arXiv subjects

Kyungmin Kim

Publications and source records attributed to Kyungmin Kim.

At least 37 records · Page 2Linked to original sources

arXiv2025

TARDiS : Text Augmentation for Refining Diversity and Separability

Text augmentation (TA) is a critical technique for text classification, especially in few-shot settings. This paper introduces a novel LLM-based TA method, TARDiS, to address challenges inherent in the generation and alignment stages of two-stage TA methods. For the generation stage, we propose two generation processes, SEG and CEG, incorporating multiple class-specific prompts to enhance diversity and separability. For the alignment stage, we introduce a class adaptation (CA) method to ensure that generated examples align with their target classes through verification and modification. Experimental results demonstrate TARDiS's effectiveness, outperforming state-of-the-art LLM-based TA methods in various few-shot text classification tasks. An in-depth analysis confirms the detailed behaviors at each stage.

cs.CL↗

arXiv2024

Realizable Continuous-Space Shields for Safe Reinforcement Learning

While Deep Reinforcement Learning (DRL) has achieved remarkable success across various domains, it remains vulnerable to occasional catastrophic failures without additional safeguards. An effective solution to prevent these failures is to use a shield that validates and adjusts the agent's actions to ensure compliance with a provided set of safety specifications. For real-world robotic domains, it is essential to define safety specifications over continuous state and action spaces to accurately account for system dynamics and compute new actions that minimally deviate from the agent's original decision. In this paper, we present the first shielding approach specifically designed to ensure the satisfaction of safety requirements in continuous state and action spaces, making it suitable for practical robotic applications. Our method builds upon realizability, an essential property that confirms the shield will always be able to generate a safe action for any state in the environment. We formally prove that realizability can be verified for stateful shields, enabling the incorporation of non-Markovian safety requirements, such as loop avoidance. Finally, we demonstrate the effectiveness of our approach in ensuring safety without compromising the policy's success rate by applying it to a navigation problem and a multi-agent particle environment.

cs.LG↗

arXiv2024

Can we discern millilensed gravitational-wave signals from signals produced by precessing binary black holes with ground-based detectors?

Millilensed gravitational waves (GWs) can potentially be identified by the interference signatures caused by $\sim\!O(10\textrm{--}100)~\textrm{ms}$ time delays between multiple overlapping lensed signals. However, distinguishing millilensed GWs from GWs generated by precessing binary black-hole mergers can be challenging due to their apparent similar waveform shapes. This morphological similarity may be an obstacle to template-based searches to correctly identifying the origin of observed GWs and poses a fundamental question, can we discern millilensed GW signals from signals produced by precessing binary black holes? In this study, we investigate the feasibility of distinguishing between these GWs by performing a proof-of-principle injection study of simulated millilensed precessing GW signals, within the context of ground-based LIGO-Virgo-KAGRA detector network detections. Our findings indicate that it is possible to differentiate between the two effects by comparing signal-to-noise ratios (SNRs) computed using templates based on different hypotheses for the target signal. We further show from the parameter estimation study that while lensing magnification is sensitive to precession, it is possible to identify millilensing in precessing GW signals with an SNR of 18. The recovery of precession in the presence of lensing is more challenging but improves significantly for signals with an SNR of 40. Nonetheless, neglecting millilensing effects results in biases in the recovered spins, revealing the importance of accounting for these effects in accurate GW signal analysis.

gr-qc↗

arXiv2024

Statistical Discrimination in Ratings-Guided Markets

We study statistical discrimination of individuals based on payoff-irrelevant social identities in markets that utilize ratings and recommendations for social learning. Even though rating/recommendation algorithms can be designed to be fair and unbiased, ratings-based social learning can still lead to discriminatory outcomes. Our model demonstrates how users' attention choices can result in asymmetric data sampling across social groups, leading to discriminatory inferences and potential discrimination based on group identities.

cs.GT↗

arXiv2024

Make the Pertinent Salient: Task-Relevant Reconstruction for Visual Control with Distractions

Recent advancements in Model-Based Reinforcement Learning (MBRL) have made it a powerful tool for visual control tasks. Despite improved data efficiency, it remains challenging to train MBRL agents with generalizable perception. Training in the presence of visual distractions is particularly difficult due to the high variation they introduce to representation learning. Building on DREAMER, a popular MBRL method, we propose a simple yet effective auxiliary task to facilitate representation learning in distracting environments. Under the assumption that task-relevant components of image observations are straightforward to identify with prior knowledge in a given task, we use a segmentation mask on image observations to only reconstruct task-relevant components. In doing so, we greatly reduce the complexity of representation learning by removing the need to encode task-irrelevant objects in the latent representation. Our method, Segmentation Dreamer (SD), can be used either with ground-truth masks easily accessible in simulation or by leveraging potentially imperfect segmentation foundation models. The latter is further improved by selectively applying the reconstruction loss to avoid providing misleading learning signals due to mask prediction errors. In modified DeepMind Control suite (DMC) and Meta-World tasks with added visual distractions, SD achieves significantly better sample efficiency and greater final performance than prior work. We find that SD is especially helpful in sparse reward tasks otherwise unsolvable by prior work, enabling the training of visually robust agents without the need for extensive reward engineering.

