A5086877012- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Adversarial Robustness in Machine Learning
- Generative Adversarial Networks and Image Synthesis
- Topic Modeling
- Anomaly Detection Techniques and Applications
- RNA and protein synthesis mechanisms
- Music and Audio Processing
- Natural Language Processing Techniques
- Advanced Memory and Neural Computing
- Genomics and Phylogenetic Studies
- Gene expression and cancer classification
- Machine Learning in Materials Science
- Speech Recognition and Synthesis
- Advanced Data Storage Technologies
- Machine Learning in Bioinformatics
- Ferroelectric and Negative Capacitance Devices
- Neural dynamics and brain function
- Computational Drug Discovery Methods
- Cancer-related molecular mechanisms research
- MicroRNA in disease regulation
- Bioinformatics and Genomic Networks
- Digital Media Forensic Detection
- Caching and Content Delivery
- Prostate Cancer Treatment and Research
Seoul National University
2016-2025
Seoul National University Hospital
2024
Seoul National University of Education
2024
Seoul Media Institute of Technology
2020-2021
Samsung (South Korea)
2019-2020
Institute of Engineering
2020
Stanford University
2004-2017
Seoul National University Bundang Hospital
2017
Jungwon University
2017
New Generation University College
2015
The main obstacle to weakly supervised semantic image segmentation is the difficulty of obtaining pixel-level information from coarse image-level annotations. Most methods based on annotations use localization maps obtained classifier, but these only focus small discriminative parts objects and do not capture precise boundaries. FickleNet explores diverse combinations locations feature created by generic deep neural networks. It selects hidden units randomly then uses them obtain activation...
Denoising diffusion probabilistic models (DDPM) have shown remarkable performance in unconditional image generation. However, due to the stochasticity of generative process DDPM, it is challenging generate images with desired semantics. In this work, we propose Iterative Latent Variable Refinement (ILVR), a method guide DDPM high-quality based on given reference image. Here, refinement enables single sample from various sets directed by The proposed ILVR generates while controlling...
As industries become automated and connectivity technologies advance, a wide range of systems continues to generate massive amounts data. Many approaches have been proposed extract principal indicators from the vast sea data represent entire system state. Detecting anomalies using these on time prevent potential accidents economic losses. Anomaly detection in multivariate series poses particular challenge because it requires simultaneous consideration temporal dependencies relationships...
Significance Self-assembling RNA molecules play critical roles throughout biology and bioengineering. To accelerate progress in design, we present EteRNA, the first internet-scale citizen science “game” scored by high-throughput experiments. A community of 37,000 nonexperts leveraged continuous remote laboratory feedback to learn new design rules that substantially improve experimental accuracy structure designs. These rules, distilled machine learning into a automated algorithm EteRNABot,...
Over the past decade, deep neural networks (DNNs) have demonstrated remarkable performance in a variety of applications. As we try to solve more advanced problems, increasing demands for computing and power resources has become inevitable. Spiking (SNNs) attracted widespread interest as third-generation due their event-driven low-powered nature. SNNs, however, are difficult train, mainly owing complex dynamics neurons non-differentiable spike operations. Furthermore, applications been...
Recently, text-to-speech (TTS) models such as FastSpeech and ParaNet have been proposed to generate mel-spectrograms from text in parallel. Despite the advantage, parallel TTS cannot be trained without guidance autoregressive their external aligners. In this work, we propose Glow-TTS, a flow-based generative model for that does not require any aligner. By combining properties of flows dynamic programming, searches most probable monotonic alignment between latent representation speech on its...
Weakly supervised semantic segmentation produces a pixel-level localization from class labels; but classifier trained on such labels is likely to restrict its focus small discriminative region of the target object. AdvCAM an attribution map image that manipulated increase classification score produced by classifier. This manipulation realized in anti-adversarial manner, which perturbs original images along pixel gradients opposite direction those used adversarial attack. It forces regions...
Abstract Background Quantitative structure-activity relationship (QSAR) is a computational modeling method for revealing relationships between structural properties of chemical compounds and biological activities. QSAR essential drug discovery, but it has many constraints. Ensemble-based machine learning approaches have been used to overcome constraints obtain reliable predictions. Ensemble builds set diversified models combines them. However, the most prevalent approach random forest other...
We evaluated SpCas9 activities at 12,832 target sequences using a high-throughput approach based on human cell library containing single-guide RNA-encoding and sequence pairs. Deep learning-based training this large dataset of SpCas9-induced indel frequencies led to the development activity-predicting model named DeepSpCas9. When tested against independently generated datasets (our own those published by other groups), DeepSpCas9 showed high generalization performance. is available...
Weakly supervised segmentation methods using bounding box annotations focus on obtaining a pixel-level mask from each containing an object. Existing typically depend class-agnostic generator, which operates the low-level information intrinsic to image. In this work, we utilize higher-level behavior of trained object detector, by seeking smallest areas image detector produces almost same result as it does whole These constitute bounding-box attribution map (BBAM), identifies target in its and...
Diffusion models learn to restore noisy data, which is corrupted with different levels of noise, by optimizing the weighted sum corresponding loss terms, i.e., denoising score matching loss. In this paper, we show that restoring data certain noise offers a proper pretext task for model rich visual concepts. We propose prioritize such over other during training, redesigning weighting scheme objective function. our simple redesign significantly improves performance diffusion regardless...
In recent years, proposed studies on time-series anomaly detection (TAD) report high F1 scores benchmark TAD datasets, giving the impression of clear improvements in TAD. However, most apply a peculiar evaluation protocol called point adjustment (PA) before scoring. this paper, we theoretically and experimentally reveal that PA has great possibility overestimating performance; even random score can easily turn into state-of-the-art method. Therefore, comparison methods after applying lead to...
Weakly supervised semantic segmentation (WSSS) methods are often built on pixel-level localization maps obtained from a classifier. However, training class labels only, classifiers suffer the spurious correlation between fore-ground and background cues (e.g. train rail), fundamentally bounding performance of WSSS. There have been previous endeavors to address this issue with additional supervision. We propose novel source information distinguish foreground background: Out-of-Distribution...
Long non-coding RNAs (lncRNAs) are important regulatory elements in biological processes. LncRNAs share similar sequence characteristics with messenger RNAs, but they play completely different roles, thus providing novel insights for studies. The development of next-generation sequencing has helped the discovery lncRNA transcripts. However, experimental verification numerous transcriptomes is time consuming and costly. To alleviate these issues, a computational approach needed to distinguish...
Electrocardiogram (ECG) signals from mobile sensors are expected to increase the availability of authentication in emerging wearable device industry. However, provide a relatively lower quality signal than conventional medical devices. This paper proposes practical procedure for ECG that collected via one-chip-solution sensors. We designed cascading bandpass filter noise cancellation and suggest eight fiducial features. For classification-based authentication, we use radial basis function...
Adversarial examples cause neural networks to produce incorrect outputs with high confidence. Although adversarial training is one of the most effective forms defense against examples, unfortunately, a large gap exists between test accuracy and in training. In this paper, we identify Feature Overfitting (AFO), which may poor adversarially robust generalization, show that can overshoot optimal point terms leading AFO our simple Gaussian model. Considering these theoretical results, present...