arXiv subjects

Minsu Kim

Publications and source records attributed to Minsu Kim.

At least 55 records · Page 3Linked to original sources

arXiv2025

Efficient Domain Generalization in Wireless Networks with Scarce Multi-Modal Data

In 6G wireless networks, multi-modal ML models can be leveraged to enable situation-aware network decisions in dynamic environments. However, trained ML models often fail to generalize under domain shifts when training and test data distributions are different because they often focus on modality-specific spurious features. In practical wireless systems, domain shifts occur frequently due to dynamic channel statistics, moving obstacles, or hardware configuration. Thus, there is a need for learning frameworks that can achieve robust generalization under scarce multi-modal data in wireless networks. In this paper, a novel and data-efficient two-phase learning framework is proposed to improve generalization performance in unseen and unfamiliar wireless environments with minimal amount of multi-modal data. In the first stage, a physics-based loss function is employed to enable each BS to learn the physics underlying its wireless environment captured by multi-modal data. The data-efficiency of the physics-based loss function is analytically investigated. In the second stage, collaborative domain adaptation is proposed to leverage the wireless environment knowledge of multiple BSs to guide under-performing BSs under domain shift. Specifically, domain-similarity-aware model aggregation is proposed to utilize the knowledge of BSs that experienced similar domains. To validate the proposed framework, a new dataset generation framework is developed by integrating CARLA and MATLAB-based mmWave channel modeling to predict mmWave RSS. Simulation results show that the proposed physics-based training requires only 13% of data samples to achieve the same performance as a state-of-the-art baseline that does not use physics-based training. Moreover, the proposed collaborative domain adaptation needs only 25% of data samples and 20% of FLOPs to achieve the convergence compared to baselines.

eess.SP↗

arXiv2025

Trajectory Balance with Asynchrony: Decoupling Exploration and Learning for Fast, Scalable LLM Post-Training

Reinforcement learning (RL) is a critical component of large language model (LLM) post-training. However, on-policy algorithms used for post-training are not naturally robust to a diversified content of experience replay buffers, which asynchronous off-policy actors can efficiently populate in parallel to training. We propose efficiently learning on such off-policy data via Trajectory Balance with Asynchrony (TBA), an approach to asynchronous RL for LLMs that leverages the principled off-policy TB objective. On math, preference-tuning, and automated red-teaming tasks, we post-train models ranging from Pythia 410M to Qwen 2.5 7B, finding TBA offers speed and performance boosts over strong baselines like Online DPO and Dr. GRPO. Beyond TBA's performance benefits (high accuracy even as asynchrony grows) and speedups ($4\times$ or more), we show its reward- and recency-prioritizing sampling enable further gains as data generation is scaled. Our code is available at https://github.com/bbartoldson/TBA.

cs.LG↗

arXiv2025

A CNN-Based Technique to Assist Layout-to-Generator Conversion for Analog Circuits

We propose a technique to assist in converting a reference layout of an analog circuit into the procedural layout generator by efficiently reusing available generators for sub-cell creation. The proposed convolutional neural network (CNN) model automatically detects sub-cells that can be generated by available generator scripts in the library, and suggests using them in the hierarchically correct places of the generator software. In experiments, the CNN model examined sub-cells of a high-speed wireline receiver that has a total of 4,885 sub-cell instances including different 145 sub-cell designs. The CNN model classified the sub-cell instances into 51 generatable and one not-generatable classes. One not-generatable class indicates that no available generator can generate the classified sub-cell. The CNN model achieved 99.3% precision in examining the 145 different sub-cell designs. The CNN model greatly reduced the examination time to 18 seconds from 88 minutes required in manual examination. Also, the proposed CNN model could correctly classify unfamiliar sub-cells that are very different from the training dataset.

cs.AR↗

arXiv2025

Energy-based generator matching: A neural sampler for general state space

We propose Energy-based generator matching (EGM), a modality-agnostic approach to train generative models from energy functions in the absence of data. Extending the recently proposed generator matching, EGM enables training of arbitrary continuous-time Markov processes, e.g., diffusion, flow, and jump, and can generate data from continuous, discrete, and a mixture of two modalities. To this end, we propose estimating the generator matching loss using self-normalized importance sampling with an additional bootstrapping trick to reduce variance in the importance weight. We validate EGM on both discrete and multimodal tasks up to 100 and 20 dimensions, respectively.

