Author name cluster

Junyoung Park

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.

8 papers

2 author rows

NeurIPS Conference 2025 Conference Paper

KeyDiff: Key Similarity-Based KV Cache Eviction for Long-Context LLM Inference in Resource-Constrained Environments

Junyoung Park
Dalton Jones
Matthew Morse
Raghavv Goel
Mingu Lee
Christopher Lott

We demonstrate that geometrically distinctive keys during LLM inference tend to have high attention scores. Based on the phenomenon we propose KeyDiff, a training-free KV cache eviction method based solely on key similarity. Unlike other KV cache eviction methods, KeyDiff can process arbitrarily long prompts within strict resource constraints and efficiently generate responses. We provide a theoretical basis for KeyDiff by relating key diversity with attention scores. These results imply KeyDiff can efficiently identify the most important tokens to retain. Notably KeyDiff does not rely on attention scores, allowing the use of optimized attention mechanisms like FlashAttention. Under a strict memory allowance, we demonstrate the effectiveness of KeyDiff for the Llama and Qwen model families by observing a performance gap of less than 0. 04\% with 8K cache budget (~23\% KV cache reduction) from the non-evicting baseline on LongBench for Llama 3. 1-8B and Llama 3. 2-3B. We also observe near baseline performance for Deepseek-R1-Distill-Llama-8B on the Math500 reasoning benchmark and decrease end-to-end inference latency by up to 30\% compared to the other token-eviction methods.

PDF Details

NeurIPS Conference 2025 Conference Paper

PARCO: Parallel AutoRegressive Models for Multi-Agent Combinatorial Optimization

Federico Berto
Chuanbo Hua
Laurin Luttmann
Jiwoo Son
Junyoung Park
Kyuree Ahn
Changhyun Kwon
Lin Xie

Combinatorial optimization problems involving multiple agents are notoriously challenging due to their NP-hard nature and the necessity for effective agent coordination. Despite advancements in learning-based methods, existing approaches often face critical limitations, including suboptimal agent coordination, poor generalization, and high computational latency. To address these issues, we propose PARCO (Parallel AutoRegressive Combinatorial Optimization), a general reinforcement learning framework designed to construct high-quality solutions for multi-agent combinatorial tasks efficiently. To this end, PARCO integrates three key novel components: (1) transformer-based communication layers to enable effective agent collaboration during parallel solution construction, (2) a multiple pointer mechanism for low-latency, parallel agent decision-making, and (3) priority-based conflict handlers to resolve decision conflicts via learned priorities. We evaluate PARCO in multi-agent vehicle routing and scheduling problems, where our approach outperforms state-of-the-art learning methods, demonstrating strong generalization ability and remarkable computational efficiency. We make our source code publicly available to foster future research: https: //github. com/ai4co/parco.

PDF Details

TMLR Journal 2025 Journal Article

RouteFinder: Towards Foundation Models for Vehicle Routing Problems

Federico Berto
Chuanbo Hua
Nayeli Gast Zepeda
André Hottung
Niels Wouda
Leon Lan
Junyoung Park
Kevin Tierney

This paper introduces RouteFinder, a comprehensive foundation model framework to tackle different Vehicle Routing Problem (VRP) variants. Our core idea is that a foundation model for VRPs should be able to represent variants by treating each as a subset of a generalized problem equipped with different attributes. We propose a unified VRP environment capable of efficiently handling any combination of these attributes. The RouteFinder model leverages a modern transformer-based encoder and global attribute embeddings to improve task representation. Additionally, we introduce two reinforcement learning techniques to enhance multi-task performance: mixed batch training, which enables training on different variants at once, and multi-variant reward normalization to balance different reward scales. Finally, we propose efficient adapter layers that enable fine-tuning for new variants with unseen attributes. Extensive experiments on 48 VRP variants show RouteFinder outperforms recent state-of-the-art learning methods. Our code is publicly available at https://github.com/ai4co/routefinder.

PDF Details

ICML Conference 2023 Conference Paper

Kernel Sufficient Dimension Reduction and Variable Selection for Compositional Data via Amalgamation

Junyoung Park
Jeongyoun Ahn
Cheolwoo Park

Compositional data with a large number of components and an abundance of zeros are frequently observed in many fields recently. Analyzing such sparse high-dimensional compositional data naturally calls for dimension reduction or, more preferably, variable selection. Most existing approaches lack interpretability or cannot handle zeros properly, as they rely on a log-ratio transformation. We approach this problem with sufficient dimension reduction (SDR), one of the most studied dimension reduction frameworks in statistics. Characterized by the conditional independence of the data to the response on the found subspace, the SDR framework has been effective for both linear and nonlinear dimension reduction problems. This work proposes a compositional SDR that can handle zeros naturally while incorporating the nonlinear nature and spurious negative correlations among components rigorously. A critical consideration of sub-composition versus amalgamation for compositional variable selection is discussed. The proposed compositional SDR is shown to be statistically consistent in constructing a sub-simplex consisting of true signal variables. Simulation and real microbiome data are used to demonstrate the performance of the proposed SDR compared to existing state-of-art approaches.

