arXiv subjects

Seungjoon Lee

Publications and source records attributed to Seungjoon Lee.

15 recordsLinked to original sources

arXiv2026

Towards Faithful Agentic XAI: A Verification Method and an Open-World Benchmark for Better Model Faithfulness

Explainable AI (XAI) helps users interpret model behavior and identify potential faults. Agentic XAI systems use Large Language Models (LLMs) to make explanations more accessible through natural-language interaction, but they can also produce plausible yet unfaithful explanations. This risk arises because unreliable XAI outputs for complex models can be amplified by LLMs and mislead users. We propose Faithful Agentic XAI (FAX), a framework that improves explanation faithfulness through explicit verification. FAX decomposes draft explanations into claims and cross-checks them against inherently faithful tools, filtering unsupported or contradictory claims before final generation. We also introduce CRAFTER-XAI-Bench, an open-world reinforcement learning benchmark with complex policies, diverse goals, and challenging scenarios for assessing model-specific faithfulness. On CRAFTER-XAI-Bench, FAX improves simulation faithfulness from 0.20 for the strongest baseline to 0.46 while maintaining high informativeness, relevance, and fluency. On three tabular benchmarks, FAX performs competitively with prior Agentic XAI baselines, but our analysis shows that these settings can conflate task accuracy with model-specific faithfulness. These findings show that explicit verification is essential for faithful Agentic XAI and that that faithfulness benchmarks must be designed to test explanations against the behavior of the target model itself.

cs.AI↗

arXiv2026

ProCompNav: Proactive Instance Navigation with Comparative Judgment for Ambiguous User Queries

Natural-language instance navigation becomes challenging when the initial user request does not uniquely specify the target instance. A practical agent should reduce the user's burden by actively asking only the information needed to distinguish the target from similar distractors, rather than requiring a detailed description upfront. Existing approaches often fall short of this goal: they may stop at the first plausible candidate before sufficiently exploring alternatives, or, even after collecting multiple candidates, ask about the target's attributes derived from individual candidates rather than questions selected to distinguish candidates in the pool. As a result, despite the dialogue, the agent may still fail to distinguish the target from distractors, leading to premature decisions and lengthy user responses. We propose Proactive Instance Navigation with Comparative Judgment (ProCompNav), a two-stage framework that first constructs a candidate pool and then identifies the target through comparative judgment. At each round, ProCompNav extracts an attribute-value pair that splits the current pool, asks a binary yes/no question, and prunes all inconsistent candidates at once. This reframes disambiguation from open-ended target description to pool-level discriminative questioning, where each question is chosen to narrow the candidate set. On CoIN-Bench, ProCompNav improves Success Rate over interactive baselines with the same minimal input and non-interactive baselines with detailed descriptions, while substantially reducing Response Length. ProCompNav also achieves state-of-the-art Success Rate on TextNav, suggesting that comparative judgment is broadly useful for instance-level navigation among similar distractors. Code is available at https://github.com/tree-jhk/procompnav.

cs.AI↗

arXiv2026

Experience-based Knowledge Correction for Robust Planning in Minecraft

Large Language Model (LLM)-based planning has advanced embodied agents in long-horizon environments such as Minecraft, where acquiring latent knowledge of goal (or item) dependencies and feasible actions is critical. However, LLMs often begin with flawed priors and fail to correct them through prompting, even with feedback. We present XENON (eXpErience-based kNOwledge correctioN), an agent that algorithmically revises knowledge from experience, enabling robustness to flawed priors and sparse binary feedback. XENON integrates two mechanisms: Adaptive Dependency Graph, which corrects item dependencies using past successes, and Failure-aware Action Memory, which corrects action knowledge using past failures. Together, these components allow XENON to acquire complex dependencies despite limited guidance. Experiments across multiple Minecraft benchmarks show that XENON outperforms prior agents in both knowledge learning and long-horizon planning. Remarkably, with only a 7B open-weight LLM, XENON surpasses agents that rely on much larger proprietary models. Project page: https://sjlee-me.github.io/XENON

cs.LG↗

arXiv2025

Using Linearized Optimal Transport to Predict the Evolution of Stochastic Particle Systems

