Searcharxiv⌕ Search

arXiv subjects

Hongseok Namkoong

Publications and source records attributed to Hongseok Namkoong.

At least 19 recordsLinked to original sources

Rare Event Estimation via Iterative Unalignment

As agents are deployed with increased autonomy, even extremely rare events along their stochastic output trajectories can occur and prove catastrophic. Safe deployment therefore does not depend on whether these events can occur, but on how often they might. We study the problem of estimating the probability of rare events that arise from stochastic variation in the agent's own actions. Estimating this type of risk requires searching over the combinatorially vast space of trajectories. Naive Monte Carlo is computationally prohibitive in this regime, and constructing effective importance sampling (IS) proposals requires coordinated changes to a context-dependent chain of conditional distributions. We develop a new IS method that perturbs the original model's weights to construct the proposal. The proposal is itself a differentiably parameterized language model, enabling gradient-based search over weight space. We formulate an objective that combines a differentiable surrogate for event amplification and an adaptive regularization scheme that dynamically balances amplification against estimator stability. We evaluate our approach on $\sim$120M and $\sim$2.6B models across three event families spanning 300+ rare events as rare as $10^{-9}$, all with independently estimated reference probabilities. In our most verifiable settings, we observe that our IS estimator achieves over $800\times$ compute-weighted efficiency gains on average over naive Monte Carlo for events with probabilities lower than $10^{-7}$. Our implementation is available at https://github.com/namkoong-lab/iterative-unalignment.

cs.LG↗

A Sensitivity Framework for Assessing External Validity

Assessing the external validity of experimental results is a pervasive challenge in causal inference as internally valid findings need not generalize beyond sites in which the experiments were conducted. Since the number of sites is often smaller than the number of site-level variables that moderate causal effects, we develop a partial identification framework that lets practitioners assess external validity under a wide range of assumptions that allow extrapolation across site-level variables. Our optimization-driven approach can accommodate any affine family of parameter restrictions, such as interaction terms being dominated by direct effects (the "effect hierarchy principle"). Since standard estimating-equation arguments do not apply in our setting---underspecified parameter directions responsible for partial identification make the covariance matrix singular---we develop a novel debiasing approach for estimands given by constrained optimization problems, giving rise to semiparametrically efficient estimators for our flexible partial identification bounds. We illustrate our approach on multi-site microcredit experiments harmonized by Meager (2019), whose lending environments differ widely in interest rates, loan sizes, and promotion. Holding out one country at a time, we construct external-validity intervals for the held-out country from the remaining source sites and show how our framework connects different extrapolation assumptions to concrete partial identification bounds.

stat.ME↗

LatentGym: A Testbed For Cross-Task Experiential Learning With Controllable Latent Structure

We envision continually learning agentic systems that become more useful over time: as they encounter sequences of related tasks, they should infer the hidden structure shared across those tasks and use it to improve future decisions. This cross-task experiential learning capability is pivotal in domains such as personalization and interactive assistance, but existing training/evaluation frameworks do not provide shared, controllable latent structures and cannot measure whether or why agents improve. We introduce LatentGym: a controllable suite in which each environment is organized around a ground-truth latent variable governing the structure across tasks. Our construction yields metrics that separate exploration (whether the agent's actions gather information about the latent) from exploitation (whether the agent uses what it has gathered). We demonstrate our suite on empirical studies addressing three questions: how and why frontier models fail to adapt across related tasks; whether post-training on related task sequences improves general cross-task adaptation, and where those gains come from; and how design choices such as inter-task feedback shape training dynamics and generalization. Together, these results establish a controlled foundation for studying how LLM agents learn from experience across tasks, and for designing agents that adapt more reliably in sequential, personalized, and interactive settings.

cs.LG↗

DRO: A Python Library for Distributionally Robust Optimization in Machine Learning