cs.LG↗

arXiv2025

On scalable and efficient training of diffusion samplers

We address the challenge of training diffusion models to sample from unnormalized energy distributions in the absence of data, the so-called diffusion samplers. Although these approaches have shown promise, they struggle to scale in more demanding scenarios where energy evaluations are expensive and the sampling space is high-dimensional. To address this limitation, we propose a scalable and sample-efficient framework that properly harmonizes the powerful classical sampling method and the diffusion sampler. Specifically, we utilize Monte Carlo Markov chain (MCMC) samplers with a novelty-based auxiliary energy as a Searcher to collect off-policy samples, using an auxiliary energy function to compensate for exploring modes the diffusion sampler rarely visits. These off-policy samples are then combined with on-policy data to train the diffusion sampler, thereby expanding its coverage of the energy landscape. Furthermore, we identify primacy bias, i.e., the preference of samplers for early experience during training, as the main cause of mode collapse during training, and introduce a periodic re-initialization trick to resolve this issue. Our method significantly improves sample efficiency on standard benchmarks for diffusion samplers and also excels at higher-dimensional problems and real-world molecular conformer generation.

cs.LG↗

arXiv2025

Self-Evolving Curriculum for LLM Reasoning

Reinforcement learning (RL) has proven effective for fine-tuning large language models (LLMs), significantly enhancing their reasoning abilities in domains such as mathematics and code generation. A crucial factor influencing RL fine-tuning success is the training curriculum: the order in which training problems are presented. While random curricula serve as common baselines, they remain suboptimal; manually designed curricula often rely heavily on heuristics, and online filtering methods can be computationally prohibitive. To address these limitations, we propose Self-Evolving Curriculum (SEC), an automatic curriculum learning method that learns a curriculum policy concurrently with the RL fine-tuning process. Our approach formulates curriculum selection as a non-stationary Multi-Armed Bandit problem, treating each problem category (e.g., difficulty level or problem type) as an individual arm. We leverage the absolute advantage from policy gradient methods as a proxy measure for immediate learning gain. At each training step, the curriculum policy selects categories to maximize this reward signal and is updated using the TD(0) method. Across three distinct reasoning domains: planning, inductive reasoning, and mathematics, our experiments demonstrate that SEC significantly improves models' reasoning capabilities, enabling better generalization to harder, out-of-distribution test problems. Additionally, our approach achieves better skill balance when fine-tuning simultaneously on multiple reasoning domains. These findings highlight SEC as a promising strategy for RL fine-tuning of LLMs.

cs.AI↗

arXiv2025

Adaptive Inference-Time Scaling via Cyclic Diffusion Search

Diffusion models have demonstrated strong generative capabilities across domains ranging from image synthesis to complex reasoning tasks. However, most inference-time scaling methods rely on fixed denoising schedules, limiting their ability to allocate computation based on instance difficulty or task-specific demands adaptively. We introduce the challenge of adaptive inference-time scaling-dynamically adjusting computational effort during inference-and propose Adaptive Bi-directional Cyclic Diffusion (ABCD), a flexible, search-based inference framework. ABCD refines outputs through bi-directional diffusion cycles while adaptively controlling exploration depth and termination. It comprises three components: Cyclic Diffusion Search, Automatic Exploration-Exploitation Balancing, and Adaptive Thinking Time. Experiments show that ABCD improves performance across diverse tasks while maintaining computational efficiency.

cs.LG↗

arXiv2025

MoME: Mixture of Matryoshka Experts for Audio-Visual Speech Recognition

Large language models (LLMs) have recently shown strong potential in audio-visual speech recognition (AVSR), but their high computational demands and sensitivity to token granularity limit their practicality in resource-constrained settings. Token compression methods can reduce inference cost, but they require fixing a compression rate in advance and produce a single fixed-length output, offering no flexibility to balance information density and efficiency at inference time. Matryoshka representation learning (MRL) addresses this by enabling a single model to operate across multiple token granularities, allowing compression rates to be adjusted dynamically. However, current MRL-based methods treat each scale independently during training, limiting cross-scale generalization, robustness at high compression, and interpretability. To overcome these limitations, we propose MoME (Mixture of Matryoshka Experts), a novel framework that integrates sparse Mixture-of-Experts (MoE) into MRL-based LLMs for AVSR. MoME augments a frozen LLM with top-k routed and shared experts, allowing dynamic capacity allocation across scales and modalities. A shared router promotes consistent expert activation across granularities, enabling compressed sequences to benefit from representations learned at lower compression. Experiments on LRS2 and LRS3 demonstrate that MoME achieves state-of-the-art performance across AVSR, ASR, and VSR tasks, while requiring significantly fewer parameters and maintaining robustness under noise. MoME unifies the adaptability of MRL with the efficiency of MoE, offering a scalable and interpretable solution for resource-aware speech recognition.

