Kyung-Joong Kim

Hoyoun Jung, Kyung-Joong Kim

Compressed prompts aid instruction-tuned language models (LMs) in overcoming context window limitations and reducing computational costs. Existing methods, which primarily based on training embeddings, face various challenges associated with interpretability, the fixed number of embedding tokens, reusability across different LMs, and inapplicability when interacting with black-box APIs. This study proposes prompt compression with reinforcement learning (PCRL), which is a discrete prompt compression method that addresses these issues. The proposed PCRL method utilizes a computationally efficient policy network that edits prompts directly. The training approach employed in the proposed PCRLs can be applied flexibly to various types of LMs, including both decoder-only and encoder-decoder architecture and it can be trained without gradient access to the LMs or labeled data. The proposed PCRL achieves an average reduction of 24.6% in terms of the token count across various instruction prompts while maintaining sufficient performance. In addition, we demonstrate that the learned policy can be transferred to larger LMs, and through a comprehensive analysis, we explore the token importance within the prompts. Our code is accessible at https://github.com/nenomigami/PromptCompressor.

Hyeon-Chang Jeon, In-Chang Baek, Cheong-mok Bae et al.

The balance of game content significantly impacts the gaming experience. Unbalanced game content diminishes engagement or increases frustration because of repetitive failure. Although game designers intend to adjust the difficulty of game content, this is a repetitive, labor-intensive, and challenging process, especially for commercial-level games with extensive content. To address this issue, the game research community has explored automated game balancing using artificial intelligence (AI) techniques. However, previous studies have focused on limited game content and did not consider the importance of the generalization ability of playtesting agents when encountering content changes. In this study, we propose RaidEnv, a new game simulator that includes diverse and customizable content for the boss raid scenario in MMORPG games. Additionally, we design two benchmarks for the boss raid scenario that can aid in the practical application of game AI. These benchmarks address two open problems in automatic content balancing, and we introduce two evaluation metrics to provide guidance for AI in automatic content balancing. This novel game research platform expands the frontiers of automatic game balancing problems and offers a framework within a realistic game production pipeline.

Jaeyoung Moon, Youjin Choi, Yucheon Park et al.

Self-annotation is the gold standard for collecting affective state labels in affective computing. Existing methods typically rely on full annotation, requiring users to continuously label affective states across entire sessions. While this process yields fine-grained data, it is time-consuming, cognitively demanding, and prone to fatigue and errors. To address these issues, we present PREFAB, a low-budget retrospective self-annotation method that targets affective inflection regions rather than full annotation. Grounded in the peak-end rule and ordinal representations of emotion, PREFAB employs a preference-learning model to detect relative affective changes, directing annotators to label only selected segments while interpolating the remainder of the stimulus. We further introduce a preview mechanism that provides brief contextual cues to assist annotation. We evaluate PREFAB through a technical performance study and a 25-participant user study. Results show that PREFAB outperforms baselines in modeling affective inflections while mitigating workload (and conditionally mitigating temporal burden). Importantly PREFAB improves annotator confidence without degrading annotation quality.

Isaac Han, Sangyeon Park, Seungwon Oh et al.

Deep neural networks trained on nonstationary data must balance stability (i.e., retaining prior knowledge) and plasticity (i.e., adapting to new tasks). Standard reinitialization methods, which reinitialize weights toward their original values, are widely used but difficult to tune: conservative reinitializations fail to restore plasticity, while aggressive ones erase useful knowledge. We propose FIRE, a principled reinitialization method that explicitly balances the stability-plasticity tradeoff. FIRE quantifies stability through Squared Frobenius Error (SFE), measuring proximity to past weights, and plasticity through Deviation from Isometry (DfI), reflecting weight isotropy. The reinitialization point is obtained by solving a constrained optimization problem, minimizing SFE subject to DfI being zero, which is efficiently approximated by Newton-Schulz iteration. FIRE is evaluated on continual visual learning (CIFAR-10 with ResNet-18), language modeling (OpenWebText with GPT-0.1B), and reinforcement learning (HumanoidBench with SAC and Atari games with DQN). Across all domains, FIRE consistently outperforms both naive training without intervention and standard reinitialization methods, demonstrating effective balancing of the stability-plasticity tradeoff.

