Recent advances in model compression have highlighted the potential of low-bit precision techniques, with Binary Neural Networks (BNNs) attracting attention for their extreme efficiency. However, extreme quantization in BNNs limits representational capacity and destabilizes training, posing significant challenges for lightweight architectures with depth-wise convolutions. To address this, we propose a 1.58-bit convolution to enhance expressiveness and a pre-BN residual connection to stabilize optimization by improving the Hessian condition number. These innovations enable, to the best of our knowledge, the first successful binarization of depth-wise convolutions in BNNs. Our method achieves 33M OPs on ImageNet with MobileNet V1, establishing a new state-of-the-art in BNNs by outperforming prior methods with comparable OPs. Moreover, it consistently outperforms existing methods across various datasets, including CIFAR-10, CIFAR-100, STL-10, Tiny ImageNet, and Oxford Flowers 102, with accuracy improvements of up to 9.3 percentage points.

As Large Language Models (LLMs) have become capable of generating long and descriptive code summaries, accurate and reliable evaluation of factual consistency has become a critical challenge.However, previous evaluation methods are primarily designed for short summaries of isolated code snippets.Consequently, they struggle to provide fine-grained evaluation of multi-sentence functionalities and fail to accurately assess dependency context commonly found in real-world code summaries.To address this, we propose ReFEree, a referencefree and fine-grained method for evaluating factual consistency in real-world code summaries.We define factual inconsistency criteria specific to code summaries and evaluate them at the segment level using these criteria along with dependency information.These segment-level results are then aggregated into a fine-grained score.We construct a code summarization benchmark with human-annotated factual consistency labels.The evaluation results demonstrate that ReFEree achieves the highest correlation with human judgment among 13 baselines, improving 15-18% over the previous state-of-the-art.Our code and data are available at https: //github.com/bsy99615/ReFEree.git.

Accurate and efficient Multivariate Time Series Forecasting (MTSF) plays a critical role in intelligent transportation systems by supporting real-time traffic management. However, achieving reliable forecasting remains challenging due to complex and dynamically evolving spatial–temporal patterns. Existing forecasting methods often fail to adapt effectively to these dynamic traffic conditions and typically incur high computational costs, significantly limiting their deployment in real-time traffic management scenarios. To address these engineering challenges, this study proposes a novel Attention-based Spatial-Temporal Network (ASTNet), explicitly designed for adaptive and efficient real-time traffic forecasting. ASTNet introduces two innovative Artificial Intelligence (AI)-driven modules: an Adaptive Spatial Graph Encoder (ASGE), which dynamically models evolving spatial dependencies from real-time traffic data, thus overcoming the limitations of static graph structures; and a Temporal Attention-Gated Unit (TAGU), which efficiently captures critical temporal dependencies through the integration of recurrent gating mechanisms and self-attention techniques. Extensive evaluations conducted on widely-used traffic benchmark datasets (PEMS04, METR-LA, etc.) confirm that ASTNet achieves superior predictive accuracy and robustness compared to state-of-the-art methods, while significantly reducing inference latency. Ablation studies further validate that the combined innovations of ASGE and TAGU are crucial for ASTNet’s outstanding performance, highlighting its practical suitability and strong potential for deployment in real-time intelligent transportation applications.

Self-supervised learning (SSL) methods based on the instance discrimination tasks with InfoNCE have achieved remarkable success. Despite their success, SSL models often struggle to generate effective representations for unseen-domain data. To address this issue, research on unsupervised domain generalization (UDG), which aims to develop SSL models that can generate domain-irrelevant features, has been conducted. Most UDG approaches utilize contrastive learning with InfoNCE to generate representations, and perform feature alignment based on strong assumptions to generalize domain-irrelevant common features from multi-source domains. However, existing methods that rely on instance discrimination tasks are not effective at extracting domain-irrelevant common features. This leads to the suppression of domain-irrelevant common features and the amplification of domain-relevant features, thereby hindering domain generalization. Furthermore, strong assumptions underlying feature alignment can lead to biased feature learning, reducing the diversity of common features. In this paper, we propose a novel approach, DomCLP, Domain-wise Contrastive Learning with Prototype Mixup. We explore how InfoNCE suppresses domain-irrelevant common features and amplifies domain-relevant features. Based on this analysis, we propose Domain-wise Contrastive Learning (DCon) to enhance domain-irrelevant common features. We also propose Prototype Mixup Learning (PMix) to generalize domain-irrelevant common features across multiple domains without relying on strong assumptions. The proposed method consistently outperforms state-of-the-art methods on the PACS and DomainNet datasets across various label fractions, showing significant improvements.