We develop an Euler-type method to predict the evolution of a time-dependent probability measure without explicitly learning an operator that governs its evolution. We use linearized optimal transport theory to prove that the measure-valued analog of Euler's method is first-order accurate when the measure evolves ``smoothly.'' In applications of interest, however, the measure is an empirical distribution of a system of stochastic particles whose behavior is only accessible through an agent-based micro-scale simulation. In such cases, this empirical measure does not evolve smoothly because the individual particles move chaotically on short time scales. However, we can still perform our Euler-type method, and when the particles' collective distribution approximates a measure that \emph{does} evolve smoothly, we observe that the algorithm still accurately predicts this collective behavior over relatively large Euler steps, thus reducing the number of micro-scale steps required to step forward in time. In this way, our algorithm provides a ``macro-scale timestepper'' that requires less micro-scale data to still maintain accuracy, which we demonstrate with three illustrative examples: a biological agent-based model, a model of a PDE, and a model of Langevin dynamics.

math.NA↗

arXiv2025

Comparison-based Active Preference Learning for Multi-dimensional Personalization

Large language models (LLMs) have shown remarkable success, but aligning them with human preferences remains a core challenge. As individuals have their own, multi-dimensional preferences, recent studies have explored multi-dimensional personalization, which aims to enable models to generate responses personalized to explicit preferences. However, human preferences are often implicit and thus difficult to articulate, limiting the direct application of this approach. To bridge this gap, we propose Active Multi-dimensional Preference Learning (AMPLe), designed to capture implicit user preferences from interactively collected comparative feedback. Building on Bayesian inference, our work introduces a modified posterior update procedure to mitigate estimation bias and potential noise in comparisons. Also, inspired by generalized binary search, we employ an active query selection strategy to minimize the number of required comparisons by a user. Through theoretical analysis and experiments on language generation tasks, we demonstrate feedback efficiency and effectiveness of our framework in personalizing model responses. Our code is publicly available at https://github.com/ml-postech/AMPLe .

cs.LG↗

arXiv2024

Multi-Agent Path Finding with Real Robot Dynamics and Interdependent Tasks for Automated Warehouses

Multi-Agent Path Finding (MAPF) is an important optimization problem underlying the deployment of robots in automated warehouses and factories. Despite the large body of work on this topic, most approaches make heavy simplifications, both on the environment and the agents, which make the resulting algorithms impractical for real-life scenarios. In this paper, we consider a realistic problem of online order delivery in a warehouse, where a fleet of robots bring the products belonging to each order from shelves to workstations. This creates a stream of inter-dependent pickup and delivery tasks and the associated MAPF problem consists of computing realistic collision-free robot trajectories fulfilling these tasks. To solve this MAPF problem, we propose an extension of the standard Prioritized Planning algorithm to deal with the inter-dependent tasks (Interleaved Prioritized Planning) and a novel Via-Point Star (VP*) algorithm to compute an optimal dynamics-compliant robot trajectory to visit a sequence of goal locations while avoiding moving obstacles. We prove the completeness of our approach and evaluate it in simulation as well as in a real warehouse.

cs.RO↗

arXiv2024

Fully Decentralized Design of Initialization-free Distributed Network Size Estimation

In this paper, we propose a distributed scheme for estimating the network size, which refers to the total number of agents in a network. By leveraging a synchronization technique for multi-agent systems, we devise an agent dynamics that ensures convergence to an equilibrium point located near the network size regardless of its initial condition. Our approach is based on an assumption that each agent has a unique identifier, and an estimation algorithm for obtaining the largest identifier value. By adopting this approach, we successfully implement the agent dynamics in a fully decentralized manner, ensuring accurate network size estimation even when some agents join or leave the network.

