Jaehyun Park

Giyeong Oh, Junghyun Lee, Jaehyun Park et al.

Modern LLMs inherit strong priors from web-scale pretraining, which can limit the headroom of post-training data-selection strategies. While Active Preference Learning (APL) seeks to optimize query efficiency in online Direct Preference Optimization (DPO), the inherent richness of on-policy candidate pools often renders simple Random sampling a surprisingly formidable baseline. We evaluate uncertainty-based APL against Random across harmlessness, helpfulness, and instruction-following settings, utilizing both reward models and LLM-as-a-judge proxies. We find that APL yields negligible improvements in proxy win-rates compared to Random. Crucially, we observe a dissociation where win-rate improves even as general capability -- measured by standard benchmarks -- degrades. APL fails to mitigate this capability collapse or reduce variance significantly better than random sampling. Our findings suggest that in the regime of strong pre-trained priors, the computational overhead of active selection is difficult to justify against the ``cheap diversity'' provided by simple random samples. Our code is available at https://github.com/BootsofLagrangian/random-vs-apl.

Jaehyun Park, Minyoung Ahn, Minkyu Kim et al.

Despite recent advances in diffusion models, AI generated images still often contain visual artifacts that compromise realism. Although more thorough pre-training and bigger models might reduce artifacts, there is no assurance that they can be completely eliminated, which makes artifact mitigation a highly crucial area of study. Previous artifact-aware methodologies depend on human-labeled artifact datasets, which are costly and difficult to scale, underscoring the need for an automated approach to reliably acquire artifact-annotated datasets. In this paper, we propose ArtiAgent, which efficiently creates pairs of real and artifact-injected images. It comprises three agents: a perception agent that recognizes and grounds entities and subentities from real images, a synthesis agent that introduces artifacts via artifact injection tools through novel patch-wise embedding manipulation within a diffusion transformer, and a curation agent that filters the synthesized artifacts and generates both local and global explanations for each instance. Using ArtiAgent, we synthesize 100K images with rich artifact annotations and demonstrate both efficacy and versatility across diverse applications. Code is available at link.

Jaehyun Park, Jaegyun Im, Sanha Hwang et al.

In the pursuit of artificial general intelligence (AGI), we tackle Abstraction and Reasoning Corpus (ARC) tasks using a novel two-pronged approach. We employ the Decision Transformer in an imitation learning paradigm to model human problem-solving, and introduce an object detection algorithm, the Push and Pull clustering method. This dual strategy enhances AI's ARC problem-solving skills and provides insights for AGI progression. Yet, our work reveals the need for advanced data collection tools, robust training datasets, and refined model structures. This study highlights potential improvements for Decision Transformers and propels future AGI research.

Alish Kanani, Sangwan Lee, Han Lyu et al.

Large language models operate in distinct compute-bound prefill followed by memory bandwidth-bound decode phases. Hybrid Mamba-Transformer models inherit this asymmetry while adding state space model (SSM) recurrences and element-wise operations that map poorly to matmul-centric accelerators. This mismatch causes performance bottlenecks, showing that a homogeneous architecture cannot satisfy all requirements. We introduce DUET, a disaggregated accelerator that assigns prefill and decode phases to specialized packages. The Prefill package utilizes systolic array chiplets with off-package memory for efficient large matrix multiplications and long-sequence SSMs. The Decode package utilizes vector-unit arrays with high-bandwidth in-package memory to accelerate token-by-token SSM and vector-matrix multiplications. Both architectures are runtime-configurable to support hybrid models with mixed Mamba and attention layers. Evaluations on Nemotron-H-56B, Zamba2-7B, and Llama3-8B across four workloads show that DUET achieves 4x faster time to first token, 1.4x higher throughput, and 1.5x lower time between tokens over the B200 GPU.

Jaehyun Park, Konyul Park, Daehun Kim et al.

In autonomous driving, transparency in the decision-making of perception models is critical, as even a single misperception can be catastrophic. Yet with multi-sensor inputs, it is difficult to determine how each modality contributes to a prediction because sensor information becomes entangled within the fusion network. We introduce Layer-Wise Modality Decomposition (LMD), a post-hoc, model-agnostic interpretability method that disentangles modality-specific information across all layers of a pretrained fusion model. To our knowledge, LMD is the first approach to attribute the predictions of a perception model to individual input modalities in a sensor-fusion system for autonomous driving. We evaluate LMD on pretrained fusion models under camera-radar, camera-LiDAR, and camera-radar-LiDAR settings for autonomous driving. Its effectiveness is validated using structured perturbation-based metrics and modality-wise visual decompositions, demonstrating practical applicability to interpreting high-capacity multimodal architectures. Code is available at https://github.com/detxter-jvb/Layer-Wise-Modality-Decomposition.