Text-to-SQL, which translates a natural language question into an SQL query, has advanced with in-context learning of Large Language Models (LLMs). However, existing methods show little improvement in performance compared to randomly chosen demonstrations, and significant performance drops when smaller LLMs (e.g., Llama 3.1-8B) are used. This indicates that these methods heavily rely on the intrinsic capabilities of hyper-scaled LLMs, rather than effectively retrieving useful demonstrations. In this paper, we propose a novel approach for effectively retrieving demonstrations and generating SQL queries. We construct a Deep Contextual Schema Link Graph, which contains key information and semantic relationship between a question and its database schema items. This graph-based structure enables effective representation of Text-to-SQL samples and retrieval of useful demonstrations for in-context learning. Experimental results on the Spider benchmark demonstrate the effectiveness of our approach, showing consistent improvements in SQL generation performance and efficiency across both hyper-scaled LLMs and small LLMs. The code is available at https://github.com/jjklle/DCG-SQL.

Suyoung Bae, YunSeok Choi, Hyojun Kim, Jee-Hyong Lee. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Suyoung Bae, YunSeok Choi, Jee-Hyong Lee. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

YunSeok Choi, CheolWon Na, Jee-Hyong Lee. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Weakly-supervised Semantic Segmentation (WSSS) aims to provide a precise semantic segmentation results without expensive pixel-wise segmentation labels. With the supervision gap between classification and segmentation, Image-level WSSS mainly relies on Class Activation Maps (CAMs) from the classification model to emulate the pixel-wise annotations. However, CAMs often fail to cover the entire object region because classification models tend to focus on narrow discriminative regions in an object. Towards accurate CAM coverage, Existing WSSS methods have tried to boost feature representation learning or impose consistency regularization to the classification models, but still there are limitation in activating non-discriminative area, where the focus of the models is weak. To tackle this issue, we propose FSAE framework, which provides explicit supervision of non-discriminative area, encouraging the CAMs to activate on various object features. We leverage weak-strong consistency with pseudo-label expansion strategy for reliable supervision and enhance learning of non-discriminative object boundaries. Specifically, we use strong perturbation to make challenging inference target, and focus on generating reliable pixel-wise supervision signal for broad object regions. Extensive experiments on the WSSS benchmark datasets show that our method boosts initial seed quality and segmentation performance by large margin, achieving new state-of-the-art performance on benchmark WSSS datasets. Our public code is available at https://github.com/obeychoi0120/FSAE.

In memory designs that adopt a full-custom design, accurately predicting the layout size of a circuit block is crucial for reducing design iterations. However, predicting the layout size is challenging due to the complex space sizes caused by wiring and layout-dependent effects between circuit elements. To address the challenge, we propose LayNet, a novel graph neural network model that predicts the layout size by constructing a weighted graph. We convert a circuit into a weighted graph to model the relationships between circuit elements. By applying graph neural networks to the weighted circuit graph, we can accurately predict the layout size. We also propose the edge selection and hierarchical graph learning approaches to reduce memory usage and inference time for large circuit blocks. LayNet achieves state-of-the-art performance on 6300 pairs of circuits and layouts in industrial memory products. Specifically, it significantly reduces the mean absolute percentage error rate by 20.82%∼88.17% for manually-generated layouts and by 7.97%∼73.39% for semiauto-generated layouts, outperforming conventional approaches. Also, the edge selection and hierarchical graph learning approaches reduce memory usage by 140. 85x and 238. 10x for these two types of layouts, respectively, and inference time by 14. 14x and 37. 84x, respectively, while maintaining performance.

