Time Series forecasting (TSF) in the modern era faces significant computational and storage cost challenges due to the massive scale of real-world data. Dataset Distillation (DD), a paradigm that synthesizes a small, compact dataset to achieve training performance comparable to that of the original dataset, has emerged as a promising solution. However, conventional DD methods are not tailored for time series and suffer from architectural overfitting and limited scalability. To address these issues, we propose Harmonic Dataset Distillation for Time Series Forecasting (HDT). HDT decomposes the time series into its sinusoidal basis through the FFT and aligns the core periodic structure by Harmonic Matching. Since this process operates in the frequency domain, all updates during distillation are applied globally without disrupting temporal dependencies of time series. Extensive experiments demonstrate that HDT achieves strong cross-architecture generalization and scalability, validating its practicality for large-scale, real-world applications.

Online Time Series Forecasting (OTSF) requires models to continuously adapt to concept drift. However, existing methods often treat concept drift as a monolithic phenomenon. To address this limitation, we first redefine concept drift by categorizing it into two distinct types: Recurring Drift, where previously seen patterns reappear, and Emergent Drift, where entirely new patterns emerge. We then propose DynaME (Dynamic Multi-period Experts), a novel hybrid framework designed to effectively address this dual nature of drift. For Recurring Drift, DynaME employs a committee of specialized experts that are dynamically fitted to the most relevant historical periodic patterns at each time step. For Emergent Drift, the framework detects high-uncertainty scenarios and shifts reliance to a stable, general expert. Extensive experiments on several benchmark datasets and backbones demonstrate that DynaME effectively adapts to both concept drifts and significantly outperforms existing baselines. Our source code is available at https://github.com/shhong97/DynaME.

This paper proposes an uncertainty quantification (UQ)-incorporated design optimization technique that integrates UQ considering the statistical characteristics of sensing margin with machine learning (ML) for the design of bitline sense amplifiers (BLSAs) in dynamic random access memory (DRAM), including offset compensation operation. The increased offset due to process variation exacerbated by device scaling degrades the reliability of DRAM read operations, making quantitative analysis of process variation essential. Conventional Monte Carlo (MC)-based UQ analysis provides high accuracy but requires extensive simulation costs. Accordingly, this study systematically defines key input and output variables for conventional BLSA and two representative offset compensation BLSA structures, introduces Sobol sequence-based quasi-Monte Carlo (QMC) sampling to improve simulation efficiency, and generalizes the Standard UQ methodology to predict statistical distributions of circuit performance parameters under process variation. Furthermore, a 3-Stage UQ algorithm that sequentially learns nominal, mean, and standard deviation values is proposed to enhance the efficiency of ML model training. HSPICE simulation results based on the TSMC 65 nm process demonstrated that the proposed 3-Stage UQ method reduced the simulation time required for ML training data generation by an average of 43% compared to the Standard UQ method and achieved excellent prediction performance even in data-limited environments. Additionally, design optimization combined with Bayesian Optimization (BO) confirmed performance improvements across all circuit structures. The methodology presented in this study enables effective process variation-aware optimization for DRAM designs requiring high reliability and yield, and is expected to contribute to next-generation memory circuit design automation.

Graph Neural Networks (GNNs) have achieved state-of-the-art performance in node classification tasks but struggle with label noise in real-world data.Existing studies on graph learning with label noise commonly rely on class-dependent label noise, overlooking the complexities of instance-dependent noise and falling short of capturing real-world corruption patterns.We introduce BeGIN (Benchmarking for Graphs with Instance-dependent Noise), a new benchmark that provides realistic graph datasets with various noise types and comprehensively evaluates noise-handling strategies across GNN architectures, noisy label detection, and noise-robust learning.To simulate instance-dependent corruptions, BeGIN introduces algorithmic methods and LLM-based simulations.Our experiments reveal the challenges of instance-dependent noise, particularly LLM-based corruption, and underscore the importance of node-specific parameterization to enhance GNN robustness.By comprehensively evaluating noise-handling strategies, BeGIN provides insights into their effectiveness, efficiency, and key performance factors.We expect that BeGIN will serve as a valuable resource for advancing research on label noise in graphs and fostering the development of robust GNN training methods.The code is available at https://github.com/kimsu55/BeGIN.