Jaehyun Park, Subin Lee, Eon Kim et al.

Feature fusion modules from encoder and self-attention module have been adopted in semantic segmentation. However, the computation of these modules is costly and has operational limitations in real-time environments. In addition, segmentation performance is limited in autonomous driving environments with a lot of contextual information perpendicular to the road surface, such as people, buildings, and general objects. In this paper, we propose an efficient feature fusion method, Feature Fusion with Different Norms (FFDN) that utilizes rich global context of multi-level scale and vertical pooling module before self-attention that preserves most contextual information while reducing the complexity of global context encoding in the vertical direction. By doing this, we could handle the properties of representation in global space and reduce additional computational cost. In addition, we analyze low performance in challenging cases including small and vertically featured objects. We achieve the mean Interaction of-union(mIoU) of 73.1 and the Frame Per Second(FPS) of 191, which are comparable results with state-of-the-arts on Cityscapes test datasets.

Daniel Rho, TaeSoo Kim, Sooill Park et al.

Contrastive learning, along with its variations, has been a highly effective self-supervised learning method across diverse domains. Contrastive learning measures the distance between representations using cosine similarity and uses cross-entropy for representation learning. Within the same framework of cosine-similarity-based representation learning, margins have played a significant role in enhancing face and speaker recognition tasks. Interestingly, despite the shared reliance on the same similarity metrics and objective functions, contrastive learning has not actively adopted margins. Furthermore, decision-boundary-based explanations are the only ones that have been used to explain the effect of margins in contrastive learning. In this work, we propose a new perspective to understand the role of margins based on gradient analysis. Based on the new perspective, we analyze how margins affect gradients of contrastive learning and separate the effect into more elemental levels. We separately analyze each and provide possible directions for improving contrastive learning. Our experimental results demonstrate that emphasizing positive samples and scaling gradients depending on positive sample angles and logits are the keys to improving the generalization performance of contrastive learning in both seen and unseen datasets, and other factors can only marginally improve performance.

Jiyong Kim, Jaeho Lee, Jiahao Lin et al.

State Space Model (SSM)-based machine learning architectures have recently gained significant attention for processing sequential data. Mamba, a recent sequence-to-sequence SSM, offers competitive accuracy with superior computational efficiency compared to state-of-the-art transformer models. While this advantage makes Mamba particularly promising for resource-constrained edge devices, no hardware acceleration frameworks are currently optimized for deploying it in such environments. This paper presents eMamba, a comprehensive end-to-end hardware acceleration framework explicitly designed for deploying Mamba models on edge platforms. eMamba maximizes computational efficiency by replacing complex normalization layers with lightweight hardware-aware alternatives and approximating expensive operations, such as SiLU activation and exponentiation, considering the target applications. Then, it performs an approximation-aware neural architecture search (NAS) to tune the learnable parameters used during approximation. Evaluations with Fashion-MNIST, CIFAR-10, and MARS, an open-source human pose estimation dataset, show eMamba achieves comparable accuracy to state-of-the-art techniques using 1.63-19.9$\times$ fewer parameters. In addition, it generalizes well to large-scale natural language tasks, demonstrating stable perplexity across varying sequence lengths on the WikiText2 dataset. We also quantize and implement the entire eMamba pipeline on an AMD ZCU102 FPGA and ASIC using GlobalFoundries (GF) 22 nm technology. Experimental results show 4.95-5.62$\times$ lower latency and 2.22-9.95$\times$ higher throughput, with 4.77$\times$ smaller area, 9.84$\times$ lower power, and 48.6$\times$ lower energy consumption than baseline solutions while maintaining competitive accuracy.

Jin Woo Lee, Jaehyun Park, Min Jun Choi et al.

While significant advancements have been made in music generation and differentiable sound synthesis within machine learning and computer audition, the simulation of instrument vibration guided by physical laws has been underexplored. To address this gap, we introduce a novel model for simulating the spatio-temporal motion of nonlinear strings, integrating modal synthesis and spectral modeling within a neural network framework. Our model leverages physical properties and fundamental frequencies as inputs, outputting string states across time and space that solve the partial differential equation characterizing the nonlinear string. Empirical evaluations demonstrate that the proposed architecture achieves superior accuracy in string motion simulation compared to existing baseline architectures. The code and demo are available online.

