler. Only for geometrically ambiguous cases do we activate a second-level stage-auto expert that performs short probing among a few candidates. On a benchmark with six source families, eleven tasks, and five random seeds per task, the proposed method improves average error by 32.0% over standard baselines and by 9.6% over the best budget-matched strong single expert. It also achieves one win and ten ties against a hindsight reference based on large-budget single experts. These results suggest that, for adaptive PINN collocation, selecting the sampling strategy according to source geometry is more effective and more robust than searching for another universal sampler.

Recently, there has been a surge of interest in extending the success of large language models (LLMs) from texts to molecules. Most existing approaches adopt a graph neural network to represent a molecule as a series of node tokens for molecule-language alignment, which, however, have overlooked the inherent hierarchical structures in molecules. Notably, higher-order molecular structures contain rich semantics of functional groups, which encode crucial biochemical functionalities of the molecules. We show that neglecting the hierarchical information in tokenization will lead to subpar molecule-language alignment and severe hallucination. To address this limitation, we propose HIerarchical GrapH Tokenization (HIGHT). HIGHT employs a hierarchical graph tokenizer that encodes the hierarchy of atom, motif, and molecular levels of informative tokens to improve the molecular perception of LLMs. HIGHT also adopts an augmented instruction tuning dataset, enriched with the hierarchical graph information, to further enhance the molecule-language alignment. Extensive experiments on 14 real-world benchmarks verify the effectiveness of HIGHT in reducing hallucination by 40%, and significant improvements in various molecule-language downstream tasks. The project is available at https: //higraphllm.github.io/.

Abstract Software engineers play a centric role throughout the software development lifecycle. Their activities directly impact the quality, performance, and successful delivery of software products, in particular for enterprises with an emphasis on high levels of quality assurance and timely delivery. Proper incentives that motivate software engineers are vital to secure and continuously improve development productivity and software quality. However, most existing research ignores the positive incentives for software engineers, especially industry‐oriented research. In addition, existing research largely relies on peer assessment and lacks objectivity and transparency. To this end, this study investigates the process of contribution measurement for software engineers in a global Information and Communications Technology (ICT) enterprise, to explore the practical experiences and significance of contribution measurement. We investigated the practices of contribution measurement through multiple methods, including archival analysis, interviews, and survey. A total of 22 software engineers were interviewed to understand the practical implementation process of measuring contributions and its impact on software processes as well as engineers. In addition, 74 responses to our questionnaire were collected and used for a comprehensive impact analysis on software engineers. The analysis results reveal five benefits for software development processes and four benefits for practitioners of contribution measurement in the studied enterprise. In addition, this study reports on the best practices of contribution measurement, such as team‐specific measurements, and provides a practical reference for researchers and organizations interested in studying or performing contribution measurement.

Taking photographs through windows is an inevitable scenario in the real world, but glass windows are not ideally clean in most cases. Although there exists various raindrop removal methods, the occlusion of dirt, as another dirty window case, has not been well valued. The vital reasons include i) the limitation of the optical imaging model proposed in previous methods, and ii) the shortage of a practical dataset for sufficient types of dirty glass windows. To fill this research gap, in this paper, we first propose a general optical imaging model that fits widely used dirty window cases. Following this, training and testing synthetic datasets are generated, and real-world dirty window data are collected to evaluate the effectiveness of our imaging model and synthetic data. For the methodology part, we propose an optics-guided Transformer network to solve this special image restoration problem, i.e., the dirt removal for images taken through a dirty window. Experimental results demonstrate that our imaging model is effective and robust. Our proposed network leads to higher performance than existing methods on both synthetic and real-world dirty window images. Code and data are available at https://github.com/Zongliang-Wu/ReDNet.

Link prediction is a fundamental task in graph learning, inherently shaped by the topology of the graph. While traditional heuristics are grounded in graph topology, they encounter challenges in generalizing across diverse graphs. Recent research efforts have aimed to leverage the potential of heuristics, yet a unified formulation accommodating both local and global heuristics remains undiscovered. Drawing insights from the fact that both local and global heuristics can be represented by adjacency matrix multiplications, we propose a unified matrix formulation to accommodate and generalize various heuristics. We further propose the Heuristic Learning Graph Neural Network (HL-GNN) to efficiently implement the formulation. HL-GNN adopts intra-layer propagation and inter-layer connections, allowing it to reach a depth of around 20 layers with lower time complexity than GCN. Extensive experiments on the Planetoid, Amazon, and OGB datasets underscore the effectiveness and efficiency of HL-GNN. It outperforms existing methods by a large margin in prediction performance. Additionally, HL-GNN is several orders of magnitude faster than heuristic-inspired methods while requiring only a few trainable parameters. The case study further demonstrates that the generalized heuristics and learned weights are highly interpretable.

In the past decades, downstream manufacturers often failed to timely adopt the security patches, resulting in some discovered vulnerabilities still posing serious risks. In the currently popular field of blockchain smart contracts, this is also a thorny issue. Although some new methods have been proposed to update and patch smart contracts deployed in blockchain networks, the binary codes of most vulnerable smart contracts are still being executed without patching. To detect the unpatched binaries as soon as possible, signature based patch presence tests and software similarity based patch presence tests have been proposed to check whether a certain patch is applied to the released software binaries. However, a large number of bug-fix patches are irrelevant to functions. They are small in size and only modify program entities other than functions. Existing signature-based patch detection methods and software similarity-based tools have limitations in detecting such patches. In this paper, we propose PPTFI, a patch presence test for function-irrelevant patches. PPTFI understands these patches and extracts code and data information as patch signatures for scanning target binaries. Being evaluated on 62 different versions of 31 real-world function-irrelevant patches and 512 binaries across 16 various compilation environments, PPTFI achieves an accuracy of 77.54%, significantly outperforming existing techniques.

