arXiv subjects

Seungwon Lee

Publications and source records attributed to Seungwon Lee.

At least 19 recordsLinked to original sources

arXiv2026

On the Asymptotics of Item Selection in Multidimensional Computerized Adaptive Testing

We study Fisher-information-based item-selection rules for multidimensional computerized adaptive testing (MCAT) with intentional and nuisance abilities. Although such rules are widely used, rigorous asymptotic theory for the coupled dynamics of adaptive selection and latent-trait estimation remains limited. We consider a weighted A-optimality criterion that prioritizes intentional abilities while controlling accuracy in nuisance dimensions. For the multidimensional two-parameter logistic model, we first establish asymptotic optimality under an item-type reuse regime: paired with the maximum likelihood estimator, the weighted A-optimal rule attains the minimum asymptotic weighted mean squared error. We then prove consistency and asymptotic normality of the maximum likelihood estimator and express its asymptotic weighted mean squared error through a limiting Fisher information matrix. We also extend the analysis to the operational setting in which each item is administered at most once. These results justify normal approximations and information-based standard errors in MCAT. A simulation study examines finite-sample behavior at practical test lengths and shows favorable weighted mean squared error, with item-selection patterns consistent with the predicted limiting behavior.

math.ST↗

arXiv2026

PIM-SHERPA: Software Method for On-device LLM Inference by Resolving PIM Memory Attribute and Layout Inconsistencies

On-device deployments of large language models (LLMs) are rapidly proliferating across mobile and edge platforms. LLM inference comprises a compute-intensive prefill phase and a memory bandwidth-intensive decode phase, and the decode phase has been widely recognized as well-suited to processing-in-memory (PIM) in both academia and industry. However, practical PIM-enabled systems face two obstacles between these phases, a memory attribute inconsistency in which prefill favors placing weights in a cacheable region for reuse whereas decode requires weights in a non-cacheable region to reliably trigger PIM, and a weight layout inconsistency between host-friendly and PIM-aware layouts. To address these problems, we introduce \textit{PIM-SHERPA}, a software-only method for efficient on-device LLM inference by resolving PIM memory attribute and layout inconsistencies. PIM-SHERPA provides two approaches, DRAM double buffering (DDB), which keeps a single PIM-aware weights in the non-cacheable region while prefetching the swizzled weights of the next layer into small cacheable buffers, and online weight rearrangement with swizzled memory copy (OWR), which performs the on-demand swizzled memory copy immediately before GEMM. Compared to a baseline PIM emulation system, PIM-SHERPA achieves approximately 47.8 - 49.7\% memory capacity savings while maintaining comparable performance to the theoretical maximum on the Llama 3.2 model. To the best of our knowledge, this is the first work to identify the memory attribute inconsistency and propose effective solutions on product-level PIM-enabled systems.

cs.DC↗

arXiv2025

Sequential Change Point Detection with FDR Control in Reconfigurable Sensor Networks

This paper investigates sequential change-point detection in reconfigurable sensor networks. In this problem, data from multiple sensors are observed sequentially. Each sensor can have a unique change point, and the data distribution differs before and after the change. We aim to detect these changes as quickly as possible once they have occurred while controlling the false discovery rate at all times. Our setting is more realistic than traditional settings in that (1) the set of active sensors - i.e., those from which data can be collected - can change over time through the deactivation of existing sensors and the addition of new sensors, and (2) dependencies can occur both between sensors and across time points. We propose powerful e-value-based detection procedures that control the false discovery rate uniformly over time. Numerical experiments demonstrate that, with the same false discovery rate target, our procedures achieve superior performance compared to existing methods, exhibiting lower false non-discovery rates and reduced detection delays.