Sowoo Lee, Dongyun Kang, Jaehyun Park et al.

This paper introduces DynaFlow, a novel framework that embeds a differentiable simulator directly into a flow matching model. By generating trajectories in the action space and mapping them to dynamically feasible state trajectories via the simulator, DynaFlow ensures all outputs are physically consistent by construction. This end-to-end differentiable architecture enables training on state-only demonstrations, allowing the model to simultaneously generate physically consistent state trajectories while inferring the underlying action sequences required to produce them. We demonstrate the effectiveness of our approach through quantitative evaluations and showcase its real-world applicability by deploying the generated actions onto a physical Go1 quadruped robot. The robot successfully reproduces diverse gait present in the dataset, executes long-horizon motions in open-loop control and translates infeasible kinematic demonstrations into dynamically executable, stylistic behaviors. These hardware experiments validate that DynaFlow produces deployable, highly effective motions on real-world hardware from state-only demonstrations, effectively bridging the gap between kinematic data and real-world execution.

Konyul Park, Daehun Kim, Jiyong Oh et al.

Reliable off-road navigation requires accurate estimation of traversable regions and robust perception under diverse terrain and sensing conditions. However, existing datasets lack both scalability and multi-modality, which limits progress in 3D traversability prediction. In this work, we introduce STONE, a large-scale multi-modal dataset for off-road navigation. STONE provides (1) trajectory-guided 3D traversability maps generated by a fully automated, annotation-free pipeline, and (2) comprehensive surround-view sensing with synchronized 128-channel LiDAR, six RGB cameras, and three 4D imaging radars. The dataset covers a wide range of environments and conditions, including day and night, grasslands, farmlands, construction sites, and lakes. Our auto-labeling pipeline reconstructs dense terrain surfaces from LiDAR scans, extracts geometric attributes such as slope, elevation, and roughness, and assigns traversability labels beyond the robot's trajectory using a Mahalanobis-distance-based criterion. This design enables scalable, geometry-aware ground-truth construction without manual annotation. Finally, we establish a benchmark for voxel-level 3D traversability prediction and provide strong baselines under both single-modal and multi-modal settings. STONE is available at: https://konyul.github.io/STONE-dataset/

Emanuele Venieri, Simone Manoni, Alberto Florian et al.

High Bandwidth Memory with Processing-in-Memory (HBM-PIM) offers an opportunity to reduce data movement by executing computation directly inside memory, but current commercial platforms expose limited instruction sets and require specialized software stacks. In this work, we investigate whether HBM-PIM can serve as a backend for ISA-level matrix acceleration, using the RISC-V Attached Matrix Extension (AME) as a semantic reference. We propose a PEP-based execution model that maps AME element-wise and matrix instructions to HBM-PIM micro-kernels and data instructions in memory operations. Differently from SoA HBM-PIM, we introduce a reduction-free outer-product dataflow that enables accumulation entirely within memory despite the lack of native reduction support. Our approach supports end-to-end execution of element-wise operations, GEMV, and GEMM in PIM mode, minimizing host involvement and off-chip transfers. An experimental evaluation on Samsung Aquabolt-XL shows that AME matrix tile multiplication achieves up to 14.9 GFLOP/s (59.4 FLOP/cycle) on a single HBM pseudo-channel.

Jaehyun Park, Dongmin Park, Jae-Gil Lee

Continual learning (CL) enables deep neural networks to adapt to ever-changing data distributions. In practice, there may be scenarios where annotation is costly, leading to active continual learning (ACL), which performs active learning (AL) for the CL scenarios when reducing the labeling cost by selecting the most informative subset is preferable. However, conventional AL strategies are not suitable for ACL, as they focus solely on learning the new knowledge, leading to catastrophic forgetting of previously learned tasks. Therefore, ACL requires a new AL strategy that can balance the prevention of catastrophic forgetting and the ability to quickly learn new tasks. In this paper, we propose AccuACL, Accumulated informativeness-based Active Continual Learning, by the novel use of the Fisher information matrix as a criterion for sample selection, derived from a theoretical analysis of the Fisher-optimality preservation properties within the framework of ACL, while also addressing the scalability issue of Fisher information-based AL. Extensive experiments demonstrate that AccuACL significantly outperforms AL baselines across various CL algorithms, increasing the average accuracy and forgetting by 23.8% and 17.0%, respectively, on average.

Yunho Kim, Jaehyun Park, Heejun Kim et al.