eess.SY↗

arXiv2022

Data-driven Discovery of Chemotactic Migration of Bacteria via Machine Learning

E. coli chemotactic motion in the presence of a chemoattractant field has been extensively studied using wet laboratory experiments, stochastic computational models as well as partial differential equation-based models (PDEs). The most challenging step in bridging these approaches, is establishing a closed form of the so-called chemotactic term, which describes how bacteria bias their motion up chemonutrient concentration gradients, as a result of a cascade of biochemical processes. Data-driven models can be used to learn the entire evolution operator of the chemotactic PDEs (black box models), or, in a more targeted fashion, to learn just the chemotactic term (gray box models). In this work, data-driven Machine Learning approaches for learning the underlying model PDEs are (a) validated through the use of simulation data from established continuum models and (b) used to infer chemotactic PDEs from experimental data. Even when the data at hand are sparse (coarse in space and/or time), noisy (due to inherent stochasticity in measurements) or partial (e.g. lack of measurements of the associated chemoattractant field), we can attempt to learn the right-hand-side of a closed PDE for an evolving bacterial density. In fact we show that data-driven PDEs including a short history of the bacterial density field (e.g. in the form of higher-order in time PDEs in terms of the measurable bacterial density) can be successful in predicting further bacterial density evolution, and even possibly recovering estimates of the unmeasured chemonutrient field. The main tool in this effort is the effective low-dimensionality of the dynamics (in the spirit of the Whitney and Takens embedding theorems). The resulting data-driven PDE can then be simulated to reproduce/predict computational or experimental bacterial density profile data, and estimate the underlying (unmeasured) chemonutrient field evolution.

q-bio.QM↗

arXiv2022

Learning black- and gray-box chemotactic PDEs/closures from agent based Monte Carlo simulation data

We propose a machine learning framework for the data-driven discovery of macroscopic chemotactic Partial Differential Equations (PDEs) -- and the closures that lead to them -- from high-fidelity, individual-based stochastic simulations of E.coli bacterial motility. The fine scale, detailed, hybrid (continuum - Monte Carlo) simulation model embodies the underlying biophysics, and its parameters are informed from experimental observations of individual cells. We exploit Automatic Relevance Determination (ARD) within a Gaussian Process framework for the identification of a parsimonious set of collective observables that parametrize the law of the effective PDEs. Using these observables, in a second step we learn effective, coarse-grained "Keller-Segel class" chemotactic PDEs using machine learning regressors: (a) (shallow) feedforward neural networks and (b) Gaussian Processes. The learned laws can be black-box (when no prior knowledge about the PDE law structure is assumed) or gray-box when parts of the equation (e.g. the pure diffusion part) is known and "hardwired" in the regression process. We also discuss data-driven corrections (both additive and functional) of analytically known, approximate closures.

q-bio.QM↗

arXiv2021

Multi-armed Bandit Algorithm against Strategic Replication

We consider a multi-armed bandit problem in which a set of arms is registered by each agent, and the agent receives reward when its arm is selected. An agent might strategically submit more arms with replications, which can bring more reward by abusing the bandit algorithm's exploration-exploitation balance. Our analysis reveals that a standard algorithm indeed fails at preventing replication and suffers from linear regret in time $T$. We aim to design a bandit algorithm which demotivates replications and also achieves a small cumulative regret. We devise Hierarchical UCB (H-UCB) of replication-proof, which has $O(\ln T)$-regret under any equilibrium. We further propose Robust Hierarchical UCB (RH-UCB) which has a sublinear regret even in a realistic scenario with irrational agents replicating careless. We verify our theoretical findings through numerical experiments.

cs.LG↗

arXiv2021

Blended Dynamics Approach to Distributed Optimization: Sum Convexity and Convergence Rate

This paper studies the application of the blended dynamics approach towards distributed optimization problem where the global cost function is given by a sum of local cost functions. The benefits include (i) individual cost function need not be convex as long as the global cost function is strongly convex and (ii) the convergence rate of the distributed algorithm is arbitrarily close to the convergence rate of the centralized one. Two particular continuous-time algorithms are presented using the proportional-integral-type couplings. One has benefit of `initialization-free,' so that agents can join or leave the network during the operation. The other one has the minimal amount of communication information. After presenting a general theorem that can be used for designing distributed algorithms, we particularly present a distributed heavy-ball method and discuss its strength over other methods.

