arXiv subjects

Taewoo Lee

Publications and source records attributed to Taewoo Lee.

At least 19 recordsLinked to original sources

arXiv2026

Llamion Technical Report

We release Llamion, a family of 14B-parameter open-weight language models obtained by transforming Orion-14B into the standardized Llama-family architecture. The transformation is performed by Efficient Knowledge Preservation for Transformation (KEPT), a recipe that combines (i) Normal Parameter Mapping (NPM) for unchanged modules, (ii) Optimized Parameter Mapping (OPM), a training-free LayerNorm-to-RMSNorm initialization we prove optimal under the near-zero-mean activation regime induced by weight decay, and (iii) Cross-architecture Knowledge Distillation (XKD), an equal-size frozen-teacher distillation that aligns the converted model's outputs with the source model's on any reasonable input distribution. Llamion recovers Orion's behaviour on H6, MT-Bench, and KoMMLU with only ~123M tokens on a single A100 in four days; Llamion-Base reaches 66.87% on KoMMLU, exceeding the next-best entry of the Open Ko LLM Leaderboard by >7.0 absolute points at submission time. Capabilities entirely absent from the transfer corpus (Python programming and 200K-token context handling) survive the architectural transition intact. We release three checkpoints (Base, Chat, LongChat) that load with trust_remote_code=False in the Hugging Face Transformers library.

cs.CL↗

arXiv2023

KoBigBird-large: Transformation of Transformer for Korean Language Understanding

This work presents KoBigBird-large, a large size of Korean BigBird that achieves state-of-the-art performance and allows long sequence processing for Korean language understanding. Without further pretraining, we only transform the architecture and extend the positional encoding with our proposed Tapered Absolute Positional Encoding Representations (TAPER). In experiments, KoBigBird-large shows state-of-the-art overall performance on Korean language understanding benchmarks and the best performance on document classification and question answering tasks for longer sequences against the competitive baseline models. We publicly release our model here.

cs.CL↗

arXiv2023

Toward accurate thermal modeling of phase change material based photonic devices

Reconfigurable or programmable photonic devices are rapidly growing and have become an integral part of many optical systems. The ability to selectively modulate electromagnetic waves through electrical stimuli is crucial in the advancement of a variety of applications from data communication and computing devices to environmental science and space explorations. Chalcogenide-based phase change materials (PCMs) are one of the most promising material candidates for reconfigurable photonics due to their large optical contrast between their different solid-state structural phases. Although significant efforts have been devoted to accurate simulation of PCM-based devices, in this paper, we highlight three important aspects which have often evaded prior models yet having significant impacts on the thermal and phase transition behavior of these devices: the enthalpy of fusion, the heat capacity change upon glass transition, as well as the thermal conductivity of liquid-phase PCMs. We further investigated the important topic of switching energy scaling in PCM devices, which also helps explain why the three above-mentioned effects have long been overlooked in electronic PCM memories but only become important in photonics. Our findings offer insight to facilitate accurate modeling of PCM-based photonic devices and can inform the development of more efficient reconfigurable optics.

physics.optics↗

arXiv2022

Approximate Submodularity and Its Implications in Discrete Optimization

Submodularity is a key property in discrete optimization. Submodularity has been widely used for analyzing the greedy algorithm to give performance bounds and providing insight into the construction of valid inequalities for mixed-integer programs. In recent years, researchers started to study approximate submodularity, with a primary focus on providing performance bounds for iterative approaches. In this paper, we study approximate submodularity from a different perspective in order to broaden its use cases in discrete optimization. We define metrics that quantify approximate submodularity, which we then use to derive new properties about both approximate submodularity preservation and the well-known Lovász extension for set functions. We also show that previous analyses of mixed-integer sets, such as the submodular knapsack polytope, can be extended to the approximate submodularity setting. Our work demonstrates that one may generalize many of the analytical tools used in submodular optimization into the approximate submodularity context.

