Yifan Zhang, Chen Huang, Yueke Zhang, Jiahao Zhang, Toby Jia-Jun Li, Collin McMillan, Kevin Leach, Yu Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Code summarization is the task of generating natural language descriptions of source code, which is critical for software comprehension and maintenance. While large language models (LLMs) have achieved remarkable progress on this task, an open question remains: can human expertise in code understanding further guide and enhance these models? We propose EyeLayer, a lightweight attention-augmentation module that incorporates human eye-gaze patterns, as a proxy of human expertise, into LLM-based code summarization. EyeLayer models human attention during code reading via a Multimodal Gaussian Mixture, redistributing token embeddings based on learned parameters \((\mu _i, \sigma _i^2)\) that capture where and how intensively developers focus. This design enables learning generalizable attention priors from eye-tracking data and incorporating them into LLMs seamlessly, without disturbing existing representations. We evaluate EyeLayer across diverse model families (i.e., LLaMA-3.2, Qwen3, and CodeBERT) covering different scales and architectures. EyeLayer consistently outperforms strong fine-tuning baselines across standard metrics, achieving gains of up to 13.17% on BLEU-4. These results demonstrate that human gaze patterns encode complementary attention signals that enhance the semantic focus of LLMs and transfer effectively across diverse models for code summarization.

Xinwei Yang, Zhaofeng Liu, Chen Huang, Jiashuai Zhang, Tong Zhang, Yifan Zhang, Wenqiang Lei. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Recent growth and proliferation of malware have tested practitioners’ ability to promptly classify new samples according to malware families. In contrast to labor-intensive reverse engineering efforts, machine learning approaches have demonstrated increased speed and accuracy. However, most existing deep-learning malware family classifiers must be calibrated using a large number of samples that are painstakingly manually analyzed before training. Furthermore, as novel malware samples arise that are beyond the scope of the training set, additional reverse engineering effort must be employed to update the training set. The sheer volume of new samples found in the wild creates substantial pressure on practitioners’ ability to reverse engineer enough malware to adequately train modern classifiers.In this paper, we present MALMIXER, a malware family classifier using semi-supervised learning that achieves high accuracy with sparse training data. We present a domain-knowledge-aware data augmentation technique for malware feature representations, enhancing few-shot performance of semi-supervised malware family classification. We show that MALMIXER achieves state-of-the-art performance in few-shot malware family classification settings. Our research confirms the feasibility and effectiveness of lightweight, domain-knowledge-aware data augmentation methods for malware features and shows the capabilities of similar semi-supervised classifiers in addressing malware classification issues.

Optimizing software performance through automated code refinement offers a promising avenue for enhancing execution speed and efficiency. Despite recent advancements in LLMs, a significant gap remains in their ability to perform indepth program analysis. This study introduces AutoPatch, an in-context learning approach designed to bridge this gap by enabling LLMs to automatically generate optimized code. Inspired by how programmers learn and apply knowledge to optimize software, AutoPatch incorporates three key components: (1) an analogy-driven framework to align LLM optimization with human cognitive processes, (2) a unified approach that integrates historical code examples and CFG analysis for context-aware learning, and (3) an automated pipeline for generating optimized code through in-context prompting. Experimental results demonstrate that AutoPatch achieves a$\mathbf{7. 3 \%}$improvement in execution efficiency over GPT-4o across common generated executable code, highlighting its potential to advance automated program runtime optimization.

Recent language models have demonstrated proficiency in summarizing source code. However, as in many other domains of machine learning, language models of code lack sufficient explainability --- informally, we lack a formulaic or intuitive understanding of what and how models learn from code. Explainability of language models can be partially provided if, as the models learn to produce higher-quality code summaries, they also align in deeming the same code parts important as those identified by human programmers. In this paper, we report negative results from our investigation of explainability of language models in code summarization through the lens of human comprehension. We measure human focus on code using eye-tracking metrics such as fixation counts and duration in code summarization tasks. To approximate language model focus, we employ a state-of-the-art model-agnostic, black-box, perturbation-based approach, SHAP (SHapley Additive exPlanations), to identify which code tokens influence that generation of summaries. Using these settings, we find no statistically significant relationship between language models' focus and human programmers' attention. Furthermore, alignment between model and human foci in this setting does not seem to dictate the quality of the LLM-generated summaries. Our study highlights an inability to align human focus with SHAP-based model focus measures. This result calls for future investigation of multiple open questions for explainable language models for code summarization and software engineering tasks in general, including the training mechanisms of language models for code, whether there is an alignment between human and model attention on code, whether human attention can improve the development of language models, and what other model focus measures are appropriate for improving explainability.