Recently, code Pre-trained Language Models (PLMs) trained on large amounts of code and comment, have shown great success in code defect detection tasks. However, most PLMs simply treated the code as a single sequence and only used the encoder of PLMs to determine if there exist defects in the entire code. For a more analyzable and explainable approach, it is crucial to identify which lines contain defects. In this paper, we propose a novel method for code defect detection that integrates line-level defect localization into a unified training process. To identify code defects at the line-level, we convert the code into a sequence separated by lines using a special token. Then, to utilize the characteristic that both the encoder and decoder of PLMs process information differently, we leverage both the encoder and decoder for line-level defect localization. By learning code defect detection and line-level defect localization tasks in a unified manner, our proposed method promotes knowledge sharing between the two tasks. We demonstrate that our proposed method significantly improves performance on four benchmark datasets for code defect detection. Additionally, we show that our method can be easily integrated with ChatGPT.

Pre-trained language models (PLMs) are widely used for various tasks, but fine-tuning them requires sufficient data. Data augmentation approaches have been proposed as alternatives, but they vary in complexity, cost, and performance. To address these challenges, we propose STAGE (Simple Text Data Augmentation by Graph Exploration), a highly effective method for data augmentation. STAGE utilizes simple modification operations such as insertion, deletion, replacement, and swap. However, what distinguishes STAGE lies in the selection of optimal words for each modification. This is achieved by leveraging a word-relation graph called the co-graph. The co-graph takes into account both word frequency and co-occurrence, providing valuable information for operand selection. To assess the performance of STAGE, we conduct evaluations using seven representative datasets and three different PLMs. Our results demonstrate the effectiveness of STAGE across diverse data domains, varying data sizes, and different PLMs. Also, STAGE demonstrates superior performance when compared to previous methods that use simple modification operations or large language models like GPT3.

Deep Neural Networks (DNNs) have demonstrated remarkable performance across diverse domains and tasks with large-scale datasets. To reduce labeling costs for large-scale datasets, semi-automated and crowdsourcing labeling methods are developed, but their labels are in-evitably noisy. Learning with Noisy Labels (LNL) approaches aim to train DNNs despite the presence of noisy labels. These approaches utilize the memorization effect to select correct labels and refine noisy ones, which are then used for subsequent training. However, these methods en-counter a significant decrease in the model's generalization performance due to the inevitably existing noise labels. To overcome this limitation, we propose a new approach to enhance learning with noisy labels by incorporating additional distribution informationstructural labels. In order to leverage additional distribution information for generalization, we employ a reverse k-NN, which helps the model in achieving a better feature manifold and mitigating over-fitting to noisy labels. The proposed method shows outperformed performance in multiple benchmark datasets with IDN and real-world noisy datasets.

Automatic processing of source code, such as code clone detection and software vulnerability detection, is very helpful to software engineers.Large pre-trained Programming Language (PL) models (such as CodeBERT, Graph-CodeBERT, CodeT5, etc.), show very powerful performance on these tasks.However, these PL models are vulnerable to adversarial examples that are generated with slight perturbation.Unlike natural language, an adversarial example of code must be semantic-preserving and compilable.Due to the requirements, it is hard to directly apply the existing attack methods for natural language models.In this paper, we propose DIP (Dead code Insertion based Blackbox Attack for Programming Language Model), a high-performance and efficient black-box attack method to generate adversarial examples using dead code insertion.We evaluate our proposed method on 9 victim downstream-task large code models.Our method outperforms the state-of-the-art black-box attack in both attack efficiency and attack quality, while generated adversarial examples are compiled preserving semantic functionality.

Deep learning models need to detect out-of-distribution (OOD) data in the inference stage because they are trained to estimate the train distribution and infer the data sampled from the distribution. Many methods have been proposed, but they have some limitations, such as requiring additional data, input processing, or high computational cost. Moreover, most methods have hyperparameters to be set by users, which have a significant impact on the detection rate. We propose a simple and effective OOD detection method by combining the feature norm and the Mahalanobis distance obtained from classification models trained with the cosine-based softmax loss. Our method is practical because it does not use additional data for training, is about three times faster when inferencing than the methods using the input processing, and is easy to apply because it does not have any hyperparameters for OOD detection. We confirm that our method is superior to or at least comparable to state-of-the-art OOD detection methods through the experiments.

