arXiv subjects

Wonseok Hwang

Publications and source records attributed to Wonseok Hwang.

27 records · Page 2Linked to original sources

arXiv2021

Cost-effective End-to-end Information Extraction for Semi-structured Document Images

A real-world information extraction (IE) system for semi-structured document images often involves a long pipeline of multiple modules, whose complexity dramatically increases its development and maintenance cost. One can instead consider an end-to-end model that directly maps the input to the target output and simplify the entire process. However, such generation approach is known to lead to unstable performance if not designed carefully. Here we present our recent effort on transitioning from our existing pipeline-based IE system to an end-to-end system focusing on practical challenges that are associated with replacing and deploying the system in real, large-scale production. By carefully formulating document IE as a sequence generation task, we show that a single end-to-end IE system can be built and still achieve competent performance.

cs.CL↗

arXiv2021

Spatial Dependency Parsing for Semi-Structured Document Information Extraction

Information Extraction (IE) for semi-structured document images is often approached as a sequence tagging problem by classifying each recognized input token into one of the IOB (Inside, Outside, and Beginning) categories. However, such problem setup has two inherent limitations that (1) it cannot easily handle complex spatial relationships and (2) it is not suitable for highly structured information, which are nevertheless frequently observed in real-world document images. To tackle these issues, we first formulate the IE task as spatial dependency parsing problem that focuses on the relationship among text tokens in the documents. Under this setup, we then propose SPADE (SPAtial DEpendency parser) that models highly complex spatial relationships and an arbitrary number of information layers in the documents in an end-to-end manner. We evaluate it on various kinds of documents such as receipts, name cards, forms, and invoices, and show that it achieves a similar or better performance compared to strong baselines including BERT-based IOB taggger.

cs.CL↗

arXiv2020

Tractable loss function and color image generation of multinary restricted Boltzmann machine

The restricted Boltzmann machine (RBM) is a representative generative model based on the concept of statistical mechanics. In spite of the strong merit of interpretability, unavailability of backpropagation makes it less competitive than other generative models. Here we derive differentiable loss functions for both binary and multinary RBMs. Then we demonstrate their learnability and performance by generating colored face images.

cs.CV↗

arXiv2020

Syntactic Question Abstraction and Retrieval for Data-Scarce Semantic Parsing

Deep learning approaches to semantic parsing require a large amount of labeled data, but annotating complex logical forms is costly. Here, we propose Syntactic Question Abstraction and Retrieval (SQAR), a method to build a neural semantic parser that translates a natural language (NL) query to a SQL logical form (LF) with less than 1,000 annotated examples. SQAR first retrieves a logical pattern from the train data by computing the similarity between NL queries and then grounds a lexical information on the retrieved pattern in order to generate the final LF. We validate SQAR by training models using various small subsets of WikiSQL train data achieving up to 4.9% higher LF accuracy compared to the previous state-of-the-art models on WikiSQL test set. We also show that by using query-similarity to retrieve logical pattern, SQAR can leverage a paraphrasing dataset achieving up to 5.9% higher LF accuracy compared to the case where SQAR is trained by using only WikiSQL data. In contrast to a simple pattern classification approach, SQAR can generate unseen logical patterns upon the addition of new examples without re-training the model. We also discuss an ideal way to create cost efficient and robust train datasets when the data distribution can be approximated under a data-hungry setting.

cs.CL↗

arXiv2019

A Comprehensive Exploration on WikiSQL with Table-Aware Word Contextualization

We present SQLova, the first Natural-language-to-SQL (NL2SQL) model to achieve human performance in WikiSQL dataset. We revisit and discuss diverse popular methods in NL2SQL literature, take a full advantage of BERT {Devlin et al., 2018) through an effective table contextualization method, and coherently combine them, outperforming the previous state of the art by 8.2% and 2.5% in logical form and execution accuracy, respectively. We particularly note that BERT with a seq2seq decoder leads to a poor performance in the task, indicating the importance of a careful design when using such large pretrained models. We also provide a comprehensive analysis on the dataset and our model, which can be helpful for designing future NL2SQL datsets and models. We especially show that our model's performance is near the upper bound in WikiSQL, where we observe that a large portion of the evaluation errors are due to wrong annotations, and our model is already exceeding human performance by 1.3% in execution accuracy.

cs.CL↗

arXiv2019

Energetic costs, precision, and efficiency of a biological motor in cargo transport