Hyeonchang Jeon, Kyung-Joong Kim

Ad-hoc team cooperation is the problem of cooperating with other players that have not been seen in the learning process. Recently, this problem has been considered in the context of Hanabi, which requires cooperation without explicit communication with the other players. While in self-play strategies cooperating on reinforcement learning (RL) process has shown success, there is the problem of failing to cooperate with other unseen agents after the initial learning is completed. In this paper, we categorize the results of ad-hoc team cooperation into Failure, Success, and Synergy and analyze the associated failures. First, we confirm that agents learning via RL converge to one strategy each, but not necessarily the same strategy and that these agents can deploy different strategies even though they utilize the same hyperparameters. Second, we confirm that the larger the behavioral difference, the more pronounced the failure of ad-hoc team cooperation, as demonstrated using hierarchical clustering and Pearson correlation. We confirm that such agents are grouped into distinctly different groups through hierarchical clustering, such that the correlation between behavioral differences and ad-hoc team performance is -0.978. Our results improve understanding of key factors to form successful ad-hoc team cooperation in multi-player games.

Won-Sang You, Tae-Gwan Ha, Seo-Young Lee et al.

Zero-shot human-AI coordination is the training of an ego-agent to coordinate with humans without human data. Most studies on zero-shot human-AI coordination have focused on enhancing the ego-agent's coordination ability in a given environment without considering the issue of generalization to unseen environments. Real-world applications of zero-shot human-AI coordination should consider unpredictable environmental changes and the varying coordination ability of co-players depending on the environment. Previously, the multi-agent UED (Unsupervised Environment Design) approach has investigated these challenges by jointly considering environmental changes and co-player policy in competitive two-player AI-AI scenarios. In this paper, our study extends a multi-agent UED approach to zero-shot human-AI coordination. We propose a utility function and co-player sampling for a zero-shot human-AI coordination setting that helps train the ego-agent to coordinate with humans more effectively than a previous multi-agent UED approach. The zero-shot human-AI coordination performance was evaluated in the Overcooked-AI environment, using human proxy agents and real humans. Our method outperforms other baseline models and achieves high performance in human-AI coordination tasks in unseen environments. The source code is available at https://github.com/Uwonsang/ACD_Human-AI

Hyeonchang Jeon, Kyungbeom Kim, Eugene Vinitsky et al.

Large-scale datasets and fast simulators have enabled improvements in driving policies that appear safe and robust, yet strong performance in nominal scenarios can still mask flawed reasoning and unsafe heuristics. Summary scores from closed-loop simulators do not give significant insight into the policy, making it difficult to determine whether they truly predict the motion of surrounding vehicles, how the ego vehicle generates future plans, or whether they merely rely on brittle heuristics that happen to succeed in nominal scenarios. To better understand the limits and weaknesses of driving policies, we focus on probing for forms of prediction, i.e., where surrounding vehicles will move next, and planning, i.e., understanding how to generate safe trajectories. We focus on these two capabilities because they reflect behaviors expected of effective driving policies, and use their presence or absence to assess policy quality across data-driven behavior cloning and simulation-driven reinforcement learning policies. To evaluate the presence of these capabilities, we investigate them as a function of scale, asking whether the closed-loop gains from larger datasets and longer simulation training reflect stronger prediction and planning or merely better behavioral heuristics. We use linear probing and targeted perturbations in both imitation learning and reinforcement learning models to track when these internal signals emerge, plateau, or fail. Despite good closed-loop performance, policies often fail to form timely surrounding-vehicle predictions during near-collision events, revealing a limitation in the predictive signals available for ego planning. Finally, causal intervention shows that correcting mistaken predictions improves ego planning toward safer trajectories.

Kyungbeom Kim, Seungwon Oh, Kyung-Joong Kim

Parameter sharing is a key strategy in multi-agent reinforcement learning (MARL) for improving scalability, yet conventional fully shared architectures often collapse into homogeneous behaviors. Recent methods introduce diversity through clustering, pruning, or masking, but typically compromise resource efficiency. We propose Prism, a parameter sharing framework that induces inter-agent diversity by representing shared networks in the spectral domain via singular value decomposition (SVD). All agents share the singular vector directions while learning distinct spectral masks on singular values. This mechanism encourages inter-agent diversity and preserves scalability. Extensive experiments on both homogeneous (LBF, SMACv2) and heterogeneous (MaMuJoCo) benchmarks show that Prism achieves competitive performance with superior resource efficiency.

In-Chang Baek, Sung-Hyun Kim, Sam Earle et al.