Session-based recommendation aims to predict the user's next action based on anonymous sessions without using side information. Most of the real-world session datasets are sparse and have long-tail item distribution. Although long-tail item recommendation plays a crucial role in improving user satisfaction, only a few methods have been proposed to take the long-tail session recommendation into consideration. Previous works in handling data sparsity problems are mostly limited to self-supervised learning techniques with heuristic augmentation which can ruin the original characteristic of session datasets, sequential and co-occurrences, and make noisier short sessions by dropping items and cropping sequences. We propose a novel method, LOAM, improving LOng-tail session-based recommendation via niche walk Augmentation and tail session Mixup, that alleviates popularity bias and enhances long-tail recommendation performance. LOAM consists of two modules, Niche Walk Augmentation (NWA) and Tail Session Mixup (TSM). NWA can generate synthetic sessions considering long-tail distribution which are likely to be found in original datasets, unlike previous heuristic methods, and expose a recommender model to various item transitions with global information. This improves the item coverage of recommendations. TSM makes the model more generalized and robust by interpolating sessions at the representation level. It encourages the recommender system to predict niche items with more diversity and relevance. We conduct extensive experiments with four real-world datasets and verify that our methods greatly improve tail performance while balancing overall performance.

Semi-supervised learning (SSL) is a method to make better models using a large number of easily accessible unlabeled data along with a small number of labeled data obtained at a high cost. Most of existing SSL studies focus on the cases where sufficient amount of labeled samples are available, tens to hundreds labeled samples for each class, which still requires a lot of labeling cost. In this paper, we focus on SSL environment with extremely scarce labeled samples, only 1 or 2 labeled samples per class, where most of existing methods fail to learn. We propose a propagation regularizer which can achieve efficient and effective learning with extremely scarce labeled samples by suppressing confirmation bias. In addition, for the realistic model selection in the absence of the validation dataset, we also propose a model selection method based on our propagation regularizer. The proposed methods show 70.9%, 30.3%, and 78.9% accuracy on CIFAR-10, CIFAR-100, SVHN dataset with just one labeled sample per class, which are improved by 8.9% to 120.2% compared to the existing approaches. And our proposed methods also show good performance on a higher resolution dataset, STL-10.

As pre-trained models have shown successful performance in program language processing as well as natural language processing, adversarial attacks on these models also attract attention.However, previous works on blackbox adversarial attacks generated adversarial examples in a very inefficient way with simple greedy search.They also failed to find out better adversarial examples because it was hard to reduce the search space without performance loss.In this paper, we propose TABS, an efficient beam search black-box adversarial attack method.We adopt beam search to find out better adversarial examples, and contextual semantic filtering to effectively reduce the search space.Contextual semantic filtering reduces the number of candidate adversarial words considering the surrounding context and the semantic similarity.Our proposed method shows good performance in terms of attack success rate, the number of queries, and semantic similarity in attacking models for two tasks: NL code search classification and retrieval tasks.

Summary Abstractive summarization is the task of creating summaries by generating a set of novel sentences based on the information extracted from the original document, while most of summarization researches are based on extractive or compressive approaches. These approaches extract phrases from the original document and concatenate them by post‐processing and cannot truly encapsulate the contents of summaries, because they only reuse the phrases in the given document. Moreover, there are limits for paraphrasing and re‐organizing of the original contents with the current Natural Language Processing (NLP) techniques. With these reasons, we propose a novel abstractive summarization method. The main goal of our paper is to generate a long sequence of words with coherent sentences by reflecting the key concepts of the original document and the contents of summaries. To achieve this goal, we propose an attention mechanism that uses Document Content Memory for learning the language model effectively. To evaluate its effectiveness, the proposed methods are compared with other language models and an extractive summarization method. We demonstrate that our proposed methods improve summarization results in ROUGE score using ACL dataset. The experimental results show that our proposed methods using keyword memory are effective to generate long sequence summary.