eess.AS↗

arXiv2025

Active Attacks: Red-teaming LLMs via Adaptive Environments

We address the challenge of generating diverse attack prompts for large language models (LLMs) that elicit harmful behaviors (e.g., insults, sexual content) and are used for safety fine-tuning. Rather than relying on manual prompt engineering, attacker LLMs can be trained with reinforcement learning (RL) to automatically generate such prompts using only a toxicity classifier as a reward. However, capturing a wide range of harmful behaviors is a significant challenge that requires explicit diversity objectives. Existing diversity-seeking RL methods often collapse to limited modes: once high-reward prompts are found, exploration of new regions is discouraged. Inspired by the active learning paradigm that encourages adaptive exploration, we introduce \textit{Active Attacks}, a novel RL-based red-teaming algorithm that adapts its attacks as the victim evolves. By periodically safety fine-tuning the victim LLM with collected attack prompts, rewards in exploited regions diminish, which forces the attacker to seek unexplored vulnerabilities. This process naturally induces an easy-to-hard exploration curriculum, where the attacker progresses beyond easy modes toward increasingly difficult ones. As a result, Active Attacks uncovers a wide range of local attack modes step by step, and their combination achieves wide coverage of the multi-mode distribution. Active Attacks, a simple plug-and-play module that seamlessly integrates into existing RL objectives, unexpectedly outperformed prior RL-based methods -- including GFlowNets, PPO, and REINFORCE -- by improving cross-attack success rates against GFlowNets, the previous state-of-the-art, from 0.07% to 31.28% (a relative gain greater than $400\ \times$) with only a 6% increase in computation. Our code is publicly available \href{https://github.com/dbsxodud-11/active_attacks}{here}.

cs.LG↗

arXiv2025

State-of-Health Prediction for EV Lithium-Ion Batteries via DLinear and Robust Explainable Feature Selection

Accurate prediction of the state-of-health (SOH) of lithium-ion batteries is essential for ensuring the safety, reliability, and efficient operation of electric vehicles (EVs). Battery packs in EVs experience nonuniform degradation due to cell-to-cell variability (CtCV), posing a major challenge for real-time battery management. In this work, we propose an explainable, data-driven SOH prediction framework tailored for EV battery management systems (BMS). The approach combines robust feature engineering with a DLinear. Using NASA's battery aging dataset, we extract twenty meaningful features from voltage, current, temperature, and time profiles, and select key features using Pearson correlation and Shapley additive explanations (SHAP). The SHAP-based selection yields consistent feature importance across multiple cells, effectively capturing CtCV. The DLinear algorithm outperforms long short-term memory (LSTM) and Transformer architectures in prediction accuracy, while requiring fewer training cycles and lower computational cost. This work offers a scalable and interpretable framework for SOH forecasting, enabling practical implementation in EV BMS and promoting safer, more efficient electric mobility.

eess.SY↗

arXiv2025

Bang-Ride Optimal Control: Monotonicity, External Positivity, and Fast Battery Charging

This work studies a class of optimal control problems with scalar inputs and general constraints, whose solutions follow a bang-ride pattern that always activates a constraint and enables efficient numerical computation. As a motivating example, fast battery charging leads to computationally demanding optimal control problems when detailed electrochemical models are used. Recently proposed optimization-free heuristics reduce this computational cost while producing input profiles observed in practice, following a bang-ride pattern and applying the maximum feasible input. We investigate when such heuristics satisfy necessary optimality conditions. By leveraging Pontryagin's maximum principle, we unify and formalize existing insights on the bang-ride structure and on the optimal control attaining the maximum feasible input under monotonicity. We further establish a novel connection between the structured optimal control and the external positivity of the costate dynamics. These results provide a rigorous theoretical foundation for heuristic charging strategies and explain the efficiency of optimization-free algorithms.