Effective long-term strategies enable AI systems to navigate complex environments by making sequential decisions over extended horizons. Similarly, reinforcement learning (RL) agents optimize decisions across sequences to maximize rewards, even without immediate feedback. To verify that Latent Diffusion-Constrained Q-learning (LDCQ), a prominent diffusion-based offline RL method, demonstrates strong reasoning abilities in multi-step decision-making, we aimed to evaluate its performance on the Abstraction and Reasoning Corpus (ARC). However, applying offline RL methodologies to enhance strategic reasoning in AI for solving tasks in ARC is challenging due to the lack of sufficient experience data in the ARC training set. To address this limitation, we introduce an augmented offline RL dataset for ARC, called Synthesized Offline Learning Data for Abstraction and Reasoning (SOLAR), along with the SOLAR-Generator, which generates diverse trajectory data based on predefined rules. SOLAR enables the application of offline RL methods by offering sufficient experience data. We synthesized SOLAR for a simple task and used it to train an agent with the LDCQ method. Our experiments demonstrate the effectiveness of the offline RL approach on a simple ARC task, showing the agent's ability to make multi-step sequential decisions and correctly identify answer states. These results highlight the potential of the offline RL approach to enhance AI's strategic reasoning capabilities.

Jaehyun Park, Yunho Kim, Sejin Kim et al.

We propose a novel offline reinforcement learning (offline RL) approach, introducing the Diffusion-model-guided Implicit Q-learning with Adaptive Revaluation (DIAR) framework. We address two key challenges in offline RL: out-of-distribution samples and long-horizon problems. We leverage diffusion models to learn state-action sequence distributions and incorporate value functions for more balanced and adaptive decision-making. DIAR introduces an Adaptive Revaluation mechanism that dynamically adjusts decision lengths by comparing current and future state values, enabling flexible long-term decision-making. Furthermore, we address Q-value overestimation by combining Q-network learning with a value function guided by a diffusion model. The diffusion model generates diverse latent trajectories, enhancing policy robustness and generalization. As demonstrated in tasks like Maze2D, AntMaze, and Kitchen, DIAR consistently outperforms state-of-the-art algorithms in long-horizon, sparse-reward environments.

Wongyu Kim, Hochang Lee, Sanghak Lee et al.

Query augmentation makes queries more meaningful by appending further information to the queries to find relevant documents. Current studies have proposed Large Language Model (LLM)-based embedders, which learn representation for embedding and generation for query augmentation in a multi-task manner by leveraging the generative capabilities of LLM. During inference, these jointly trained embedders have conducted query augmentation followed by embedding, showing effective results. However, augmenting every query leads to substantial embedding latency and query augmentation can be detrimental to performance for some queries. Also, previous methods have not been explored in multimodal environments. To tackle these problems, we propose M-Solomon, a universal multimodal embedder that can adaptively determine when to augment queries. Our approach first divides the queries of the training datasets into two groups at the dataset level. One includes queries that require augmentation and the other includes queries that do not. Then, we introduces a synthesis process that generates appropriate augmentations for queries that require them by leveraging a powerful Multimodal LLM (MLLM). Next, we present adaptive query augmentation. Through this step, M-Solomon can conduct query augmentation only when necessary by learning to generate synthetic augmentations with the prefix /augment for queries that demand them and to generate the simple string /embed for others. Experimental results showed that M-Solomon not only surpassed the baseline without augmentation by a large margin but also outperformed the baseline that always used augmentation, providing much faster embedding latency.

Jaehyun Park, Junyeop Kwon, Dabeen Lee

We study model-based reinforcement learning with non-linear function approximation where the transition function of the underlying Markov decision process (MDP) is given by a multinomial logistic (MNL) model. We develop a provably efficient discounted value iteration-based algorithm that works for both infinite-horizon average-reward and discounted-reward settings. For average-reward communicating MDPs, the algorithm guarantees a regret upper bound of $\tilde{\mathcal{O}}(dD\sqrt{T})$ where $d$ is the dimension of feature mapping, $D$ is the diameter of the underlying MDP, and $T$ is the horizon. For discounted-reward MDPs, our algorithm achieves $\tilde{\mathcal{O}}(d(1-γ)^{-2}\sqrt{T})$ regret where $γ$ is the discount factor. Then we complement these upper bounds by providing several regret lower bounds. We prove a lower bound of $Ω(d\sqrt{DT})$ for learning communicating MDPs of diameter $D$ and a lower bound of $Ω(d(1-γ)^{3/2}\sqrt{T})$ for learning discounted-reward MDPs with discount factor $γ$. Lastly, we show a regret lower bound of $Ω(dH^{3/2}\sqrt{K})$ for learning $H$-horizon episodic MDPs with MNL function approximation where $K$ is the number of episodes, which improves upon the best-known lower bound for the finite-horizon setting.