math.OC↗

arXiv2021

Optimality-Based Clustering: An Inverse Optimization Approach

We propose a new clustering approach, called optimality-based clustering, that clusters data points based on their latent decision-making preferences. We assume that each data point is a decision generated by a decision-maker who (approximately) solves an optimization problem and cluster the data points by identifying a common objective function of the optimization problems for each cluster such that the worst-case optimality error is minimized. We propose three different clustering models and test them in the diet recommendation application.

math.OC↗

arXiv2021

Adaptable Multi-Domain Language Model for Transformer ASR

We propose an adapter based multi-domain Transformer based language model (LM) for Transformer ASR. The model consists of a big size common LM and small size adapters. The model can perform multi-domain adaptation with only the small size adapters and its related layers. The proposed model can reuse the full fine-tuned LM which is fine-tuned using all layers of an original model. The proposed LM can be expanded to new domains by adding about 2% of parameters for a first domain and 13% parameters for after second domain. The proposed model is also effective in reducing the model maintenance cost because it is possible to omit the costly and time-consuming common LM pre-training process. Using proposed adapter based approach, we observed that a general LM with adapter can outperform a dedicated music domain LM in terms of word error rate (WER).

eess.AS↗

arXiv2020

Quantile Inverse Optimization: Improving Stability in Inverse Linear Programming

Inverse linear programming (LP) has received increasing attention due to its potential to generate efficient optimization formulations that can closely replicate the behavior of a complex system. However, inversely inferred parameters and corresponding forward solutions from the existing inverse LP method can be highly sensitive to noise, errors, and uncertainty in the input data, limiting its applicability in data-driven settings. We introduce the notion of inverse and forward stability in inverse LP and propose a novel inverse LP method that determines a set of objective functions that are stable under data imperfection and generate solutions close to the relevant subset of the data. We formulate the inverse model as a mixed-integer program and elucidate its connection to bi-clique problems, which we exploit to develop efficient heuristics. We also show how this method can be used for online learning. We numerically evaluate the stability of the proposed method and demonstrate its practical use in the diet recommendation and transshipment applications.

math.OC↗

arXiv2020

An Ensemble Learning Framework for Model Fitting and Evaluation in Inverse Linear Optimization

We develop a generalized inverse optimization framework for fitting the cost vector of a single linear optimization problem given multiple observed decisions. This setting is motivated by ensemble learning, where building consensus from base learners can yield better predictions. We unify several models in the inverse optimization literature under a single framework and derive assumption-free and exact solution methods for each one. We extend a goodness-of-fit metric previously introduced for the problem with a single observed decision to this new setting, and demonstrate several important properties. Finally, we demonstrate our framework in a novel inverse optimization-driven procedure for automated radiation therapy treatment planning. Here, the inverse optimization model leverages an ensemble of dose predictions from different machine learning models to construct a consensus treatment plan that outperforms baseline methods. The consensus plan yields better trade-offs between the competing clinical criteria used for plan evaluation.

math.OC↗

arXiv2017

Inverse Optimization: Closed-form Solutions, Geometry and Goodness of fit

In classical inverse linear optimization, one assumes a given solution is a candidate to be optimal. Real data is imperfect and noisy, so there is no guarantee this assumption is satisfied. Inspired by regression, this paper presents a unified framework for cost function estimation in linear optimization comprising a general inverse optimization model and a corresponding goodness-of-fit metric. Although our inverse optimization model is nonconvex, we derive a closed-form solution and present the geometric intuition. Our goodness-of-fit metric, $ρ$, the coefficient of complementarity, has similar properties to $R^2$ from regression and is quasiconvex in the input data, leading to an intuitive geometric interpretation. While $ρ$ is computable in polynomial-time, we derive a lower bound that possesses the same properties, is tight for several important model variations, and is even easier to compute. We demonstrate the application of our framework for model estimation and evaluation in production planning and cancer therapy.