Label noise, commonly found in real-world datasets, has a detrimental impact on a model's generalization. To effectively detect incorrectly labeled instances, previous works have mostly relied on distinguishable training signals, such as training loss, as indicators to differentiate between clean and noisy labels. However, they have limitations in that the training signals incompletely reveal the model's behavior and are not effectively generalized to various noise types, resulting in limited detection accuracy. In this paper, we propose DynaCor framework that distinguishes incorrectly labeled instances from correctly labeled ones based on the dynamics of the training signals. To cope with the absence of supervision for clean and noisy labels, DynaCor first in-troduces a label corruption strategy that augments the original dataset with intentionally corrupted labels, enabling indirect simulation of the model's behavior on noisy labels. Then, DynaCor learns to identify clean and noisy instances by inducing two clearly distinguishable clusters from the latent representations of training dynamics. Our compre-hensive experiments show that DynaCor outperforms the state-of-the-art competitors and shows strong robustness to various noise types and noise rates.

Neurons have a polarized structure, with dendrites and axons, and compartment-specific functions can be affected by the dwelling mitochondria. Recent studies have shown that the morphology of mitochondria is closely related to the functions of neurons and neurodegenerative diseases. However, the conventional mitochondria analysis workflow mainly relies on manual annotations and generic image-processing software. Moreover, even though there have been recent developments in automatic mitochondria analysis using deep learning, the application of existing methods in a daily analysis remains challenging because the performance of a pretrained deep learning model can vary depending on the target data, and there are always errors in inference time, requiring human proofreading. To address these issues, we introduce MitoVis, a novel visualization system for end-to-end data processing and an interactive analysis of the morphology of neuronal mitochondria. MitoVis introduces a novel active learning framework based on recent contrastive learning, which allows accurate fine-tuning of the neural network model. MitoVis also provides novel visual guides for interactive proofreading so that users can quickly identify and correct errors in the result with minimal effort. We demonstrate the usefulness and efficacy of the system via case studies conducted by neuroscientists. The results show that MitoVis achieved up to 13.3× faster total analysis time in the case study compared to the conventional manual analysis workflow.

Recently, graph neural networks (GNNs) have been successfully applied to predicting molecular properties, which is one of the most classical cheminformatics tasks with various applications. Despite their effectiveness, we empirically observe that training a single GNN model for diverse molecules with distinct structural patterns limits its prediction performance. In this paper, motivated by this observation, we propose TopExpert to leverage topology-specific prediction models (referred to as experts), each of which is responsible for each molecular group sharing similar topological semantics. That is, each expert learns topology-specific discriminative features while being trained with its corresponding topological group. To tackle the key challenge of grouping molecules by their topological patterns, we introduce a clustering-based gating module that assigns an input molecule into one of the clusters and further optimizes the gating module with two different types of self-supervision: topological semantics induced by GNNs and molecular scaffolds, respectively. Extensive experiments demonstrate that TopExpert has boosted the performance for molecular property prediction and also achieved better generalization for new molecules with unseen scaffolds than baselines. The code is available at https://github.com/kimsu55/ToxExpert.

The advancement of artificial intelligence (AI) has drawn educators' attention to its educational potential. However, the efficacy of AI depends on learners' effective interaction, avoiding either neglect or excessive dependence. Despite the significance of learner-AI interaction, research on interaction patterns remains limited. This study investigates how EFL (English as a foreign language) learners interact with an AI tool for English writing and explores the effects of different interaction patterns on their writing performance. Through an experiment involving 29 EFL undergraduates, three distinct interaction patterns emerged, exhibiting significant differences in their engagement with the AI tool. The comparisons of the three clusters indicate that different AI interaction patterns lead to varied interaction approaches, and not all learners equally benefit from AI's potential. In order to promote productive learner-AI interaction in educational environments, instructors should provide personalized support and feedback.

With the recent advances in sequencing technology, researchers now have opportunities to study microbiomes associated with various environments. Recent studies have shown that the composition of microbiomes in our bodies and our environments play a significant role in our health. For example, 90% of human DNA is composed of bacterial microbiomes. In this study, we propose a systems biology approach using split graphs to analyze the composition of microbiomes and the impact of such composition on the health and growth of organisms living in associated environments. We focus on a case study related to the composition of microbiomes in fish guts and its impact on various growth parameters for three types of fish. The proposed model explores features in the aquatic ecosystem including correlations among its microorganisms and their abundance levels. The results of the study show that single or groups of bacteria are significantly associated with multiple growth phenotypes in different gut portions of the fish. We also identify bacterial clusters that provide new insight to functional relevance of these bacteria and their contribution to the fish gut microbial ecosystem.

To understand the current state of a discipline and to discover new knowledge of a certain theme, one builds bibliometric content networks based on the present knowledge entities. However, such networks can vary according to the collection of data sets relevant to the theme by querying knowledge entities. In this study we classify three different bibliometric content networks. The primary bibliometric network is based on knowledge entities relevant to a keyword of the theme, the secondary network is based on entities associated with the lower concepts of the keyword, and the tertiary network is based on entities influenced by the theme. To explore the content and properties of these networks, we propose a tomographic content analysis that takes a slice‐and‐dice approach to analyzing the networks. Our findings indicate that the primary network is best suited to understanding the current knowledge on a certain topic, whereas the secondary network is good at discovering new knowledge across fields associated with the topic, and the tertiary network is appropriate for outlining the current knowledge of the topic and relevant studies.