Dohoon Lee, Jaehyun Park, Hyunwoo J. Kim et al.

Flow and diffusion models have demonstrated strong performance and training stability across various tasks but lack two critical properties of simulation-based methods: freedom of dimensionality and adaptability to different inference trajectories. To address this limitation, we propose the Multidimensional Adaptive Coefficient (MAC), a plug-in module for flow and diffusion models that extends conventional unidimensional coefficients to multidimensional ones and enables inference trajectory-wise adaptation. MAC is trained via simulation-based feedback through adversarial refinement. Empirical results across diverse frameworks and datasets demonstrate that MAC enhances generative quality with high training efficiency. Consequently, our work offers a new perspective on inference trajectory optimality, encouraging future research to move beyond vector field design and to leverage training-efficient, simulation-based optimization.

Jaehyun Park, Chansoo Kim, Kichun Jo

The autonomous car must recognize the driving environment quickly for safe driving. As the Light Detection And Range (LiDAR) sensor is widely used in the autonomous car, fast semantic segmentation of LiDAR point cloud, which is the point-wise classification of the point cloud within the sensor framerate, has attracted attention in recognition of the driving environment. Although the voxel and fusion-based semantic segmentation models are the state-of-the-art model in point cloud semantic segmentation recently, their real-time performance suffer from high computational load due to high voxel resolution. In this paper, we propose the fast voxel-based semantic segmentation model using Point Convolution and 3D Sparse Convolution (PCSCNet). The proposed model is designed to outperform at both high and low voxel resolution using point convolution-based feature extraction. Moreover, the proposed model accelerates the feature propagation using 3D sparse convolution after the feature extraction. The experimental results demonstrate that the proposed model outperforms the state-of-the-art real-time models in semantic segmentation of SemanticKITTI and nuScenes, and achieves the real-time performance in LiDAR point cloud inference.

Michael Jaemin Kim, Jaehyun Park, Yeonhong Park et al.

Since its public introduction in the mid-2010s, the Row Hammer (RH) phenomenon has drawn significant attention from the research community due to its security implications. Although many RH-protection schemes have been proposed by processor vendors, DRAM manufacturers, and academia, they still have shortcomings. Solutions implemented in the memory controller (MC) incur increasingly higher costs due to their conservative design for the worst case in terms of the number of DRAM banks and RH threshold to support. Meanwhile, DRAM-side implementation either has a limited time margin for RH-protection measures or requires extensive modifications to the standard DRAM interface. Recently, a new command for RH-protection has been introduced in the DDR5/LPDDR5 standards, referred to as refresh management (RFM). RFM enables the separation of the tasks for RHprotection to both MC and DRAM by having the former generate an RFM command at a specific activation frequency and the latter take proper RH-protection measures within a given time window. Although promising, no existing study presents and analyzes RFM-based solutions for RH-protection. In this paper, we propose Mithril, the first RFM interfacecompatible, DRAM-MC cooperative RH-protection scheme providing deterministic protection guarantees. Mithril has minimal energy overheads for common use cases without adversarial memory access patterns. We also introduce Mithril+, an optional extension to provide minimal performance overheads at the expense of a tiny modification to the MC, while utilizing existing DRAM commands.

Yigit Tuncel, Ganapati Bhat, Jaehyun Park et al.

Energy harvesting offers an attractive and promising mechanism to power low-energy devices. However, it alone is insufficient to enable an energy-neutral operation, which can eliminate tedious battery charging and replacement requirements. Achieving an energy-neutral operation is challenging since the uncertainties in harvested energy undermine the quality of service requirements. To address this challenge, we present a runtime energy-allocation framework that optimizes the utility of the target device under energy constraints using a rollout algorithm, which is a sequential approach to solve dynamic optimization problems. The proposed framework uses an efficient iterative algorithm to compute initial energy allocations at the beginning of a day. The initial allocations are then corrected at every interval to compensate for the deviations from the expected energy harvesting pattern. We evaluate this framework using solar and motion energy harvesting modalities and American Time Use Survey data from 4772 different users. Compared to prior techniques, the proposed framework achieves up to 35% higher utility even under energy-limited scenarios. Moreover, measurements on a wearable device prototype show that the proposed framework has 1000x smaller energy overhead than iterative approaches with a negligible loss in utility.