stat.ME↗

arXiv2025

Assessing Validity of ICD-10 Administrative Data in Coding Comorbidities

Objectives: Administrative data is commonly used to inform chronic disease prevalence and support health informatics research. This study assessed the validity of coding comorbidity in the International Classification of Diseases, 10th Revision (ICD-10) administrative data. Methods: We analyzed three chart review cohorts (4,008 patients in 2003, 3,045 in 2015, and 9,024 in 2022) in Alberta, Canada. Nurse reviewers assessed the presence of 17 clinical conditions using a consistent protocol. The reviews were linked with administrative data using unique identifiers. We compared the accuracy in coding comorbidity by ICD-10, using chart review data as the reference standard. Results: Our findings showed that the mean difference in prevalence between chart reviews and ICD-10 for these 17 conditions was 2.1% in 2003, 7.6% in 2015, and 6.3% in 2022. Some conditions were relatively stable, such as diabetes (1.9%, 2.1%, and 1.1%) and metastatic cancer (0.3%, 1.1%, and 0.4%). For these 17 conditions, the sensitivity ranged from 39.6-85.1% in 2003, 1.3-85.2% in 2015, and 3.0-89.7% in 2022. The C-statistics for predicting in-hospital mortality using comorbidities by ICD-10 were 0.84 in 2003, 0.81 in 2015, and 0.78 in 2022. Discussion: The under-coding could be primarily due to the increase of hospital patient volumes and the limited time allocated to coders. There is a potential to develop artificial intelligence methods based on electronic health records to support coding practices and improve coding quality. Conclusion: Comorbidities were increasingly under-coded over 20 years. The validity of ICD-10 decreased but remained relatively stable for certain conditions mandated for coding. The under-coding exerted minimal impact on in-hospital mortality prediction.

q-bio.QM↗

arXiv2025

Integrating Large Language Models with Human Expertise for Disease Detection in Electronic Health Records

Objective: Electronic health records (EHR) are widely available to complement administrative data-based disease surveillance and healthcare performance evaluation. Defining conditions from EHR is labour-intensive and requires extensive manual labelling of disease outcomes. This study developed an efficient strategy based on advanced large language models to identify multiple conditions from EHR clinical notes. Methods: We linked a cardiac registry cohort in 2015 with an EHR system in Alberta, Canada. We developed a pipeline that leveraged a generative large language model (LLM) to analyze, understand, and interpret EHR notes by prompts based on specific diagnosis, treatment management, and clinical guidelines. The pipeline was applied to detect acute myocardial infarction (AMI), diabetes, and hypertension. The performance was compared against clinician-validated diagnoses as the reference standard and widely adopted International Classification of Diseases (ICD) codes-based methods. Results: The study cohort accounted for 3,088 patients and 551,095 clinical notes. The prevalence was 55.4%, 27.7%, 65.9% and for AMI, diabetes, and hypertension, respectively. The performance of the LLM-based pipeline for detecting conditions varied: AMI had 88% sensitivity, 63% specificity, and 77% positive predictive value (PPV); diabetes had 91% sensitivity, 86% specificity, and 71% PPV; and hypertension had 94% sensitivity, 32% specificity, and 72% PPV. Compared with ICD codes, the LLM-based method demonstrated improved sensitivity and negative predictive value across all conditions. The monthly percentage trends from the detected cases by LLM and reference standard showed consistent patterns.

cs.CL↗

arXiv2024

Learning Service Selection Decision Making Behaviors During Scientific Workflow Development

Increasingly, more software services have been published onto the Internet, making it a big challenge to recommend services in the process of a scientific workflow composition. In this paper, a novel context-aware approach is proposed to recommending next services in a workflow development process, through learning service representation and service selection decision making behaviors from workflow provenance. Inspired by natural language sentence generation, the composition process of a scientific workflow is formalized as a step-wise procedure within the context of the goal of workflow, and the problem of next service recommendation is mapped to next word prediction. Historical service dependencies are first extracted from scientific workflow provenance to build a knowledge graph. Service sequences are then generated based on diverse composition path generation strategies. Afterwards, the generated corpus of composition paths are leveraged to study previous decision making strategies. Such a trained goal-oriented next service prediction model will be used to recommend top K candidate services during workflow composition process. Extensive experiments on a real-word repository have demonstrated the effectiveness of this approach.