cs.LG↗

arXiv2024

Rethinking Data Augmentation for Robust LiDAR Semantic Segmentation in Adverse Weather

Existing LiDAR semantic segmentation methods often struggle with performance declines in adverse weather conditions. Previous work has addressed this issue by simulating adverse weather or employing universal data augmentation during training. However, these methods lack a detailed analysis and understanding of how adverse weather negatively affects LiDAR semantic segmentation performance. Motivated by this issue, we identified key factors of adverse weather and conducted a toy experiment to pinpoint the main causes of performance degradation: (1) Geometric perturbation due to refraction caused by fog or droplets in the air and (2) Point drop due to energy absorption and occlusions. Based on these findings, we propose new strategic data augmentation techniques. First, we introduced a Selective Jittering (SJ) that jitters points in the random range of depth (or angle) to mimic geometric perturbation. Additionally, we developed a Learnable Point Drop (LPD) to learn vulnerable erase patterns with a Deep Q-Learning Network to approximate the point drop phenomenon from adverse weather conditions. Without precise weather simulation, these techniques strengthen the LiDAR semantic segmentation model by exposing it to vulnerable conditions identified by our data-centric analysis. Experimental results confirmed the suitability of the proposed data augmentation methods for enhancing robustness against adverse weather conditions. Our method achieves a notable 39.5 mIoU on the SemanticKITTI-to-SemanticSTF benchmark, improving the baseline by 8.1\%p and establishing a new state-of-the-art. Our code will be released at \url{https://github.com/engineerJPark/LiDARWeather}.

cs.CV↗

arXiv2024

Reinforcement Learning from Delayed Observations via World Models

In standard reinforcement learning settings, agents typically assume immediate feedback about the effects of their actions after taking them. However, in practice, this assumption may not hold true due to physical constraints and can significantly impact the performance of learning algorithms. In this paper, we address observation delays in partially observable environments. We propose leveraging world models, which have shown success in integrating past observations and learning dynamics, to handle observation delays. By reducing delayed POMDPs to delayed MDPs with world models, our methods can effectively handle partial observability, where existing approaches achieve sub-optimal performance or degrade quickly as observability decreases. Experiments suggest that one of our methods can outperform a naive model-based approach by up to 250%. Moreover, we evaluate our methods on visual delayed environments, for the first time showcasing delay-aware reinforcement learning continuous control with visual observations.

cs.LG↗

arXiv2024

Binary microlensing by high eccentric stellar-mass black hole binaries

Microlensing is one of the most promising tools for discovering stellar-mass black holes (BHs) in the Milky Way because it allows us to probe dark or faint celestial compact objects. While the existence of stellar-mass BHs has been confirmed through observation of X-ray binaries within our galaxy and gravitational waves from extragalactic BH binaries, a conclusive observation of microlensing events caused by Galactic BH binaries has yet to be achieved. In this study, we focus on those with high eccentricity, including unbound orbits, which can dynamically form in star clusters and could potentially increase the observation rate. We demonstrate parameter estimation for simulated light curves supposing various orbital configurations of BH binary lenses. We employ a model-based fitting using the Nelder-Mead method and Bayesian inference based on the Markov chain Monte Carlo method for the demonstration. The results show that we can retrieve true values of the parameters of high eccentric BH binary lenses within the 1$σ$ uncertainty of inferred values. We conclude it is feasible to find high eccentric Galactic BH binaries from the observation of binary microlensing events.

astro-ph.HE↗

arXiv2024

Gravitational lensing aided luminosity distance estimation for compact binary coalescences

The luminosity distance is a key observable of gravitational-wave (GW) observations. We demonstrate how one can correctly retrieve the luminosity distance of compact binary coalescences (CBCs) if the GW signal is strongly lensed. We perform a proof-of-concept parameter estimation for the luminosity distance supposing (i) strong lensing produces two lensed GW signals emitted from a CBC, (ii) the Advanced LIGO-Virgo network detects both lensed signals as independent events, and (iii) the two events are identified as strongly lensed signals originated from the same source. Taking into account the maximum magnification allowed in two lensing scenarios and simulated GW signals emitted from four different binary black holes, we find that the strong lensing can improve the precision of the distance estimation of a CBC by up to a factor of a few compared to that can be expected without lensing.