We introduce dro, an open-source Python library for distributionally robust optimization (DRO) for regression and classification problems. The library implements 15 DRO formulations and 9 backbone models, enabling 85 distinct DRO methods. Furthermore, dro is compatible with both scikit-learn and PyTorch through a unified interface, and uses approximation techniques to reduce runtime for some computationally demanding formulations. Comprehensive documentation is available at https://python-dro.org.

cs.LG↗

Seq@K: Evaluating Language Models Under Sequential Feedback

Language-model agents differ fundamentally from prior intelligent systems in their ability to learn from natural language in context. This capability enables agents to iteratively refine their answers in response to feedback, a prevalent mode of operation in modern agentic workflows. However, the community primarily ranks models by metrics such as Pass@1 or Pass@K, which cannot capture how they utilize feedback on prior attempts. We introduce Seq@K, a natural evaluation protocol that measures whether a model can solve a problem within K sequential attempts, provided feedback after each previous failure. Unlike multi-turn agentic benchmarks that examine long-horizon planning, Seq@K isolates and formalizes the core capability of refinement from feedback. We evaluate Seq@K on a suite of 7 benchmarks and 16 frontier models, and show that Seq@K reveals a materially different model-selection landscape: under Seq@K, (1) model rankings often differ, (2) the Pareto frontiers admit more efficient configurations and (3) the gains from choosing the optimal model configuration can widen or shrink relative to those of Pass@K. We further show that analyzing model trajectories under Seq@K helps characterize iterative refinement behavior, such as the rate at which models address issues raised in feedback, or how rigidly they hold onto their own answers. Finally, we study how the nature of feedback affects models' iterative refinement, showing that practitioners must carefully design and benchmark the feedback loops of iterative agentic systems. Together, these results establish how Seq@K offers a more faithful metric for intelligent systems that typically operate in such feedback loop, and the necessity to consider Seq@K in evaluating models for these systems.

cs.AI↗

MBABench-2.0: Evaluating AI Agents on Constructing Financial Spreadsheets Beyond Simple Verifiability

Spreadsheets are the lingua franca of modern finance, driving valuation, forecast, audit, and many other core workflows of the industry. As LLM agents advance to automate knowledge work, spreadsheet construction is a natural and pertinent testbed. However, most existing spreadsheet benchmarks evaluate agents on fill-in-the-blank style tasks, where an agent performs minor edits in otherwise complete spreadsheets. This protocol fails to capture the most central and time-consuming part of the spreadsheet workflow: the modeling decisions, format design, and sheet layout regular knowledge workers need to toil through to build a workbook from scratch. We introduce MBABench-2.0, a benchmark of 101 original spreadsheet construction tasks, the hardest of which is estimated to take 33 hours for a modeler with 3+ years of experience. Unlike coding or math, where a unit test or a final answer can verify an agent's output, the quality of a financial spreadsheet that concerns readability, maintainability, or auditability, is not so simply verifiable. To holistically evaluate spreadsheet construction, we introduce 129 criteria spanning 12 categories that operationalize quality beyond the accuracy of the spreadsheet. Evaluating over 20 agents across 8 models and 4 harness types, we find that frontier agents rarely construct spreadsheets that are both accurate and meet our quality criteria: the top agent achieves a pass rate of only 16% on the benchmark. The benchmark demonstrates a substantial gap in spreadsheet building capability of current agents.