Subgraph-based methods have proven to be effective and interpretable in predicting drug-drug interactions (DDIs), which are essential for medical practice and drug development. Subgraph selection and encoding are critical stages in these methods, yet customizing these components remains underexplored due to the high cost of manual adjustments. In this study, inspired by the success of neural architecture search (NAS), we propose a method to search for data-specific components within subgraph-based frameworks. Specifically, we introduce extensive subgraph selection and encoding spaces that account for the diverse contexts of drug interactions in DDI prediction. To address the challenge of large search spaces and high sampling costs, we design a relaxation mechanism that uses an approximation strategy to efficiently explore optimal subgraph configurations. This approach allows for robust exploration of the search space. Extensive experiments demonstrate the effectiveness and superiority of the proposed method, with the discovered subgraphs and encoding functions highlighting the model’s adaptability.

The success of a software organization relies heavily on the quality of its products and services, which in turn are influenced by the knowledge, capability, and experience of the software engineers involved in development processes. It is popular to apply quantitative assessments of software engineers for quality assurance. However, the extent to which it benefits software organizations and how it can be effectively implemented in industrial settings remains unclear. One global Information and Communications Technology (ICT) enterprise has implemented a quantitative assessment practice of software engineer’s outputs to improve its engineering capability and product and service quality. To investigate the benefits and experiences of adopting this practice in industrial settings, we conducted an empirical study using a mixed-method approach (i.e., archive analysis, interviews, and surveys). The results indicate that this practice can benefit the ICT enterprise in terms of standardizing development processes, optimizing team structures, and offering suggestions for training and management, etc. Meanwhile, this paper reports on the best practices to tackle the challenges during the adoption of the practice in the ICT enterprise, e.g., customization for teams and synergy of quantitative and qualitative assessment. In addition, we discuss the implications and recommendations of institutionalizing quantitative engineer assessment in software organizations. For organizations intending to improve software quality from the human aspect, this study provides empirical references on how to implement quantitative engineer assessment meanwhile mitigate potential risks.

In the coming era of social companions, many researches have been pursuing natural dialog interactions and long-term relations between social companions and users. With respect to the quick decrease of user interests after the first few interactions, various emotion and memory models are developed and integrated with social companions for better user engagement. This paper reviews related works in the effort of combining memory and emotion with natural language dialog on social companions. We separate these works into three categories: (1) Affective system with dialog, (2) Task-driven memory with dialog, (3) Chat-driven memory with dialog. In addition, we discussed limitations and challenging issues to be solved. Finally, we also introduced our framework of social companions.

Abstract How to endow the virtual characters with personalized behavior patterns remains a challenging problem. Instead of heuristically designing behaviors for certain personalities, this paper bridges the gap between personalities and behaviors using a medium concept, mood, to make the behaviors of the characters consistent enough to convey their personalities, while flexible enough to make appropriate response in various situations. We propose a personality‐characterized mood dynamics model, in which the emotion weights are computed as a solution of a convex optimization problem that is constructed to make the overall mood approaches the personality after sufficient interactions. The convergence of the algorithm is demonstrated by numerical simulations. The implementation of the personality‐characterized mood dynamics model enables an emotion‐oriented virtual human, Sophie, to show personalized behaviors in the emotional interactions with users. Copyright © 2015 John Wiley & Sons, Ltd.

Integrated Gasification Combined Cycle with embedded membrane reactor modules (IGCC-MR) represents a new technology option for the co-production of electricity and pure hydrogen endowed with enhanced environmental performance capacity. It is an alternative to conventional coaland gas-fired power generation technologies. As a new technology, the IGCC-MR power plant needs to be evaluated in the presence of irreducible regulatory and fuel market uncertainties for the potential deployment of an initial fleet of demonstration plants at the commercial scale. This paper presents the development of a systematic and comprehensive three-step methodological framework to assess the economic value of flexible alternatives in the design and operations of an IGCC-MR plant under the aforementioned sources of uncertainty. The main objective is to demonstrate the potential value enhancements stemming to the long-term economic performance of flexible IGCC-MR project investments, by managing the uncertainty associated with future environmental regulations and fuel costs. The paper provides an overview of promising design flexibility concepts for IGCC-MR power plants and focuses on operational and constructional flexibility. The operational flexibility is realized through the option of a temporary shutdown of the plant with considerations of regulatory and market uncertainties. This option reduces the probability of loss and the downside risk compared to the base case. The constructional flexibility considers installation of a Carbon Capture and Storage (CCS) unit in the plant under three different alternatives: 1) installing CCS in the initial construction phase, 2) retrofitting CCS at a later stage and 3) retrofitting CCS with pre-investment at a later stage. Monte Carlo simulations and financial analysis are used to demonstrate that the most economically advantageous flexibility option is to install CCS in the initial IGCC-MR construction phase.