Journal rankings, frequently determined by the journal impact factor or similar indices, are quantitative measures for evaluating a journal's performance in its discipline, which is presently a major research thrust in the bibliometrics field. Recently, text mining was adopted to augment journal ranking‐based evaluation with the content analysis of a discipline taking a time‐variant factor into consideration. However, previous studies focused mainly on a silo analysis of a discipline using either citation‐or content‐oriented approaches, and no attempt was made to analyze topical journal ranking and its change over time in a seamless and integrated manner. To address this issue, we propose a journal‐time‐topic model, an extension of Dirichlet multinomial regression, which we applied to the field of bioinformatics to understand journal contribution to topics in a field and the shift of topic trends. The journal‐time‐topic model allows us to identify which journals are the major leaders in what topics and the manner in which their topical focus. It also helps reveal an interesting distinct pattern in the journal impact factor of high‐ and low‐ranked journals. The study results shed a new light for understanding topic specific journal rankings and shifts in journals' concentration on a subject.

Mobile data offloading through WiFi is an essential requirement to reduce cellular network traffic. While extensive attempts have been made at mobile data offloading, previous studies have rarely addressed practical issues, such as dealing with diverse user contexts. In this paper, we propose a personalized data offloading scheme to provide maximum throughput within the cellular budget in daily life. We propose an adaptive policy that considers a user's mobility patterns, cellular budget, and network usage for applications. The proposed system employs an adaptive model to predict the throughput of WiFi APs and the network usage of smartphones. Among the three types of predictor model (i.e., spatial, temporal, and spatio-temporal), the system automatically chooses the optimal model for each mobile user without user intervention. The experimental results from 10 mobile users show that the proposed system provides 29% higher throughput than previous systems and minimizes extra data charges.

BACKGROUND: Thyroid carcinomas are known to harbor oncogenic driver mutations and advances in sequencing technology now allow the detection of these in fine needle aspiration biopsies (FNA). Recent work by The Cancer Genome Atlas (TCGA) Research Network has expanded the number of genetic alterations detected in papillary thyroid carcinomas (PTC). We sought to investigate the prevalence of these and other genetic alterations in diverse subtypes of thyroid nodules beyond PTC, including a variety of samples with benign histopathology. This is the first clinical evaluation of a large panel of TCGA-reported genomic alterations in thyroid FNAs. RESULTS: In FNAs, genetic alterations were detected in 19/44 malignant samples (43% sensitivity) and in 7/44 histopathology benign samples (84% specificity). Overall, after adding a cohort of tissue samples, 38/76 (50%) of histopathology malignant samples were found to harbor a genetic alteration, while 15/75 (20%) of benign samples were also mutated. The most frequently mutated malignant subtypes were medullary thyroid carcinoma (9/12, 75%) and PTC (14/30, 47%). Additionally, follicular adenoma, a benign subtype of thyroid neoplasm, was also found to harbor mutations (12/29, 41%). Frequently mutated genes in malignant samples included BRAF (20/76, 26%) and RAS (9/76, 12%). Of the TSHR variants detected, (6/7, 86%) were in benign nodules. In a direct comparison of the same FNA also tested by an RNA-based gene expression classifier (GEC), the sensitivity of genetic alterations alone was 42%, compared to the 91% sensitivity achieved by the GEC. The specificity based only on genetic alterations was 84%, compared to 77% specificity with the GEC. CONCLUSIONS: While the genomic landscape of all thyroid neoplasm subtypes will inevitably be elucidated, caution should be used in the early adoption of published mutations as the sole predictor of malignancy in thyroid. The largest set of such mutations known to date detects only a portion of thyroid carcinomas in preoperative FNAs in our cohort and thus is not sufficient to rule out cancer. Due to the finding that variants are also found in benign nodules, testing only GEC suspicious nodules may be helpful in avoiding false positives and altering the extent of treatment when selected mutations are found.

Mobile sensing systems employ various sensors in smartphones to extract human-related information. As the demand for sensing systems increases, a more effective mechanism is required to sense information about human life. In this paper, we present a systematic study on the feasibility and gaining properties of a crowdsensing system that primarily concerns sensing WiFi packets in the air. We propose that this method is effective for estimating urban mobility by using only a small number of participants. During a seven-week deployment, we collected smartphone sensor data, including approximately four million WiFi packets from more than 130,000 unique devices in a city. Our analysis of this dataset examines core issues in urban mobility monitoring, including feasibility, spatio-temporal coverage, scalability, and threats to privacy. Collectively, our findings provide valuable insights to guide the development of new mobile sensing systems for urban life monitoring.