As semiconductor micro-fabrication process continues to advance, the size of probing pads also become smaller in a chip. A probe needle contacts each probing pad for electrical test. However, probe needle may incorrectly touch probing pad. Such contact failures damage probing pads and cause qualification problems. In order to detect contact failures, the current system observes the probing marks on pads. Due to a low accuracy of the system, engineers have to redundantly verify the result of the system once more, which causes low efficiency. We suggest an approach for automatic defect detection to solve these problems using image processing and CSVM. We develop significant features of probing marks to classify contact failures more correctly. We reduce 38% of the workload of engineers.

Semiconductor is a major component of electronic devices and is required very high reliability and productivity. If defective chip predict in advance, the product quality will be improved and productivity will increases by reduction of test cost. However, the performance of the classifiers about defective chips is very poor due to semiconductor data is extremely imbalance, as roughly 1:1000. In this paper, the iterative undersampling method using CSVM is employed to deal with the class imbalanced. The main idea is to select the informative majority class samples around the decision boundary determined by classify. Our experimental results are reported to demonstrate that our method outperforms the other sampling methods in regard with the accuracy of defective chip in highly imbalanced data.

In the semiconductor manufacturing process, it is important to predict defective chips in advance for reduction of test cost and early stabilization of the production process. However, highly imbalanced datasets in the semiconductor test process degrade the performance of prediction. In order to enhance an SVM Ensemble, this study presents an improved methodology using the K-means, which clusters the majority class and the minority class before training an SVM. A result of the experiment with the actual data of the semiconductor test process is reported to demonstrate that our approach outperforms other methods in terms of classifying the imbalanced dataset.

Large scale, automatic fingerprint identification systems (AFISs) perform fingerprint classification to improve matching accuracy and reduce the matching time before fingerprint matching. Fingerprints are classified into several classes such as arch (A), whorl (W), left loop (L) and right loop (L). The existing systems generally classify fingerprints based on the information of singular points. This approach is well suited for fingerprints acquired using paper and ink. However, it is not as efficient with recent automatic fingerprint systems because it cannot guarantee that singular points are well extracted since the recent systems have various sized sensors and use multifarious fingerprint acquisition methods. In this paper, a novel approach is proposed to use the fingerprint ridge direction, which is one of the global features. It is a probabilistic approach based on the fingerprint ridge characteristics of each class. FVC2000 DB1 and FVC2002 DB1 databases were used to evaluate the performance of our classification. Furthermore, the effectiveness of applying the probabilistic model to the classification of various exceptional fingerprint patterns was verified.

In these days, Linux system is widely used because of its freedom to use and develop. With this trend, to find vulnerabilities in Linux kernel has become more important. Linux kernel is so huge that we need a machine based error detecting approach. There are some former studies about error detection in software code. However, they are not suitable for detecting unknown vulnerabilities related to Linux kernel variables. We suggest a vulnerability recommendation system for Linux kernel variables. First, we propose a methodology by analyzing 368 reported vulnerabilities in Linux kernel. We focus on two elements to find vulnerabilities in Linux kernel variables. Those are the kernel variables which are concerned about privilege escalating and the system call tree information that shows which system calls may modify which kernel variables. We tested our recommendation system with two representative Linux versions. Through experiments, we confirm that our system can find potential vulnerabilities including known ones.

Web page evaluation is an important issue in the Internet. The page view count is a widely used criterion for the Web page evaluation because of its easiness. But, the evaluation methods based on the page view count cannot reflect whether the Web page content corresponds with userspsila needs because users click a page after looking at only the title or the small part of the page. If the page content does not satisfy a user, the user generally does not spend much time nor take any actions to look at the page so therefore we developed an Ajax log system. Using this system, we collect userspsila visiting time and action on Web pages such as clicks, scrolling, etc. Users are not interrupted while Ajax works. But the collected data are continuous values. We cannot determine adaptive criteria to each user data. To solve this problem, the evaluation module of the system is based on the neural network. The system with neural network learns userspsila action pattern while reading useful Web pages and evaluates the usefulness of Web pages from userspsila actions. Our system can more accurately find pages which satisfy users than a search engine.