gr-qc↗

arXiv2023

Enhancing the Electron Pair Approximation with Measurements on Trapped Ion Quantum Computers

The electron pair approximation offers a resource efficient variational quantum eigensolver (VQE) approach for quantum chemistry simulations on quantum computers. With the number of entangling gates scaling quadratically with system size and a constant energy measurement overhead, the orbital optimized unitary pair coupled cluster double (oo-upCCD) ansatz strikes a balance between accuracy and efficiency on today's quantum computers. However, the electron pair approximation makes the method incapable of producing quantitatively accurate energy predictions. In order to improve the accuracy without increasing the circuit depth, we explore the idea of reduced density matrix (RDM) based second order perturbation theory (PT2) as an energetic correction to electron pair approximation. The new approach takes into account of the broken-pair energy contribution that is missing in pair-correlated electron simulations, while maintaining the computational advantages of oo-upCCD ansatz. In dissociations of N$_2$, Li$_2$O, and chemical reactions such as the unimolecular decomposition of CH$_2$OH$^+$ and the \snTwo reaction of CH$_3$I $+$ Br$^-$, the method significantly improves the accuracy of energy prediction. On two generations of the IonQ's trapped ion quantum computers, Aria and Forte, we find that unlike the VQE energy, the PT2 energy correction is highly noise-resilient. By applying a simple error mitigation approach based on post-selection solely on the VQE energies, the predicted VQE-PT2 energy differences between reactants, transition state, and products are in excellent agreement with noise-free simulators.

physics.chem-ph↗

arXiv2023

Variational quantum eigensolver for closed-shell molecules with non-bosonic corrections

The realization of quantum advantage with noisy-intermediate-scale quantum (NISQ) machines has become one of the major challenges in computational sciences. Maintaining coherence of a physical system with more than ten qubits is a critical challenge that motivates research on compact system representations to reduce algorithm complexity. Toward this end, quantum simulations based on the variational quantum eigensolver (VQE) is considered to be one of the most promising algorithms for quantum chemistry in the NISQ era. We investigate reduced mapping of one spatial orbital to a single qubit to analyze the ground state energy in a way that the Pauli operators of qubits are mapped to the creation/annihilation of singlet pairs of electrons. To include the effect of non-bosonic (or non-paired) excitations, we introduce a simple correction scheme in the electron correlation model approximated by the geometrical mean of the bosonic (or paired) terms. Employing it in a VQE algorithm, we assess ground state energies of H2O, N2, and Li2O in good agreements with full configuration interaction (FCI) models respectively, using only 6, 8, and 12 qubits with quantum gate depths proportional to the squares of the qubit counts. With the adopted seniority-zero approximation that uses only one half of the qubit counts of a conventional VQE algorithm, we find our non-bosonic correction method reaches reliable quantum chemistry simulations at least for the tested systems.

quant-ph↗

arXiv2023

Persuasion in Veto Bargaining

We consider the classic veto bargaining model but allow the agenda setter to engage in persuasion to convince the veto player to approve her proposal. We fully characterize the optimal proposal and experiment when Vetoer has quadratic loss, and show that the proposer-optimal can be achieved either by providing no information or with a simple binary experiment. Proposer chooses to reveal partial information when there is sufficient expected misalignment with Vetoer. In this case the opportunity to engage in persuasion strictly benefits Proposer and increases the scope to exercise agenda power.

econ.TH↗

arXiv2023

Selective Perception: Optimizing State Descriptions with Reinforcement Learning for Language Model Actors

Large language models (LLMs) are being applied as actors for sequential decision making tasks in domains such as robotics and games, utilizing their general world knowledge and planning abilities. However, previous work does little to explore what environment state information is provided to LLM actors via language. Exhaustively describing high-dimensional states can impair performance and raise inference costs for LLM actors. Previous LLM actors avoid the issue by relying on hand-engineered, task-specific protocols to determine which features to communicate about a state and which to leave out. In this work, we propose Brief Language INputs for DEcision-making Responses (BLINDER), a method for automatically selecting concise state descriptions by learning a value function for task-conditioned state descriptions. We evaluate BLINDER on the challenging video game NetHack and a robotic manipulation task. Our method improves task success rate, reduces input size and compute costs, and generalizes between LLM actors.