Molecular motors play pivotal roles in organizing the interior of cells. A motor efficient in cargo transport would move along cytoskeletal filaments with a high speed and a minimal error in transport distance (or time) while consuming a minimal amount of energy. The travel distance of the motor and its variance are, however, physically constrained by the free energy being consumed. A recently formulated \emph{thermodynamic uncertainty relation} offers a theoretical framework for the energy-accuracy trade-off relation ubiquitous in biological processes. According to the relation, a measure $\mathcal{Q}$, the product between the heat dissipated from a motor and the squared relative error in the displacement, has a minimal theoretical bound ($\mathcal{Q} \geq 2 k_B T$), which is approached when the time trajectory of the motor is maximally regular for a given amount of free energy input. Here, we use $\mathcal{Q}$ to quantify the transport efficiency of biological motors. Analyses on the motility data from several types of molecular motors reveal that $\mathcal{Q}$ is a complex function of ATP concentration and load ($f$). For kinesin-1, $\mathcal{Q}$ approaches the theoretical bound at $f\approx 4$ pN and over a broad range of ATP concentration (1 $μ$M - 10 mM), and is locally minimized at [ATP] $\approx$ 200 $μ$M. In stark contrast, this local minimum vanishes for a mutant that has a longer neck-linker, and the value of $\mathcal{Q}$ is significantly greater, which underscores the importance of molecular structure. Transport efficiencies of the biological motors studied here are semi-optimized under the cellular condition ([ATP] $\approx 1$ mM, $f=0-1$ pN). Our study indicates that among many possible directions of optimization, cytoskeletal motors are designed to operate at a high speed with a minimal error while leveraging their energy resources.

physics.bio-ph↗

arXiv2017

Physical insight into the thermodynamic uncertainty relation using Brownian motion in tilted periodic potentials

Using Brownian motion in periodic potentials $V(x)$ tilted by a force $f$, we provide physical insight into the thermodynamic uncertainty relation, a recently conjectured principle for statistical errors and irreversible heat dissipation in nonequilibrium steady states. According to the relation, nonequilibrium output generated from dissipative processes necessarily incurs an energetic cost or heat dissipation $q$, and in order to limit the output fluctuation within a relative uncertainty $ε$, at least $2k_BT/ε^2$ of heat must be dissipated. Our model shows that this bound is attained not only at near-equilibrium ($f\ll V'(x)$) but also at far-from-equilibrium $(f\gg V'(x))$, more generally when the dissipated heat is normally distributed. Furthermore, the energetic cost is maximized near the critical force when the barrier separating the potential wells is about to vanish and the fluctuation of Brownian particle is maximized. These findings indicate that the deviation of heat distribution from Gaussianity gives rise to the inequality of the uncertainty relation, further clarifying the meaning of the uncertainty relation. Our derivation of the uncertainty relation also recognizes a new bound of nonequilibrium fluctuations that the variance of dissipated heat ($σ_q^2$) increases with its mean ($μ_q$) and cannot be smaller than $2k_BTμ_q$.

cond-mat.stat-mech↗

arXiv2016

Quantifying the Heat Dissipation from a Molecular Motor's Transport Properties in Nonequilibrium Steady States

Theoretical analysis, which maps single molecule time trajectories of a molecular motor onto unicyclic Markov processes, allows us to evaluate the heat dissipated from the motor and to elucidate its dependence on the mean velocity and diffusivity. Unlike passive Brownian particles in equilibrium, the velocity and diffusion constant of molecular motors are closely inter-related to each other. In particular, our study makes it clear that the increase of diffusivity with the heat production is a natural outcome of active particles, which is reminiscent of the recent experimental premise that the diffusion of an exothermic enzyme is enhanced by the heat released from its own catalytic turnover. Compared with freely diffusing exothermic enzymes, kinesin-1 whose dynamics is confined on one-dimensional tracks is highly efficient in transforming conformational fluctuations into a locally directed motion, thus displaying a significantly higher enhancement in diffusivity with its turnover rate. Putting molecular motors and freely diffusing enzymes on an equal footing, our study offers thermodynamic basis to understand the heat enhanced self-diffusion of exothermic enzymes.

q-bio.SC↗

arXiv2016

Decoding Single Molecule Time Traces with Dynamic Disorder

Single molecule time trajectories of biomolecules provide glimpses into complex folding landscapes that are difficult to visualize using conventional ensemble measurements. Recent experiments and theoretical analyses have highlighted dynamic disorder in certain classes of biomolecules, whose dynamic pattern of conformational transitions is affected by slower transition dynamics of internal state hidden in a low dimensional projection. A systematic means to analyze such data is, however, currently not well developed. Here we report a new algorithm - Variational Bayes-double chain Markov model (VB-DCMM) - to analyze single molecule time trajectories that display dynamic disorder. The proposed analysis employing VB-DCMM allows us to detect the presence of dynamic disorder, if any, in each trajectory, identify the number of internal states, and estimate transition rates between the internal states as well as the rates of conformational transition within each internal state. Applying VB-DCMM algorithm to single molecule FRET data of H-DNA in 100 mM-Na$^+$ solution, followed by data clustering, we show that at least 6 kinetic paths linking 4 distinct internal states are required to correctly interpret the duplex-triplex transitions of H-DNA.

q-bio.QM↗