Linux kernel provides several advantages to system developers and is widely used as an operating system in a variety of systems, including embedded systems, access routers and servers. These advantages are due to the fact that the Linux kernel is publicly available, however, this feature of openness can have negative impacts on system security. If an attacker wished to exploit Linux-based systems, the attacker could easily do so by finding and abusing the vulnerabilities of the systemspsila Linux kernel sources. There are several methods available that can find source-level vulnerabilities, but they are not always suitable for the Linux kernel. In this paper, we propose a two-step Onion mechanism as a methodology to find source-level vulnerabilities of the Linux kernel variables. The first step of the Onion mechanism is to select variables that may be vulnerable by exploiting their usage patterns. The second step is to inspect the vulnerabilities of the selected variables by making and analyzing system call trees. We also evaluate our proposed methodology by applying it to two well-known source-level vulnerabilities.

A learning diagnosis system collects data from a learner's learning process, and analyzes it to build a suitable model for the learner, which can then be incorporated into an intelligent tutoring system to provide customized tutoring services. However, if the collected data reflects inconsistent learner behaviors or unpredictable learning tendencies, then the reliability of the learner model is degraded. In this paper, the outliers in the learner's data are eliminated by a k-NN method. We apply this method to an experimental data set obtained using DOLLS-HI, a learner diagnosis system that uses housing interior learning contents to diagnose learning styles. The resulting diagnosis model shows improved reliability than before eliminating the outliers.

A large number of people download music files easily from Web sites. But rare music sites provide personalized services. So, we suggest a method for personalized services. We extract the properties of music from music's sound wave. We use STFT (shortest time fourier form) to analyze music's property. And we infer users' preferences from users' music list. To analyze users' preferences we propose a dynamic K-means clustering algorithm. The dynamic K-means clustering algorithm clusters the pieces in the music list dynamically adapting the number of clusters. We recommend pieces of music based on the clusters. The previous recommendation systems analyze a user's preference by simply averaging the properties of music in the user's list. So those cannot recommend correctly if a user prefers several genres of music. By using our K-means clustering algorithm, we can recommend pieces of music which are close to user's preference even though he likes several genres. We perform experiments with one hundred pieces of music. In this paper we present and evaluate algorithms to recommend music.

Ranking fuzzy numbers is one of the very important research topics in fuzzy set theory because it is a base of decision-making in all application areas. However, fuzzy numbers cannot be easily arranged in order of magnitude because they represent uncertain and vague values. When two fuzzy numbers overlap with each other, a fuzzy number may not be considered absolutely larger than the other. That is, even though a fuzzy number may be considered larger than the other, it may also be considered smaller than the other. It means that, even when we consider only two fuzzy numbers, two sequences may be the ranking result at the same time. However, most of the existing ranking methods produce only one ranking sequence. They ignore other possible sequences due to the overlap between fuzzy numbers. We propose a ranking method which generates possible ranking sequences of fuzzy numbers, and represents them with fuzzy sets. Some numeric examples are also presented to show how our method ranks fuzzy numbers.

Since fuzzy numbers represent uncertain numeric values, it is difficult to rank them according to their magnitude. In the paper, a method for ranking fuzzy numbers is proposed. The method considers the overall possibility distributions of fuzzy numbers in their evaluations for ranking and provides users with a method of changing viewpoints for evaluations. Users represent their viewpoints with fuzzy sets. The method evaluates fuzzy numbers with a satisfaction function and the viewpoint given by users and then ranks the numbers according to their evaluation values. The satisfaction function is a measure of comparisons between fuzzy numbers. In order to illustrate the ranking method, two numeric examples are shown, and for the comparative study, our method is compared with four existing ranking methods through eight examples. As an example of potential applications, the proposed method is applied to a decision-making problem: a two-person game with fuzzy profit and loss. The ranking method is used to analyze player choices.

This paper presents a fuzzy traffic controller for a set of intersections and its simulation results. The controller of an intersection controls its own traffic and cooperates with its neighbors. It gets information from its traffic detectors and its neighbors. Using this information, the fuzzy rule base system gives optimal signals. It manages phase sequences and phase lengths adaptively to its neighbors' as well as its own traffic conditions. To carry out the performance evaluation of the controller, a simulator for intersection groups has been developed. The proposed method is compared with the vehicle actuated method which is one of the typical conventional methods. The average delay time of a vehicle is used as a performance index. The simulation results show good performance in the case of time-varying traffic patterns and heavy traffic conditions.