math.OC↗

arXiv2017

Trade-off preservation in inverse multi-objective convex optimization

We present a new inverse optimization methodology for multi-objective convex optimization that accommodates an input solution that may not be Pareto optimal and determines a weight vector that produces a Pareto optimal solution that approximates the input solution and preserves the decision maker's intention encoded in it. We introduce a notion of trade-off preservation, which we use as a measure of similarity for approximating the input solution, and show its connection with minimizing an optimality gap. Our inverse model maintains the complexity of the traditional inverse convex models. We propose a linear approximation to the model and a successive linear programming algorithm that balance between trade-off preservation and computational efficiency, and show that our model encompasses many of the existing models from the literature. We demonstrate the proposed method using clinical data from prostate cancer radiation therapy.

math.OC↗

arXiv2017

Imaging of director fields in liquid crystals using stimulated Raman scattering microscopy

We demonstrate an approach for background-free three-dimensional imaging of director fields in liquid crystals using stimulated Raman scattering microscopy. This imaging technique is implemented using a single femtosecond pulsed laser and a photonic crystal fiber, providing Stokes and pump frequencies needed to access Raman shifts of different chemical bonds of molecules and allowing for chemically selective and broadband imaging of both pristine liquid crystals and composite materials. Using examples of model three-dimensional structures of director fields, we show that the described technique is a powerful tool for mapping of long-range molecular orientation patterns in soft matter via polarized chemical-selective imaging.

cond-mat.soft↗

arXiv2017

Optical manipulation of shape-morphing elastomeric liquid crystal microparticles doped with gold nanocrystals

We demonstrate facile optical manipulation of shape of birefringent colloidal microparticles made from liquid crystal elastomers. Using soft lithography and polymerization, we fabricate elastomeric microcylinders with weakly undulating director oriented on average along their long axes. These particles are infiltrated with gold nanospheres acting as heat transducers that allow for an efficient localized transfer of heat from a focused infrared laser beam to a submicrometer region within a microparticle. Photothermal control of ordering in the liquid crystal elastomer using scanned beams allows for a robust control of colloidal particles, enabling both reversible and irreversible changes of shape. Possible applications include optomechanics, microfluidics, and reconfigurable colloidal composites with shape-dependent self-assembly.

cond-mat.soft↗

arXiv2016

Active shape-morphing elastomeric colloids in short-pitch cholesteric liquid crystals

Active elastomeric liquid crystal particles with initial cylindrical shapes are obtained by means of soft lithography and polymerization in a strong magnetic field. Gold nanocrystals infiltrated into these particles mediate energy transfer from laser light to heat, so that the inherent coupling between the temperature-dependent order and shape allows for dynamic morphing of these particles and well-controlled stable shapes. Continuous changes of particle shapes are followed by their spontaneous realignment and transformations of director structures in the surrounding cholesteric host, as well as locomotion in the case of a nonreciprocal shape morphing. These findings bridge the fields of liquid crystal solids and active colloids, may enable shape-controlled self-assembly of adaptive composites and light-driven micromachines, and can be understood by employing simple symmetry considerations along with electrostatic analogies.

cond-mat.soft↗

arXiv2016

Reconfigurable interactions and three-dimensional patterning of colloidal particles and defects in lamellar soft media

Colloidal systems find important applications ranging from fabrication of photonic crystals to direct probing of phenomena typically encountered in atomic crystals and glasses. New applications - such as nanoantennas, plasmonic sensors, and nanocircuits - pose a challenge of achieving sparse colloidal assemblies with tunable interparticle separations that can be controlled at will. We demonstrate reconfigurable multiscale interactions and assembly of colloids mediated by defects in cholesteric liquid crystals that are probed by means of laser manipulation and three-dimensional imaging. We find that colloids attract via distance-independent elastic interactions when pinned to the ends of cholesteric oily streaks, line defects at which one or more layers are interrupted. However, dislocations and oily streaks can also be optically manipulated to induce kinks, allowing one to lock them into the desired configurations that are stabilized by elastic energy barriers for structural transformation of the particle-connecting defects. Under the influence of elastic energy landscape due to these defects, sublamellar-sized colloids self-assemble into structures mimicking the cores of dislocations and oily streaks. Interactions between these defect embedded colloids can be varied from attractive to repulsive by optically introducing dislocation kinks. The reconfigurable nature of defect-particle interactions allows for patterning of defects by manipulation of colloids and, in turn, patterning of particles by these defects, thus achieving desired colloidal configurations on scales ranging from the size of defect core to the sample size. This defect-colloidal sculpturing may be extended to other lamellar media, providing the means for optically guided self-assembly of mesoscopic composites with predesigned properties.