Reward design plays a pivotal role in the training of game AIs, requiring substantial domain-specific knowledge and human effort. In recent years, several studies have explored reward generation for training game agents and controlling robots using large language models (LLMs). In the content generation literature, there has been early work on generating reward functions for reinforcement learning agent generators. This work introduces PCGRLLM, an extended architecture based on earlier work, which employs a feedback mechanism and several reasoning-based prompt engineering techniques. We evaluate the proposed method on a story-to-reward generation task in a two-dimensional environment using two state-of-the-art LLMs, demonstrating the generalizability of our approach. Our experiments provide insightful evaluations that demonstrate the capabilities of LLMs essential for content generation tasks. The results highlight significant performance improvements of 415% and 40% respectively, depending on the zero-shot capabilities of the language model. Our work demonstrates the potential to reduce human dependency in game AI development, while supporting and enhancing creative processes.

Sangyeon Park, Isaac Han, Seungwon Oh et al.

Plasticity loss, a critical challenge in neural network training, limits a model's ability to adapt to new tasks or shifts in data distribution. This paper introduces AID (Activation by Interval-wise Dropout), a novel method inspired by Dropout, designed to address plasticity loss. Unlike Dropout, AID generates subnetworks by applying Dropout with different probabilities on each preactivation interval. Theoretical analysis reveals that AID regularizes the network, promoting behavior analogous to that of deep linear networks, which do not suffer from plasticity loss. We validate the effectiveness of AID in maintaining plasticity across various benchmarks, including continual learning tasks on standard image classification datasets such as CIFAR10, CIFAR100, and TinyImageNet. Furthermore, we show that AID enhances reinforcement learning performance in the Arcade Learning Environment benchmark.

In-Chang Baek, Sung-Hyun Kim, Seo-Young Lee et al.

Recent research has highlighted the significance of natural language in enhancing the controllability of generative models. While various efforts have been made to leverage natural language for content generation, research on deep reinforcement learning (DRL) agents utilizing text-based instructions for procedural content generation remains limited. In this paper, we propose IPCGRL, an instruction-based procedural content generation method via reinforcement learning, which incorporates a sentence embedding model. IPCGRL fine-tunes task-specific embedding representations to effectively compress game-level conditions. We evaluate IPCGRL in a two-dimensional level generation task and compare its performance with a general-purpose embedding method. The results indicate that IPCGRL achieves up to a 21.4% improvement in controllability and a 17.2% improvement in generalizability for unseen instructions. Furthermore, the proposed method extends the modality of conditional input, enabling a more flexible and expressive interaction framework for procedural content generation.

Seungwon Oh, Sangyeon Park, Isaac Han et al.

Recent studies have shown that as training progresses, neural networks gradually lose their capacity to learn new information, a phenomenon known as plasticity loss. An unbounded weight growth is one of the main causes of plasticity loss. Furthermore, it harms generalization capability and disrupts optimization dynamics. Re-initializing the network can be a solution, but it results in the loss of learned information, leading to performance drops. In this paper, we propose Soft Weight Rescaling (SWR), a novel approach that prevents unbounded weight growth without losing information. SWR recovers the plasticity of the network by simply scaling down the weight at each step of the learning process. We theoretically prove that SWR bounds weight magnitude and balances weight magnitude between layers. Our experiment shows that SWR improves performance on warm-start learning, continual learning, and single-task learning setups on standard image classification benchmarks.

In-Chang Baek, Tae-Hwa Park, Jin-Ha Noh et al.

Driven by the rapid growth of machine learning, recent advances in game artificial intelligence (AI) have significantly impacted productivity across various gaming genres. Reward design plays a pivotal role in training game AI models, wherein researchers implement concepts of specific reward functions. However, despite the presence of AI, the reward design process predominantly remains in the domain of human experts, as it is heavily reliant on their creativity and engineering skills. Therefore, this paper proposes ChatPCG, a large language model (LLM)-driven reward design framework.It leverages human-level insights, coupled with game expertise, to generate rewards tailored to specific game features automatically. Moreover, ChatPCG is integrated with deep reinforcement learning, demonstrating its potential for multiplayer game content generation tasks. The results suggest that the proposed LLM exhibits the capability to comprehend game mechanics and content generation tasks, enabling tailored content generation for a specified game. This study not only highlights the potential for improving accessibility in content generation but also aims to streamline the game AI development process.