cs.SE↗

GameBoyWorlds: A Testbed for Self-Improvement in Embodied Video Games

Powered by expert guidance, agents can operate in interactive environments; however, it is unclear whether they can learn autonomously from their own experience. To evaluate such self-improvement methods, we introduce GameBoyWorlds, a testbed for agentic self-improvement in video games. GameBoyWorlds-Execution evaluates task execution on a collection of 5 distinct game series. Agents are allowed access to dedicated training games but are provided no demonstrations, documentation, or rewards. Agents must ground themselves in the environment through self-directed exploration and by inferring actionable knowledge from their own experience. At test time, agents must complete short-horizon tasks that evaluate their ability to navigate, interact, and engage with game-specific mechanics in unseen games. Out-of-the-box frontier models complete fewer than 50% of the 500 tasks due to failures in multimodal grounding, establishing that self-improvement methods have room to push performance. We demonstrate that contemporary approaches to self-improvement are lacking, with world modelling and autonomous skill discovery failing, and a novel strategy that uses curiosity-based exploration to write guides achieving only partial success. GameBoyWorlds-Playthrough tests end-to-end game completion in two fan-made Pokémon games. We show that while frontier models have been pre-exposed to official releases such as Pokémon Red, they lack essential information on the games in our testbed. Instead of relying on their parametric knowledge to succeed, agents must learn from their own experience and autonomously improve over the course of the playthrough. We show that a sophisticated agentic pipeline with multimodal memory and hierarchical subgoals fails to reach even the first major milestone in both games, establishing GameBoyWorlds as an ambitious target for self-improving agents.

cs.AI↗

C-Learner: Constrained Learning for Causal Inference

Debiasing methods such as augmented inverse propensity weighting (AIPW), and targeted maximum likelihood estimation (TMLE) enjoy asymptotic properties like semiparametric efficiency and double robustness, but can produce unstable estimates in practice that require ad hoc adjustments (e.g., truncating propensity scores). In contrast, simple plug-ins can remain stable but lack these asymptotic guarantees. To achieve the best of both worlds---a plug-in that enjoys strong asymptotic guarantees---we propose a constrained learning framework that trains a nuisance model to minimize prediction error subject to the constraint that the estimated first-order error of the resulting plug-in is zero. To compare different debiasing methods that share the same classical limit, we study a stylized high-dimensional regression problem where nuisance estimation errors do not vanish asymptotically. Our unified analysis covers both $d n$, as well as ridge regularization, and characterizes how overlap affects the estimators' limiting distributions. Under sufficient overlap, our estimator has smaller asymptotic variance than AIPW and TMLE, whereas when overlap deteriorates so much that AIPW and TMLE are no longer root-$n$ consistent, constrained learning still retains the direct plug-in's root-$n$ limit. Empirically, across a range of experimental settings including those with text-based covariates and language models, we observe our estimator outperforms classical debiasing methods in challenging settings with limited overlap between treatment and control, and performs similarly otherwise.

stat.ML↗

MBABench: Evaluating LLM Agents on End-to-End Spreadsheet Tasks in Finance

LLM agents are increasingly expected to carry out end-to-end workflows, producing complete artifacts from high-level user instructions. To meet enterprise needs, frontier AI labs have developed agents that can construct entire spreadsheets from scratch. This is especially relevant in finance, where core workflows such as financial modeling, forecasting, and scenario analysis are commonly conducted through spreadsheets. Yet, existing spreadsheet benchmarks do not measure this new capability, focusing instead on question-answering or single-formula edits. To address this gap, we provide one of the first evaluations of agents on end-to-end spreadsheet tasks, focusing on economically critical financial workflows such as modeling and scenario analysis. Since deliverables therein are routinely reviewed and revised by multiple stakeholders, judging their quality necessarily involves high-level criteria such as readability or ease of modification. To reflect the multidimensional nature of solution quality, we develop an evaluation taxonomy comprising three dimensions: Accuracy, Formula, and Format, each comprising fine-grained criteria that reflect professional standards. Evaluating over 18 agents, the benchmark reveals that even the strongest agents fall short of basic professional finance standards, and their performance degrade sharply as the difficulty increases beyond a few chained calculations. This suggests that current agents are not yet able to reliably produce professional-quality spreadsheets at the level of complexity real-world workflows demand.

cs.AI↗

Benchmarking In-context Experiential Learning Through Repeated Product Recommendations