BACKGROUND: Accurate somatic mutation-calling is essential for insightful mutation analyses in cancer studies. Several mutation-callers are publicly available and more are likely to appear. Nonetheless, mutation-calling is still challenging and there is unlikely to be one established caller that systematically outperforms all others. Therefore, fully utilizing multiple callers can be a powerful way to construct a list of final calls for one's research. RESULTS: Using a set of mutations from multiple callers that are impartially validated, we present a statistical approach for building a combined caller, which can be applied to combine calls in a wider dataset generated using a similar protocol. Using the mutation outputs and the validation data from The Cancer Genome Atlas endometrial study (6,746 sites), we demonstrate how to build a statistical model that predicts the probability of each call being a somatic mutation, based on the detection status of multiple callers and a few associated features. CONCLUSION: The approach allows us to build a combined caller across the full range of stringency levels, which outperforms all of the individual callers.

Bioinformatics is a fast‐growing field based on the optimal use of “big data” gathered in genomic, proteomics, and functional genomics research. In this paper, we conduct a comprehensive and in‐depth bibliometric analysis of the field of bioinformatics by extracting citation data from PubMed Central full‐text. Citation data for the period 2000 to 2011, comprising 20,869 papers with 546,245 citations, was used to evaluate the productivity and influence of this emerging field. Four measures were used to identify productivity; most productive authors, most productive countries, most productive organizations, and most popular subject terms. Research impact was analyzed based on the measures of most cited papers, most cited authors, emerging stars, and leading organizations. Results show the overall trends between the periods 2000 to 2003 and 2004 to 2007 were dissimilar, while trends between the periods 2004 to 2007 and 2008 to 2011 were similar. In addition, the field of bioinformatics has undergone a significant shift, co‐evolving with other biomedical disciplines.

BACKGROUND: Somatic mutation-calling based on DNA from matched tumor-normal patient samples is one of the key tasks carried by many cancer genome projects. One such large-scale project is The Cancer Genome Atlas (TCGA), which is now routinely compiling catalogs of somatic mutations from hundreds of paired tumor-normal DNA exome-sequence data. Nonetheless, mutation calling is still very challenging. TCGA benchmark studies revealed that even relatively recent mutation callers from major centers showed substantial discrepancies. Evaluation of the mutation callers or understanding the sources of discrepancies is not straightforward, since for most tumor studies, validation data based on independent whole-exome DNA sequencing is not available, only partial validation data for a selected (ascertained) subset of sites. RESULTS: To provide guidelines to comparing outputs from multiple callers, we have analyzed two sets of mutation-calling data from the TCGA benchmark studies and their partial validation data. Various aspects of the mutation-calling outputs were explored to characterize the discrepancies in detail. To assess the performances of multiple callers, we introduce four approaches utilizing the external sequence data to varying degrees, ranging from having independent DNA-seq pairs, RNA-seq for tumor samples only, the original exome-seq pairs only, or none of those. CONCLUSIONS: Our analyses provide guidelines to visualizing and understanding the discrepancies among the outputs from multiple callers. Furthermore, applying the four evaluation approaches to the whole exome data, we illustrate the challenges and highlight the various circumstances that require extra caution in assessing the performances of multiple callers.

BACKGROUND: Estimation of allele frequency is of fundamental importance in population genetic analyses and in association mapping. In most studies using next-generation sequencing, a cost effective approach is to use medium or low-coverage data (e.g., < 15X). However, SNP calling and allele frequency estimation in such studies is associated with substantial statistical uncertainty because of varying coverage and high error rates. RESULTS: We evaluate a new maximum likelihood method for estimating allele frequencies in low and medium coverage next-generation sequencing data. The method is based on integrating over uncertainty in the data for each individual rather than first calling genotypes. This method can be applied to directly test for associations in case/control studies. We use simulations to compare the likelihood method to methods based on genotype calling, and show that the likelihood method outperforms the genotype calling methods in terms of: (1) accuracy of allele frequency estimation, (2) accuracy of the estimation of the distribution of allele frequencies across neutrally evolving sites, and (3) statistical power in association mapping studies. Using real re-sequencing data from 200 individuals obtained from an exon-capture experiment, we show that the patterns observed in the simulations are also found in real data. CONCLUSIONS: Overall, our results suggest that association mapping and estimation of allele frequencies should not be based on genotype calling in low to medium coverage data. Furthermore, if genotype calling methods are used, it is usually better not to filter genotypes based on the call confidence score.