The vanilla Graph Convolutional Network (GCN) uses a low-pass filter to extract low-frequency signals from graph topology, which may lead to the over-smoothing problem when GCN goes deep. To this end, various methods have been proposed to create an adaptive filter by incorporating an extra filter (e.g., a high-pass filter) extracted from the graph topology. However, these methods heavily rely on topological information and ignore the node attribute space, which severely sacrifices the expressive power of the deep GCNs, especially when dealing with disassortative graphs. In this paper, we propose a cross-space adaptive filter, called CSF, to produce the adaptive-frequency information extracted from both the topology and attribute spaces. Specifically, we first derive a tailored attribute-based high-pass filter that can be interpreted theoretically as a minimizer for semi-supervised kernel ridge regression. Then, we cast the topology-based low-pass filter as a Mercer's kernel within the context of GCNs. This serves as a foundation for combining it with the attribute-based filter to capture the adaptive-frequency information. Finally, we derive the cross-space filter via an effective multiple-kernel learning strategy, which unifies the attribute-based high-pass filter and the topology-based low-pass filter. This helps to address the over-smoothing problem while maintaining effectiveness. Extensive experiments demonstrate that CSF not only successfully alleviates the over-smoothing problem but also promotes the effectiveness of the node classification task. Our code is available at https://github.com/huangzichun/Cross-Space-Adaptive-Filter.

Code summarization is the task of creating short, natural language descriptions of source code. It is an important part of code comprehension and a powerful method of documentation. Previous work has made progress in identifying where programmers focus in code as they write their own summaries (i.e., Writing). However, there is currently a gap in studying programmers’ attention as they read code with pre-written summaries (i.e., Reading). As a result, it is currently unknown how these two forms of code comprehension compare: Reading and Writing. Also, there is a limited understanding of programmer attention with respect to program semantics. We address these shortcomings with a human eye-tracking study ( n = 27) comparing Reading and Writing. We examined programmers’ attention with respect to fine-grained program semantics, including their attention sequences (i.e., scan paths). We find distinctions in programmer attention across the comprehension tasks, similarities in reading patterns between them, and differences mediated by demographic factors. This can help guide code comprehension in both computer science education and automated code summarization. Furthermore, we mapped programmers’ gaze data onto the Abstract Syntax Tree to explore another representation of human attention. We find that visual behavior on this structure is not always consistent with that on source code.

This paper launches a new effort at modeling programmer attention by predicting eye movement scanpaths. Programmer attention refers to what information people intake when performing programming tasks. Models of programmer attention refer to machine prediction of what information is important to people. Models of programmer attention are important because they help researchers build better interfaces, assistive technologies, and more human-like AI. For many years, researchers in SE have built these models based on features such as mouse clicks, key logging, and IDE interactions. Yet the holy grail in this area is scanpath prediction - the prediction of the sequence of eye fixations a person would take over a visual stimulus. A person's eye movements are considered the most concrete evidence that a person is taking in a piece of information. Scanpath prediction is a notoriously difficult problem, but we believe that the emergence of lower-cost, higheraccuracy eye tracking equipment and better large language models of source code brings a solution within grasp. We present an eye tracking experiment with 27 programmers and a prototype scanpath predictor to present preliminary results and obtain early community feedback.

Anomaly detection (AD) aims to determine if an instance has properties different from those seen in normal cases. The success of this technique depends on how well a neural network learns from normal instances. We observe that the learning difficulty scales exponentially with the input resolution, making it infeasible to apply AD to high-resolution images. Resizing them to a lower resolution is a compromising solution and does not align with clinical practice where the diagnosis could depend on image details. In this work, we propose to train the network and perform inference at the patch level, through the sliding window algorithm. This simple operation allows the network to receive high-resolution images but introduces additional training difficulties, including inconsistent image structure and higher variance. We address these concerns by setting the network's objective to learn augmentation-invariant features. We further study the augmentation function in the context of medical imaging. In particular, we observe that the resizing operation, a key augmentation in general computer vision literature, is detrimental to detection accuracy, and the inverting operation can be beneficial. We also propose a new module that encourages the network to learn from adjacent patches to boost detection performance. Extensive experiments are conducted on breast tomosynthesis and chest X-ray datasets and our method improves 8.03% and 5.66% AUC on image-level classification respectively over the current leading techniques. The experimental results demonstrate the effectiveness of our approach.

Understanding binary code is an essential but complex software engineering task for reverse engineering, malware analysis, and compiler optimization. Unlike source code, binary code has limited semantic information, which makes it challenging for human comprehension. At the same time, compiling source to binary code, or transpiling among different programming languages (PLs) can provide a way to introduce external knowledge into binary comprehension. We propose to develop Artificial Intelligence (AI) models that aid human comprehension of binary code. Specifically, we propose to incorporate domain knowledge from large corpora of source code (e.g., variable names, comments) to build AI models that capture a generalizable representation of binary code. Lastly, we will investigate metrics to assess the performance of models that apply to binary code by using human studies of comprehension.