To navigate ever-shifting real-world environments, agents must grapple with incomplete knowledge and adapt their strategies through experience. However, current evaluations of LLM-based agents largely overlook this capability. Crucially, we stress not just the ability to contend with uncertainty within a task (episode), as episode-specific information is progressively revealed across turns, but also the refinement of such adaptive ability across similar episodes, as agents accumulate experiences and infer their shared latent structure. Repeated product recommendation offers a natural setting to isolate this need for experiential learning: within each interaction, an intelligent recommender must elicit unknown customer preferences through questions; across multiple interactions, it should tailor its questioning strategy to the observed customer and product distributions. We instantiate the Benchmark for Experiential Learning and Active exploration (BELA) by combining (1) a rich catalog of real-world products from Amazon, (2) a diverse collection of synthetic customer personas aimed to capture heterogeneous latent preferences, and (3) an LLM-based customer simulator framework that emulate preference-revealing interactions. Rather than aiming to faithfully replicate real consumer behavior, BELA provides a controlled and scalable testbed for whether agents can exploit consistent latent preferences across episodes. Benchmarking current models reveals that they can learn across turns, but struggle to improve across episodes. This underscore the need for frontier models to advance in experiential learning capabilities.

cs.LG↗

Active Exploration via Autoregressive Generation of Missing Data

We pose uncertainty quantification and exploration in online decision-making as a problem of training and generation from an autoregressive sequence model, an area experiencing rapid innovation. Our approach rests on viewing uncertainty as arising from missing future outcomes that could be revealed through action choices, rather than from unobservable latent parameters of the environment. This reformulation aligns naturally with modern machine learning capabilities: we can i) train generative models through next-outcome prediction rather than fit explicit priors, ii) assess uncertainty through autoregressive generation rather than sampling latent parameters from posteriors, and iii) adapt to new information by extending the sequence model's context rather than explicit posterior updating. Our main theoretical result establishes a reduction from online decision-making to offline next-outcome prediction: Bayesian regret is controlled directly by the sequence model's offline prediction loss, without requiring an explicit latent-variable posterior. Experiments, including a semi-synthetic news recommendation task, show that autoregressive generation produces calibrated epistemic uncertainty and enables effective exploration by using article text as prior information to focus exploration on resolving remaining uncertainties.

cs.LG↗

Design and Scheduling of an AI-based Queueing System

To leverage prediction models to make optimal scheduling decisions in service systems, we must understand how predictive errors impact congestion due to externalities on the delay of other jobs. Motivated by applications where prediction models interact with human servers (e.g., content moderation), we consider a large queueing system comprising of many single server queues where the class of a job is estimated using a prediction model. By characterizing the impact of mispredictions on congestion cost in heavy traffic, we design an index-based policy that incorporates the predicted class information in a near-optimal manner. Our theoretical results guide the design of predictive models by providing a simple model selection procedure with downstream queueing performance as a central concern, and offer novel insights on how to design queueing systems with AI-based triage. We illustrate our framework on a content moderation task based on real online comments, where we construct toxicity classifiers by finetuning large language models.

math.OC↗

When Representative Samples Produce Worse Outcomes: Scale-up Decisions and Testing in Small-Budget RCTs

Small randomized controlled trials are often used to screen interventions before running larger follow-up studies. This is a critical phase of experimentation, as missing effective interventions or scaling up harmful ones can be very costly. A common proposal to mitigate these errors is to recruit samples that are representative of the target population, but this is often challenging in resource-constrained pilots. We challenge the narrative that representative samples are always superior by showing that when statistical significance testing determines whether interventions receive further study, the pilot trial composition that maximizes the downstream expected improvement in outcomes depends critically on its budget size. In the large-budget limit, the optimal pilot design converges to a sample that is representative of the target population. However, in the small-budget regime, the pilot designer maximizes expected impact by sampling only from a single homogeneous sub-population, chosen in a manner that depends on sampling costs and the designer's prior beliefs about heterogeneous treatment effects. Our proof of the small-budget result applies more generally when an RCT and significance test are used to decide whether to receive any non-adaptive downstream payoff, a result that may be applicable to other settings with constrained experimentation budgets.