math.OC↗

arXiv2025

ExploreGS: Explorable 3D Scene Reconstruction with Virtual Camera Samplings and Diffusion Priors

Recent advances in novel view synthesis (NVS) have enabled real-time rendering with 3D Gaussian Splatting (3DGS). However, existing methods struggle with artifacts and missing regions when rendering from viewpoints that deviate from the training trajectory, limiting seamless scene exploration. To address this, we propose a 3DGS-based pipeline that generates additional training views to enhance reconstruction. We introduce an information-gain-driven virtual camera placement strategy to maximize scene coverage, followed by video diffusion priors to refine rendered results. Fine-tuning 3D Gaussians with these enhanced views significantly improves reconstruction quality. To evaluate our method, we present Wild-Explore, a benchmark designed for challenging scene exploration. Experiments demonstrate that our approach outperforms existing 3DGS-based methods, enabling high-quality, artifact-free rendering from arbitrary viewpoints. https://exploregs.github.io

cs.CV↗

arXiv2025

Adaptive Audio-Visual Speech Recognition via Matryoshka-Based Multimodal LLMs

Audio-Visual Speech Recognition (AVSR) leverages audio and visual modalities to improve robustness in noisy environments. Recent advances in Large Language Models (LLMs) show strong performance in speech recognition, including AVSR. However, the long speech representations lead to high computational costs for LLMs. Prior methods compress inputs before feeding them to LLMs, but high compression often harms accuracy. To address this, we propose Llama-MTSK, the first Matryoshka-based Multimodal LLM for AVSR, which flexibly adapts audio-visual token allocation under varying compute constraints. Inspired by Matryoshka Representation Learning, our model encodes representations at multiple granularities with a single architecture, avoiding the need for separate models. For efficient fine-tuning, we introduce three LoRA-based strategies using global and scale-specific modules. Evaluations on major AVSR datasets show Llama-MTSK matches or outperforms models trained at fixed compression levels.

cs.CV↗

arXiv2025

Unveiling the nature and fate of the almost-dark cloud AGC 226178 through HI mapping

The origin of extragalactic, almost dark HI clouds with extreme gas-to-stellar mass ratios remains poorly understood. We investigate the nature and fate of the "almost dark" cloud AGC 226178, projected within the Virgo cluster, with an HI-to-stellar mass ratio of ~1000. We present deep single-dish HI mapping from the Five-hundred-meter Aperture Spherical Telescope (FAST), complemented by high-resolution interferometric data from the Very Large Array (VLA), as part of the Atomic gas in Virgo Interacting Dwarf galaxies (AVID) project. These observations provide the highest-quality HI analysis to date of such a cloud, combining resolution and sensitivity. FAST data reveal a short, low-velocity tail toward the dwarf galaxy VCC 2034, previously proposed as a possible origin for AGC 226178. However, VCC 2034 shows a line-of-sight asymmetric HI feature and cometary morphology indicating a stripping event unrelated to AGC 226178. VLA data reveal a velocity gradient across AGC 226178 and a clumpy internal structure. The velocity dispersion exceeds the thermal linewidth, implying turbulence or unresolved motions. The cloud cannot be gravitationally bound by atomic gas alone. The resolved HI clumps follow standard HI mass-star formation rate and mass-size relations, with those forming stars reaching surface densities above the threshold for self-shielding. We conclude that AGC 226178 is a free-floating HI cloud of unknown origin. The system appears to be in the process of disintegration. It is likely located well outside the Virgo cluster, as the preservation of its extended HI morphology within the cluster environment would otherwise require a substantial reservoir of unseen molecular gas with a mass exceeding that of the observed HI content. While confinement pressure from the hot intracluster medium may aid its stability, it is unlikely to be the dominant factor preventing its disruption.

astro-ph.GA↗

arXiv2025

Zero-AVSR: Zero-Shot Audio-Visual Speech Recognition with LLMs by Learning Language-Agnostic Speech Representations

