Author name cluster

Dongsoo Har

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

6 papers

2 author rows

AAAI Conference 2026 Short Paper

Dynamics-Aware Planning Representation for Zero-Shot Reinforcement Learning (Student Abstract)

Jungho An
Taeyoung Kim
Haeun Kim
Dongsoo Har

Offline Zero-Shot Reinforcement Learning requires an agent to solve unseen tasks using only a fixed offline dataset without explicit rewards. A central challenge is learning representations that capture both high-level long-term planning and low-level physical dynamics. We propose a novel framework, Dynamics-Aware Planning Representation (DAPR), which disentangles these two aspects via complementary contrastive objectives. Specifically, DAPR learns goal-oriented planning directions and local dynamics-consistent directions in the latent space. By jointly enforcing these constraints, DAPR yields representations that balance “where to go” with “how to move.” Experiments on standard locomotion benchmarks (Walker, Cheetah, Quadruped) demonstrate that DAPR consistently improves performance and generalization over strong baselines, achieving substantial gains on precision demanding tasks.

PDF Details DOI

AAAI Conference 2026 Short Paper

Steering Sparse Autoencoder Latents to Control Dynamic Head Pruning in Vision Transformers (Student Abstract)

Yousung Lee
Dongsoo Har

Dynamic head pruning in Vision Transformers (ViTs) improves efficiency by removing redundant attention heads, but existing pruning policies are often difficult to interpret and control. In this work, we propose a novel framework by integrating Sparse Autoencoders (SAEs) with dynamic pruning, leveraging their ability to disentangle dense embeddings into interpretable and controllable sparse latents. Specifically, we train an SAE on the final-layer residual embedding of the ViT and amplify the sparse latents with different strategies to alter pruning decisions. Among them, per-class steering reveals compact, class-specific head subsets that preserve accuracy. For example, bowl improves accuracy (76%→82%) while reducing head usage (0.72→0.33) via heads h2 and h5. These results show that sparse latent features enable class-specific control of dynamic pruning, effectively bridging pruning efficiency and mechanistic interpretability in ViTs.

PDF Details DOI

AAAI Conference 2024 Short Paper

Cluster-Based Sampling in Hindsight Experience Replay for Robotic Tasks (Student Abstract)

Taeyoung Kim
Dongsoo Har

In multi-goal reinforcement learning with a sparse binary reward, training agents is particularly challenging, due to a lack of successful experiences. To solve this problem, hindsight experience replay (HER) generates successful experiences even from unsuccessful ones. However, generating successful experiences from uniformly sampled ones is not an efficient process. In this paper, the impact of exploiting the property of achieved goals in generating successful experiences is investigated and a novel cluster-based sampling strategy is proposed. The proposed sampling strategy groups episodes with different achieved goals by using a cluster model and samples experiences in the manner of HER to create the training batch. The proposed method is validated by experiments with three robotic control tasks of the OpenAI Gym. The results of experiments demonstrate that the proposed method is substantially sample efficient and achieves better performance than baseline approaches.

PDF Details DOI

AAAI Conference 2024 Short Paper

Enhanced Optical Character Recognition by Optical Sensor Combined with BERT and Cosine Similarity Scoring (Student Abstract)

Woohyeon Moon
Sarvar Nengroo
Taeyoung Kim
Jihui Lee
Seungah Son
Dongsoo Har

Optical character recognition(OCR) is the technology to identify text characters embedded within images. Conventional OCR models exhibit performance degradation when performing with noisy images. To solve this problem, we propose a novel model, which combines computer vision using optical sensor with natural language processing by bidirectional encoder representations from transformers(BERT) and cosine similarity scoring. The proposed model uses a confidence rate to determine whether to utilize optical sensor alone or BERT/cosine similarity scoring combined with the optical sensor. Experimental results show that the proposed model outperforms approximately 4.34 times better than the conventional OCR.

PDF Details DOI

ECAI Conference 2024 Conference Paper

Robust Monocular Depth Estimation in Adverse Weather Conditions by Unsupervised Domain Adaptation

Jihui Lee
Quoc-Vinh Lai-Dang
Neha Sengar
Dongsoo Har

Robust monocular depth estimation is essential for various applications relying on visual cues to understand the real world. To ensure robustness, unsupervised domain adaptation is widely used for monocular depth estimation. Despite recent advances, existing methods often struggle in outdoor environments due to adverse environmental conditions and limited datasets. Intentionally corrupted images obtained from real images captured in clear weather conditions for unsupervised domain adaptation often fail to accurately represent the complex characteristics of diverse environments, leading to unrealistic training data. From this viewpoint, simulation data offering more plausible representation of adverse weather conditions are used. However, it still presents drawbacks due to potentially degrading adaptation capabilities. To address the limitations of using simulation data, we propose a wild-condition pass filtering module that extracts wild-condition features and captures cross-domain relationships from both real and simulation datasets. This enables comprehensive learning of different conditions from each dataset and improved performance on real adversarial target images. The proposed method achieves a notable 22% improvement over the baseline on the Foggy Cityscapes dataset, highlighting the importance of employing realistic domain adaptation techniques to effectively address the challenges posed by adverse environmental conditions. The code is available at https: //github. com/JH2-LEE/wide.

Details

AAAI Conference 2024 Short Paper

Virtual Action Actor-Critic Framework for Exploration (Student Abstract)

Bumgeun Park
Taeyoung Kim
Quoc-Vinh Lai-Dang
Dongsoo Har

Efficient exploration for an agent is challenging in reinforcement learning (RL). In this paper, a novel actor-critic framework namely virtual action actor-critic (VAAC), is proposed to address the challenge of efficient exploration in RL. This work is inspired by humans' ability to imagine the potential outcomes of their actions without actually taking them. In order to emulate this ability, VAAC introduces a new actor called virtual actor (VA), alongside the conventional actor-critic framework. Unlike the conventional actor, the VA takes the virtual action to anticipate the next state without interacting with the environment. With the virtual policy following a Gaussian distribution, the VA is trained to maximize the anticipated novelty of the subsequent state resulting from a virtual action. If any next state resulting from available actions does not exhibit high anticipated novelty, training the VA leads to an increase in the virtual policy entropy. Hence, high virtual policy entropy represents that there is no room for exploration. The proposed VAAC aims to maximize a modified Q function, which combines cumulative rewards and the negative sum of virtual policy entropy. Experimental results show that the VAAC improves the exploration performance compared to existing algorithms.

PDF Details DOI