Saemi Moon

Seunghyuk Cho, Sunghyun Choi, Jaeseung Heo et al.

Multi-agent systems (MAS) have substantially advanced autonomous software engineering (SWE), but their growing inference energy demands raise sustainability concerns. In this paper, we demonstrate that this cost is concentrated in an overlooked source: redundant output tokens generated across agents. Two empirical findings ground this claim. First, our per-token energy attribution for MAS reveals a sharp asymmetry: an output token consumes 30 to 1,000 times more energy than an input or cached token. Second, MAS inflate per-episode output because agents repeatedly re-explore overlapping repository regions. To address this inefficiency, we propose Librarian, a persistent search sub-agent that tracks repository-search history and suppresses redundant exploration actions across agents. By returning short references to file regions instead of full file excerpts, Librarian further reduces output-token volume. On SWE-Bench Verified, Librarian reduces per-episode GPU energy consumption of existing multi-agent SWE systems by up to 25% while preserving task performance.

Saemi Moon, Suhyeon Jun, Seoyeon Lee et al.

Concept unlearning aims to erase a target concept from a pretrained text-to-image diffusion model without retraining. Closed-form methods are attractive in this setting because they apply a single deterministic edit to the cross-attention weights and add no inference-time cost. Existing closed-form methods, however, represent the target concept through the text encoder's response to a few short anchor prompts that name it, and paraphrased prompts that evoke the concept without naming it consistently bypass the edit. We argue that the target should instead be represented in the cross-attention activation space. Text embeddings describe the user's prompt, while cross-attention activations describe what the model is about to render, and the latter generalize to paraphrase the anchor templates do not cover. Building on this observation, we propose PURE (Projection in U-Net Rendering for Erasure), a closed-form method that builds the forget and retain bases from per-layer cross-attention activations captured along a short denoising trajectory and applies a single linear projector to the cross-attention key and value weights. On a recent holistic concept-unlearning benchmark covering ten concepts across artistic style, intellectual property, celebrity, and NSFW categories, PURE significantly reduces target leakage under paraphrased and adversarial prompts while preserving retain concepts close to the unedited model, yielding the best overall forget-retain trade-off among evaluated methods.

Saemi Moon, Seunghyuk Cho, Dongwoo Kim

We tackle the problem of feature unlearning from a pre-trained image generative model: GANs and VAEs. Unlike a common unlearning task where an unlearning target is a subset of the training set, we aim to unlearn a specific feature, such as hairstyle from facial images, from the pre-trained generative models. As the target feature is only presented in a local region of an image, unlearning the entire image from the pre-trained model may result in losing other details in the remaining region of the image. To specify which features to unlearn, we collect randomly generated images that contain the target features. We then identify a latent representation corresponding to the target feature and then use the representation to fine-tune the pre-trained model. Through experiments on MNIST, CelebA, and FFHQ datasets, we show that target features are successfully removed while keeping the fidelity of the original models. Further experiments with an adversarial attack show that the unlearned model is more robust under the presence of malicious parties.

Saemi Moon, Myeonghyeon Kim, Zhenyue Qin et al.

Skeleton-based action recognition attracts practitioners and researchers due to the lightweight, compact nature of datasets. Compared with RGB-video-based action recognition, skeleton-based action recognition is a safer way to protect the privacy of subjects while having competitive recognition performance. However, due to improvements in skeleton recognition algorithms as well as motion and depth sensors, more details of motion characteristics can be preserved in the skeleton dataset, leading to potential privacy leakage. We first train classifiers to categorize private information from skeleton trajectories to investigate the potential privacy leakage from skeleton datasets. Our preliminary experiments show that the gender classifier achieves 87% accuracy on average, and the re-identification classifier achieves 80% accuracy on average with three baseline models: Shift-GCN, MS-G3D, and 2s-AGCN. We propose an anonymization framework based on adversarial learning to protect potential privacy leakage from the skeleton dataset. Experimental results show that an anonymized dataset can reduce the risk of privacy leakage while having marginal effects on action recognition performance even with simple anonymizer architectures. The code used in our experiments is available at https://github.com/ml-postech/Skeleton-anonymization/

Youngbin Choi, Minjong Lee, Saemi Moon et al.

Large language models (LLMs) are increasingly studied in the context of multi-turn reasoning, where models iteratively refine their outputs based on user-provided feedback. Such settings are crucial for tasks that require complex reasoning, yet existing feedback paradigms often rely on issuing new messages. LLMs struggle to integrate these reliably, leading to inconsistent improvements. In this work, we introduce in-place feedback, a novel interaction paradigm in which users directly edit an LLM's previous response, and the model conditions on this modified response to generate its revision. Empirical evaluations on diverse reasoning-intensive benchmarks reveal that in-place feedback achieves better performance than conventional multi-turn feedback while using $79.1\%$ fewer tokens. Complementary analyses on controlled environments further demonstrate that in-place feedback resolves a core limitation of multi-turn feedback: models often fail to apply feedback precisely to erroneous parts of the response, leaving errors uncorrected and sometimes introducing new mistakes into previously correct content. These findings suggest that in-place feedback offers a more natural and effective mechanism for guiding LLMs in reasoning-intensive tasks.