math.OC↗

arXiv2021

Coarse-scale PDEs from fine-scale observations via machine learning

Complex spatiotemporal dynamics of physicochemical processes are often modeled at a microscopic level (through e.g. atomistic, agent-based or lattice models) based on first principles. Some of these processes can also be successfully modeled at the macroscopic level using e.g. partial differential equations (PDEs) describing the evolution of the right few macroscopic observables (e.g. concentration and momentum fields). Deriving good macroscopic descriptions (the so-called "closure problem") is often a time-consuming process requiring deep understanding/intuition about the system of interest. Recent developments in data science provide alternative ways to effectively extract/learn accurate macroscopic descriptions approximating the underlying microscopic observations. In this paper, we introduce a data-driven framework for the identification of unavailable coarse-scale PDEs from microscopic observations via machine learning algorithms. Specifically, using Gaussian Processes, Artificial Neural Networks, and/or Diffusion Maps, the proposed framework uncovers the relation between the relevant macroscopic space fields and their time evolution (the right-hand-side of the explicitly unavailable macroscopic PDE). Interestingly, several choices equally representative of the data can be discovered. The framework will be illustrated through the data-driven discovery of macroscopic, concentration-level PDEs resulting from a fine-scale, Lattice Boltzmann level model of a reaction/transport process. Once the coarse evolution law is identified, it can be simulated to produce long-term macroscopic predictions. Different features (pros as well as cons) of alternative machine learning algorithms for performing this task (Gaussian Processes and Artificial Neural Networks), are presented and discussed.

cs.LG↗

arXiv2019

Some manifold learning considerations towards explicit model predictive control

Model predictive control (MPC) is a de facto standard control algorithm across the process industries. There remain, however, applications where MPC is impractical because an optimization problem is solved at each time step. We present a link between explicit MPC formulations and manifold learning to enable facilitated prediction of the MPC policy. Our method uses a similarity measure informed by control policies and system state variables, to "learn" an intrinsic parametrization of the MPC controller using a diffusion maps algorithm, which will also discover a low-dimensional control law when it exists as a smooth, nonlinear combination of the state variables. We use function approximation algorithms to project points from state space to the intrinsic space, and from the intrinsic space to policy space. The approach is illustrated first by "learning" the intrinsic variables for MPC control of constrained linear systems, and then by designing controllers for an unstable nonlinear reactor.

math.OC↗

arXiv2019

Distributed Algorithm for Economic Dispatch Problem with Separable Losses

Economic dispatch problem for a networked power system has been considered. The objective is to minimize the total generation cost while meeting the overall supply-demand balance and generation capacity. In particular, a more practical scenario has been studied by considering the power losses. A non-convex optimization problem has been formulated where the non-convexity comes from the nonlinear equality constraint representing the supply-demand balance with the power losses. It is shown that the optimization problem can be solved using convex relaxation and dual decomposition. A simple distributed algorithm is proposed to solve the optimization problem. Specifically, the proposed algorithm does not require any initialization process and hence robust to various changes in operating condition. In addition, the behavior of the proposed algorithm is analyzed when the problem is infeasible.

eess.SY↗

arXiv2018

Linking Gaussian Process regression with data-driven manifold embeddings for nonlinear data fusion

In statistical modeling with Gaussian Process regression, it has been shown that combining (few) high-fidelity data with (many) low-fidelity data can enhance prediction accuracy, compared to prediction based on the few high-fidelity data only. Such information fusion techniques for multifidelity data commonly approach the high-fidelity model $f_h(t)$ as a function of two variables $(t,y)$, and then using $f_l(t)$ as the $y$ data. More generally, the high-fidelity model can be written as a function of several variables $(t,y_1,y_2....)$; the low-fidelity model $f_l$ and, say, some of its derivatives, can then be substituted for these variables. In this paper, we will explore mathematical algorithms for multifidelity information fusion that use such an approach towards improving the representation of the high-fidelity function with only a few training data points. Given that $f_h$ may not be a simple function -- and sometimes not even a function -- of $f_l$, we demonstrate that using additional functions of $t$, such as derivatives or shifts of $f_l$, can drastically improve the approximation of $f_h$ through Gaussian Processes. We also point out a connection with "embedology" techniques from topology and dynamical systems.

stat.ML↗