cs.LG↗

arXiv2023

Relation-Aware Language-Graph Transformer for Question Answering

Question Answering (QA) is a task that entails reasoning over natural language contexts, and many relevant works augment language models (LMs) with graph neural networks (GNNs) to encode the Knowledge Graph (KG) information. However, most existing GNN-based modules for QA do not take advantage of rich relational information of KGs and depend on limited information interaction between the LM and the KG. To address these issues, we propose Question Answering Transformer (QAT), which is designed to jointly reason over language and graphs with respect to entity relations in a unified manner. Specifically, QAT constructs Meta-Path tokens, which learn relation-centric embeddings based on diverse structural and semantic relations. Then, our Relation-Aware Self-Attention module comprehensively integrates different modalities via the Cross-Modal Relative Position Bias, which guides information exchange between relevant entites of different modalities. We validate the effectiveness of QAT on commonsense question answering datasets like CommonsenseQA and OpenBookQA, and on a medical question answering dataset, MedQA-USMLE. On all the datasets, our method achieves state-of-the-art performance. Our code is available at http://github.com/mlvlab/QAT.

cs.CL↗

arXiv2022

Search for Microlensing Signature in Gravitational Waves from Binary Black Hole Events

In a recent search (Kim et al. 2022), we looked for microlensing signature in gravitational waves from spectrograms of the binary black hole events in the first and second gravitational-wave transient catalogs. For the search, we have implemented a deep learning-based method (Kim et al. 2021) and figured out that one event, GW190707 093326, out of forty-six events, is classified into the lensed class. However, upon estimating the p-value of this event, we observed that the uncertainty of the p-value still includes the possibility of the event being unlensed. Therefore, we concluded that no significant evidence of beating patterns from the evaluated binary black hole events has found from the search. For a consequence study, we discuss the distinguishability between microlensed gravitational waves and the signal from precessing black hole binaries.

astro-ph.IM↗

arXiv2022

Meta-node: A Concise Approach to Effectively Learn Complex Relationships in Heterogeneous Graphs

Existing message passing neural networks for heterogeneous graphs rely on the concepts of meta-paths or meta-graphs due to the intrinsic nature of heterogeneous graphs. However, the meta-paths and meta-graphs need to be pre-configured before learning and are highly dependent on expert knowledge to construct them. To tackle this challenge, we propose a novel concept of meta-node for message passing that can learn enriched relational knowledge from complex heterogeneous graphs without any meta-paths and meta-graphs by explicitly modeling the relations among the same type of nodes. Unlike meta-paths and meta-graphs, meta-nodes do not require any pre-processing steps that require expert knowledge. Going one step further, we propose a meta-node message passing scheme and apply our method to a contrastive learning model. In the experiments on node clustering and classification tasks, the proposed meta-node message passing method outperforms state-of-the-arts that depend on meta-paths. Our results demonstrate that effective heterogeneous graph learning is possible without the need for meta-paths that are frequently used in this field.

cs.LG↗

arXiv2022

Keeping the Listener Engaged: a Dynamic Model of Bayesian Persuasion

We consider a dynamic model of Bayesian persuasion in which information takes time and is costly for the sender to generate and for the receiver to process, and neither player can commit to their future actions. Persuasion may totally collapse in a Markov perfect equilibrium (MPE) of this game. However, for persuasion costs sufficiently small, a version of a folk theorem holds: outcomes that approximate Kamenica and Gentzkow (2011)'s sender-optimal persuasion as well as full revelation and everything in between are obtained in MPE, as the cost vanishes.

econ.TH↗

arXiv2022

Deep Learning-based Search for Microlensing Signature from Binary Black Hole Events in GWTC-1 and -2

We present the result of the first deep learning-based search for the signature of microlensing in gravitational waves. This search seeks the signature induced by lenses with masses between $10^3M_\odot$--$10^5M_\odot$ from spectrograms of the binary black hole events in the first and second gravitational-wave transient catalogs. We use a deep learning model trained with spectrograms of simulated noisy gravitational-wave signals to classify the events into two classes, lensed or unlensed. We introduce ensemble learning and a majority voting-based consistency test for the predictions of ensemble learners. The classification scheme of this search primarily classifies one event, GW190707_093326, into the lensed class. To verify the primary classification of this event, we also examine the median probability to the lensed class and observe the resulting value, $0.984^{+0.012}_{-0.342}$, agrees with an empirical criterion $>\!0.6$ for claiming the detection of a lensed signal. However, the uncertainty of the estimated $p$-value for the median probability and error, ranging from 0 to 0.1, convinces us GW190707_093326 is less likely a lensed event because it includes $p\!\geq\!0.05$ where the unlensed hypothesis is true. Therefore, we conclude our search finds no significant evidence of microlensing signature from the evaluated binary black hole events.

gr-qc↗