cs.SE↗

arXiv2024

Breaking MLPerf Training: A Case Study on Optimizing BERT

Speeding up the large-scale distributed training is challenging in that it requires improving various components of training including load balancing, communication, optimizers, etc. We present novel approaches for fast large-scale training of BERT model which individually ameliorates each component thereby leading to a new level of BERT training performance. Load balancing is imperative in distributed BERT training since its training datasets are characterized by samples with various lengths. Communication cost, which is proportional to the scale of distributed training, needs to be hidden by useful computation. In addition, the optimizers, e.g., ADAM, LAMB, etc., need to be carefully re-evaluated in the context of large-scale distributed training. We propose two new ideas, (1) local presorting based on dataset stratification for load balancing and (2) bucket-wise gradient clipping before allreduce which allows us to benefit from the overlap of gradient computation and synchronization as well as the fast training of gradient clipping before allreduce. We also re-evaluate existing optimizers via hyperparameter optimization and utilize ADAM, which also contributes to fast training via larger batches than existing methods. Our proposed methods, all combined, give the fastest MLPerf BERT training of 25.1 (22.3) seconds on 1,024 NVIDIA A100 GPUs, which is 1.33x (1.13x) and 1.57x faster than the other top two (one) submissions to MLPerf v1.1 (v2.0). Our implementation and evaluation results are available at MLPerf v1.1~v2.1.

cs.LG↗

arXiv2023

A Domain-Agnostic Approach for Characterization of Lifelong Learning Systems

Despite the advancement of machine learning techniques in recent years, state-of-the-art systems lack robustness to "real world" events, where the input distributions and tasks encountered by the deployed systems will not be limited to the original training context, and systems will instead need to adapt to novel distributions and tasks while deployed. This critical gap may be addressed through the development of "Lifelong Learning" systems that are capable of 1) Continuous Learning, 2) Transfer and Adaptation, and 3) Scalability. Unfortunately, efforts to improve these capabilities are typically treated as distinct areas of research that are assessed independently, without regard to the impact of each separate capability on other aspects of the system. We instead propose a holistic approach, using a suite of metrics and an evaluation framework to assess Lifelong Learning in a principled way that is agnostic to specific domains or system techniques. Through five case studies, we show that this suite of metrics can inform the development of varied and complex Lifelong Learning systems. We highlight how the proposed suite of metrics quantifies performance trade-offs present during Lifelong Learning system development - both the widely discussed Stability-Plasticity dilemma and the newly proposed relationship between Sample Efficient and Robust Learning. Further, we make recommendations for the formulation and use of metrics to guide the continuing development of Lifelong Learning systems and assess their progress in the future.

cs.LG↗

arXiv2022

Goal-Driven Context-Aware Next Service Recommendation for Mashup Composition

As service-oriented architecture becoming one of the most prevalent techniques to rapidly deliver functionalities to customers, increasingly more reusable software components have been published online in forms of web services. To create a mashup, it gets not only time-consuming but also error-prone for developers to find suitable services from such a sea of services. Service discovery and recommendation has thus attracted significant momentum in both academia and industry. This paper proposes a novel incremental recommend-as-you-go approach to recommending next potential service based on the context of a mashup under construction, considering services that have been selected to the current step as well as its mashup goal. The core technique is an algorithm of learning the embedding of services, which learns their past goal-driven context-aware decision making behaviors in addition to their semantic descriptions and co-occurrence history. A goal exclusionary negative sampling mechanism tailored for mashup development is also developed to improve training performance. Extensive experiments on a real-world dataset demonstrate the effectiveness of our approach.

cs.SE↗

arXiv2022

CO2-driven surface changes in the Hapi region on Comet 67P/Churyumov-Gerasimenko

Between 2014 December 31 and 2015 March 17, the OSIRIS cameras on Rosetta documented the growth of a 140m wide and 0.5m deep depression in the Hapi region on Comet 67P/Churyumov-Gerasimenko. This shallow pit is one of several that later formed elsewhere on the comet, all in smooth terrain that primarily is the result of airfall of coma particles. We have compiled observations of this region in Hapi by the microwave instrument MIRO on Rosetta, acquired during October and November 2014. We use thermophysical and radiative transfer models in order to reproduce the MIRO observations. This allows us to place constraints on the thermal inertia, diffusivity, chemical composition, stratification, extinction coefficients, and scattering properties of the surface material, and how they evolved during the months prior to pit formation. The results are placed in context through long-term comet nucleus evolution modelling. We propose that: 1) MIRO observes signatures that are consistent with a solid-state greenhouse effect in airfall material; 2) CO2 ice is sufficiently close to the surface to have a measurable effect on MIRO antenna temperatures, and likely is responsible for the pit formation in Hapi observed by OSIRIS; 3) the pressure at the CO2 sublimation front is sufficiently strong to expel dust and water ice outwards, and to compress comet material inwards, thereby causing the near-surface compaction observed by CONSERT, SESAME, and groundbased radar, manifested as the "consolidated terrain" texture observed by OSIRIS.