We explore a novel zero-shot Audio-Visual Speech Recognition (AVSR) framework, dubbed Zero-AVSR, which enables speech recognition in target languages without requiring any audio-visual speech data in those languages. Specifically, we introduce the Audio-Visual Speech Romanizer (AV-Romanizer), which learns language-agnostic speech representations by predicting Roman text. Then, by leveraging the strong multilingual modeling capabilities of Large Language Models (LLMs), we propose converting the predicted Roman text into language-specific graphemes, forming the proposed Cascaded Zero-AVSR. Taking it a step further, we explore a unified Zero-AVSR approach by directly integrating the audio-visual speech representations encoded by the AV-Romanizer into the LLM. This is achieved through finetuning the adapter and the LLM using our proposed multi-task learning scheme. To capture the wide spectrum of phonetic and linguistic diversity, we also introduce a Multilingual Audio-Visual Romanized Corpus (MARC) consisting of 2,916 hours of audio-visual speech data across 82 languages, along with transcriptions in both language-specific graphemes and Roman text. Extensive analysis and experiments confirm that the proposed Zero-AVSR framework has the potential to expand language support beyond the languages seen during the training of the AV-Romanizer.

cs.CV↗

arXiv2025

RL4CO: an Extensive Reinforcement Learning for Combinatorial Optimization Benchmark

Combinatorial optimization (CO) is fundamental to several real-world applications, from logistics and scheduling to hardware design and resource allocation. Deep reinforcement learning (RL) has recently shown significant benefits in solving CO problems, reducing reliance on domain expertise and improving computational efficiency. However, the absence of a unified benchmarking framework leads to inconsistent evaluations, limits reproducibility, and increases engineering overhead, raising barriers to adoption for new researchers. To address these challenges, we introduce RL4CO, a unified and extensive benchmark with in-depth library coverage of 27 CO problem environments and 23 state-of-the-art baselines. Built on efficient software libraries and best practices in implementation, RL4CO features modularized implementation and flexible configurations of diverse environments, policy architectures, RL algorithms, and utilities with extensive documentation. RL4CO helps researchers build on existing successes while exploring and developing their own designs, facilitating the entire research process by decoupling science from heavy engineering. We finally provide extensive benchmark studies to inspire new insights and future work. RL4CO has already attracted numerous researchers in the community and is open-sourced at https://github.com/ai4co/rl4co.

cs.LG↗

arXiv2025

Non-Adaptive Adversarial Face Generation

Adversarial attacks on face recognition systems (FRSs) pose serious security and privacy threats, especially when these systems are used for identity verification. In this paper, we propose a novel method for generating adversarial faces-synthetic facial images that are visually distinct yet recognized as a target identity by the FRS. Unlike iterative optimization-based approaches (e.g., gradient descent or other iterative solvers), our method leverages the structural characteristics of the FRS feature space. We figure out that individuals sharing the same attribute (e.g., gender or race) form an attributed subsphere. By utilizing such subspheres, our method achieves both non-adaptiveness and a remarkably small number of queries. This eliminates the need for relying on transferability and open-source surrogate models, which have been a typical strategy when repeated adaptive queries to commercial FRSs are impossible. Despite requiring only a single non-adaptive query consisting of 100 face images, our method achieves a high success rate of over 93% against AWS's CompareFaces API at its default threshold. Furthermore, unlike many existing attacks that perturb a given image, our method can deliberately produce adversarial faces that impersonate the target identity while exhibiting high-level attributes chosen by the adversary.

cs.CV↗

arXiv2025

Dark states of electrons in a quantum system with two pairs of sublattices

A quantum state of matter that is forbidden to interact with photons and is therefore undetectable by spectroscopic means is called a dark state. This basic concept can be applied to condensed matter where it suggests that a whole band of quantum states could be undetectable across a full Brillouin zone. Here we report the discovery of such condensed matter dark states in palladium diselenide as a model system that has two pairs of sublattices in the primitive cell. By using angle-resolved photoemission spectroscopy, we find valence bands that are practically unobservable over the whole Brillouin zone at any photon energy, polarisation, and scattering plane. Our model shows that two pairs of sublattices located at half-translation positions and related by multiple glide-mirror symmetries make their relative quantum phases polarised into only four kinds, three of which become dark due to double destructive interference. This mechanism is generic to other systems with two pairs of sublattices, and we show how the phenomena observed in cuprates, lead-halide perovskites, and density wave systems can be resolved by the mechanism of dark states. Our results suggest that the sublattice degree of freedom, which has been overlooked so far, should be considered in the study of correlated phenomena and optoelectronic characteristics.

cond-mat.supr-con↗