Details

AAMAS Conference 2023 Conference Paper

Learn to Solve the Min-max Multiple Traveling Salesmen Problem with Reinforcement Learning

Junyoung Park
Changhyun Kwon
Jinkyoo Park

We propose ScheduleNet, a scalable scheduler that minimizes task completion time by coordinating multiple agents. We formulate the min-max Multiple Traveling Salesmen Problem (mTSP) as a Markov decision process with an episodic reward and derive a scalable decision-making policy using Reinforcement Learning (RL). The decision-making procedure of ScheduleNet includes (1) representing the state of a problem with the agent-task graph, (2) extracting node embedding for agents and tasks by employing the type-aware graph attention, (3) and computing the task assignment probability with the computed node embedding. We show that ScheduleNet can outperform other heuristic approaches and existing deep RL approaches, particularly validating its exceptional effectiveness in solving large and practical problems. We also confirm that ScheduleNet can effectively solve practical mTSP variants, which include limited observation and online mTSP.

PDF

ICML Conference 2022 Conference Paper

Kernel Methods for Radial Transformed Compositional Data with Many Zeros

Junyoung Park
Changwon Yoon
Cheolwoo Park
Jeongyoun Ahn

Compositional data analysis with a high proportion of zeros has gained increasing popularity, especially in chemometrics and human gut microbiomes research. Statistical analyses of this type of data are typically carried out via a log-ratio transformation after replacing zeros with small positive values. We should note, however, that this procedure is geometrically improper, as it causes anomalous distortions through the transformation. We propose a radial transformation that does not require zero substitutions and more importantly results in essential equivalence between domains before and after the transformation. We show that a rich class of kernels on hyperspheres can successfully define a kernel embedding for compositional data based on this equivalence. To the best of our knowledge, this is the first work that theoretically establishes the availability of the extensive library of kernel-based machine learning methods for compositional data. The applicability of the proposed approach is demonstrated with kernel principal component analysis.

Details

NeurIPS Conference 2022 Conference Paper

Sym-NCO: Leveraging Symmetricity for Neural Combinatorial Optimization

Minsu Kim
Junyoung Park
Jinkyoo Park

Deep reinforcement learning (DRL)-based combinatorial optimization (CO) methods (i. e. , DRL-NCO) have shown significant merit over the conventional CO solvers as DRL-NCO is capable of learning CO solvers less relying on problem-specific expert domain knowledge (heuristic method) and supervised labeled data (supervised learning method). This paper presents a novel training scheme, Sym-NCO, which is a regularizer-based training scheme that leverages universal symmetricities in various CO problems and solutions. Leveraging symmetricities such as rotational and reflectional invariance can greatly improve the generalization capability of DRL-NCO because it allows the learned solver to exploit the commonly shared symmetricities in the same CO problem class. Our experimental results verify that our Sym-NCO greatly improves the performance of DRL-NCO methods in four CO tasks, including the traveling salesman problem (TSP), capacitated vehicle routing problem (CVRP), prize collecting TSP (PCTSP), and orienteering problem (OP), without utilizing problem-specific expert domain knowledge. Remarkably, Sym-NCO outperformed not only the existing DRL-NCO methods but also a competitive conventional solver, the iterative local search (ILS), in PCTSP at 240$\times$ faster speed. Our source code is available at https: //github. com/alstn12088/Sym-NCO.

PDF Details

ICRA Conference 2018 Conference Paper

Stiffness Decomposition and Design Optimization of Under-Actuated Tendon-Driven Robotic Systems

Minji Kim
Junyoung Park
Juhyeok Kim
Myungsin Kim

We present a novel systematic design framework for general under-actuated tendon-driven (UATD) robotic systems to exhibit desired behaviors both during the free motion and the contact task. For this, we propose stiffness decomposition, which enables us to completely decompose the configuration space of the UATD robotic systems into the actuated space (with full actuation via active tendons) and the un-actuated space (with no actuation, only with passive compliance and contact wrench). The behavior in the actuated space is then fully-controllable, thus, the attainment of the desired behaviors, particularly those during the contact task, hinges upon that in the un-actuated space. For this, relying on the stiffness decomposition, we optimize the design parameters (e. g. , tendon routing, pulley radius, passive compliance, etc.) to ensure the deformation in the un-actuated space as directional (e. g. , for adaptive grasping) and minimized (e. g. , pushing with posture maintained) for different contact wrench sets as possible, while also rendering the free motion to be as compliant and backdrivable as possible. The presented framework is then applied to design a UATD robotic finger and experimentally verified with the robot able to mimic the behavior of human index finger both during the free motion and pinch-pushing.

Details