cond-mat.soft↗

arXiv2016

Plasmonic complex fluids of nematiclike and helicoidal self-assemblies of gold nanorods with a negative order parameter

We describe a soft matter system of self-organized oblate micelles and plasmonic gold nanorods that exhibit a negative orientational order parameter. Because of anisotropic surface anchoring interactions, colloidal gold nanorods tend to align perpendicular to the director describing the average orientation of normals to the discoidal micelles. Helicoidal structures of highly concentrated nanorods with a negative order parameter are realized by adding a chiral additive and are further controlled by means of confinement and mechanical stress. Polarization-sensitive absorption, scattering, and two-photon luminescence are used to characterize orientations and spatial distributions of nanorods. Self-alignment and effective-medium optical properties of these hybrid inorganic-organic complex fluids match predictions of a simple model based on anisotropic surface anchoring interactions of nanorods with the structured host medium.

cond-mat.soft↗

arXiv2016

Shape-dependent oriented trapping and scaffolding of plasmonic nanoparticles by topological defects for self-assembly of colloidal dimers in liquid crystals

We demonstrate scaffolding of plasmonic nanoparticles by topological defects induced by colloidal microspheres to match their surface boundary conditions with a uniform far-field alignment in a liquid crystal host. Displacing energetically costly liquid crystal regions of reduced order, anisotropic nanoparticles with concave or convex shapes not only stably localize in defects but also self-orient with respect to the microsphere surface. Using laser tweezers, we manipulate the ensuing nanoparticle-microsphere colloidal dimers, probing the strength of elastic binding and demonstrating self-assembly of hierarchical colloidal superstructures such as chains and arrays.

cond-mat.soft↗

arXiv2016

Nonlinear photoluminescence imaging of isotropic and liquid crystalline dispersions of graphene oxide

We report a visible-range nonlinear photoluminescence (PL) from graphene oxide (GO) flakes excited by near-infrared femtosecond laser light. PL intensity has nonlinear dependence on the laser power, implying a multiphoton excitation process, and also strongly depends on a linear polarization orientation of excitation light, being at maximum when it is parallel to flakes. We show that PL can be used for a fully three-dimensional label-free imaging of isotropic, nematic, and lamellar liquid crystalline dispersions of GO flakes in water. This nonlinear PL is of interest for applications in direct label-free imaging of composite materials and study of orientational ordering in mesomorphic phases formed by these flakes, as well as in biomedical and sensing applications utilizing GO.

cond-mat.mtrl-sci↗

arXiv2016

Three-dimensional patterning of solid microstructures through laser reduction of colloidal graphene oxide in liquid-crystalline dispersions

Graphene materials and structures have become an essential part of modern electronics and photovoltaics. However, despite many production methods, applications of graphene-based structures are hindered by high costs, lack of scalability and limitations in spatial patterning. Here we fabricate three-dimensional functional solid microstructures of reduced graphene oxide in a lyotropic nematic liquid crystal of graphene oxide flakes using a pulsed near-infrared laser. This reliable, scalable approach is mask-free, does not require special chemical reduction agents, and can be implemented at ambient conditions starting from aqueous graphene oxide flakes. Orientational ordering of graphene oxide flakes in self-assembled liquid-crystalline phases enables laser patterning of complex, three-dimensional reduced graphene oxide structures and colloidal particles, such as trefoil knots, with "frozen" orientational order of flakes. These structures and particles are mechanically rigid and range from hundreds of nanometres to millimetres in size, as needed for applications in colloids, electronics, photonics and display technology.

cond-mat.mtrl-sci↗