astro-ph.EP↗

arXiv2022

Learning Context-Aware Service Representation for Service Recommendation in Workflow Composition

As increasingly more software services have been published onto the Internet, it remains a significant challenge to recommend suitable services to facilitate scientific workflow composition. This paper proposes a novel NLP-inspired approach to recommending services throughout a workflow development process, based on incrementally learning latent service representation from workflow provenance. A workflow composition process is formalized as a step-wise, context-aware service generation procedure, which is mapped to next-word prediction in a natural language sentence. Historical service dependencies are extracted from workflow provenance to build and enrich a knowledge graph. Each path in the knowledge graph reflects a scenario in a data analytics experiment, which is analogous to a sentence in a conversation. All paths are thus formalized as composable service sequences and are mined, using various patterns, from the established knowledge graph to construct a corpus. Service embeddings are then learned by applying deep learning model from the NLP field. Extensive experiments on the real-world dataset demonstrate the effectiveness and efficiency of the approach.

cs.SE↗

arXiv2021

Spatial Distribution of Ultraviolet Emission from Cometary Activity at 67P/Churyumov-Gerasimenko

The Alice ultraviolet spectrograph on board the \textit{Rosetta} orbiter provided the first near-nucleus ultraviolet observations of a cometary coma from arrival at comet 67P/Churyumov-Gerasimenko in 2014 August through 2016 September. The characterization of atomic and molecular emissions in the coma revealed the unexpected contribution of dissociative electron impact emission at large heliocentric distances and during some outbursts. This mechanism also proved useful for compositional analysis, and Alice observed many cases that suggested elevated levels of the supervolatile \ce{O2}, identifiable in part to their emissions resulting from dissociative electron impact. In this paper we present the first two-dimensional UV maps constructed from Alice observations of atomic emission from 67P during an increase in cometary activity on 2015 November 7-8. Comparisons to observations of background coma and of an earlier collimated jet are used to describe possible changes to the near-nucleus coma and plasma. To verify the mapping method and place the Alice observations in context, comparisons to images derived from the MIRO and VIRTIS-H instruments are made. The spectra and maps we present show an increase in dissociative electron impact emission and an \ce{O2}/\ce{H2O} ratio of $\sim$0.3 for the activity; these characteristics have been previously identified with cometary outbursts seen in Alice data. Further, UV maps following the increases in activity show the spatial extent and emission variation experienced by the near-nucleus coma, informing future UV observations of comets that lack the same spatial resolution.

astro-ph.EP↗

arXiv2021

Data-free mixed-precision quantization using novel sensitivity metric

Post-training quantization is a representative technique for compressing neural networks, making them smaller and more efficient for deployment on edge devices. However, an inaccessible user dataset often makes it difficult to ensure the quality of the quantized neural network in practice. In addition, existing approaches may use a single uniform bit-width across the network, resulting in significant accuracy degradation at extremely low bit-widths. To utilize multiple bit-width, sensitivity metric plays a key role in balancing accuracy and compression. In this paper, we propose a novel sensitivity metric that considers the effect of quantization error on task loss and interaction with other layers. Moreover, we develop labeled data generation methods that are not dependent on a specific operation of the neural network. Our experiments show that the proposed metric better represents quantization sensitivity, and generated data are more feasible to be applied to mixed-precision quantization.

cs.LG↗

arXiv2019

Revisiting Classical Bagging with Modern Transfer Learning for On-the-fly Disaster Damage Detector

Automatic post-disaster damage detection using aerial imagery is crucial for quick assessment of damage caused by disaster and development of a recovery plan. The main problem preventing us from creating an applicable model in practice is that damaged (positive) examples we are trying to detect are much harder to obtain than undamaged (negative) examples, especially in short time. In this paper, we revisit the classical bootstrap aggregating approach in the context of modern transfer learning for data-efficient disaster damage detection. Unlike previous classical ensemble learning articles, our work points out the effectiveness of simple bagging in deep transfer learning that has been underestimated in the context of imbalanced classification. Benchmark results on the AIST Building Change Detection dataset show that our approach significantly outperforms existing methodologies, including the recently proposed disentanglement learning.

cs.LG↗

arXiv2018

Training Deep Neural Network in Limited Precision

Energy and resource efficient training of DNNs will greatly extend the applications of deep learning. However, there are three major obstacles which mandate accurate calculation in high precision. In this paper, we tackle two of them related to the loss of gradients during parameter update and backpropagation through a softmax nonlinearity layer in low precision training. We implemented SGD with Kahan summation by employing an additional parameter to virtually extend the bit-width of the parameters for a reliable parameter update. We also proposed a simple guideline to help select the appropriate bit-width for the last FC layer followed by a softmax nonlinearity layer. It determines the lower bound of the required bit-width based on the class size of the dataset. Extensive experiments on various network architectures and benchmarks verifies the effectiveness of the proposed technique for low precision training.

cs.NE↗

arXiv2018

Quantization for Rapid Deployment of Deep Neural Networks

This paper aims at rapid deployment of the state-of-the-art deep neural networks (DNNs) to energy efficient accelerators without time-consuming fine tuning or the availability of the full datasets. Converting DNNs in full precision to limited precision is essential in taking advantage of the accelerators with reduced memory footprint and computation power. However, such a task is not trivial since it often requires the full training and validation datasets for profiling the network statistics and fine tuning the networks to recover the accuracy lost after quantization. To address these issues, we propose a simple method recognizing channel-level distribution to reduce the quantization-induced accuracy loss and minimize the required image samples for profiling. We evaluated our method on eleven networks trained on the ImageNet classification benchmark and a network trained on the Pascal VOC object detection benchmark. The results prove that the networks can be quantized into 8-bit integer precision without fine tuning.

cs.NE↗

arXiv2006

Magnetic-field dependence of valley splitting for Si quantum wells grown on tilted SiGe substrates

The valley splitting of the first few Landau levels is calculated as a function of the magnetic field for electrons confined in a strained silicon quantum well grown on a tilted SiGe substrate, using a parameterized tight-binding method. For a zero substrate tilt angle, the valley splitting slightly decreases with increasing magnetic field. In contrast, the valley splitting for a finite substrate tilt angle exhibits a strong and non-monotonous dependence on the magnetic field strength. The valley splitting of the first Landau level shows an exponential increase followed by a slow saturation as the magnetic field strength increases. The valley splitting of the second and third Landau levels shows an oscillatory behavior. The non-monotonous dependence is explained by the phase variation of the Landau level wave function along the washboard-like interface between the tilted quantum well and the buffer material. The phase variation is the direct consequence of the misorientation between the crystal axis and the confinement direction of the quantum well. This result suggests that the magnitude of the valley splitting can be tuned by controlling the Landau-level filling factor through the magnetic field and the doping concentration.

cond-mat.mtrl-sci↗

arXiv2006

Zero valley splitting at zero magnetic field for strained Si/SiGe quantum wells grown on tilted substrates

The electronic structure for a strained Si/SiGe quantum well grown on a tilted substrate with periodic steps is calculated using a parameterized tight-binding method. For a zero tilt angle the energy difference between the two lowest minima of the conduction band at the center of the Brillouin zone defines a non-zero valley splitting. At finite tilt angles, the two lowest conduction band minima shift to k0 and -k0 in the Brillouin zone and have equal energy. The valley splitting for quantum wells grown on a tilted substrate is therefore equal to zero, which is a direct consequence of the periodicity of the steps at the interfaces between the quantum well and the buffer materials.

cond-mat.mtrl-sci↗