stat.ME↗

Rethinking Distribution Shifts: Empirical Analysis and Modeling for Tabular Data

Different distribution shifts require different interventions, and algorithms must be grounded in the specific shifts they address. However, methodological development for robust algorithms typically relies on structural assumptions that lack empirical validation. Advocating for an empirically grounded data-driven approach to algorithm development, we build an empirical testbed comprising natural shifts across 8 tabular datasets, 172 distribution pairs over 45 methods and 90,000 method configurations encompassing empirical risk minimization and distributionally robust optimization (DRO) methods. We find $Y|X$-shifts are most prevalent in our testbed, in stark contrast to the heavy focus on $X$ (covariate)-shifts in the ML literature, and that the performance of robust algorithms is no better than that of vanilla methods. To understand why, we conduct an in-depth empirical analysis of DRO methods and find that underlooked implementation details -- such as the choice of underlying model class (e.g., LightGBM) and hyperparameter selection -- have a bigger impact on performance than the ambiguity set or its radius. We illustrate via case studies how a data-driven, inductive understanding of distribution shifts can provide a new approach to algorithm development.

cs.LG↗

Local Sensitivity Under Transport Restrictions

We quantify the value of structural knowledge, restrictions a modeler places on the world before seeing data. Our analytic workhorse is the local sensitivity of an estimand to distributional perturbations in the Otto-Wasserstein geometry: the largest first-order change in the estimand per unit displacement of probability mass (transport), equal to the dual norm of the spatial gradient of the efficient influence function. The modeler encodes structural knowledge by restricting the class of transports, and the resulting reduction in sensitivity is the value of their inductive bias. We illustrate by shedding light on a longstanding puzzle in causal inference where classical semiparametric efficiency bounds remain identical regardless of whether the propensity score is known, despite observed practical difficulties when it is unknown. Our approach characterizes how known propensities significantly reduce sensitivity to misspecification.

math.ST↗

Learning-To-Measure: In-Context Active Feature Acquisition

Active feature acquisition (AFA) is a sequential decision-making problem where the goal is to improve model performance for test instances by adaptively selecting which features to acquire. In practice, AFA methods often learn from retrospective data with systematic missingness in the features and limited task-specific labels. Most prior work addresses acquisition for a single predetermined task, limiting scalability. To address this limitation, we formalize the meta-AFA problem, where the goal is to learn acquisition policies across various tasks. We introduce Learning-to-Measure (L2M), which consists of i) reliable uncertainty quantification over unseen tasks, and ii) an uncertainty-guided greedy feature acquisition agent that maximizes conditional mutual information. We demonstrate a sequence-modeling or autoregressive pre-training approach that underpins reliable uncertainty quantification for tasks with arbitrary missingness. L2M operates directly on datasets with retrospective missingness and performs the meta-AFA task in-context, eliminating per-task retraining. Across synthetic and real-world tabular benchmarks, L2M matches or surpasses task-specific baselines, particularly under scarce labels and high missingness.

cs.LG↗

A Broader View of Thompson Sampling

Thompson Sampling is one of the most widely used and studied bandit algorithms, known for its simple structure, low regret performance, and solid theoretical guarantees. Yet, in stark contrast to most other families of bandit algorithms, the exact mechanism through which posterior sampling (as introduced by Thompson) is able to "properly" balance exploration and exploitation, remains a mystery. In this paper, we show that the core insight to address this question stems from recasting Thompson Sampling as an online optimization algorithm. To distill this, we introduce a suitable time invariant notion of regret that leads to a stationarized bandit problem, and a stationary Bellman-optimal policy. We then show that Thompson Sampling admits an online optimization form that mimics the structure of the aforementioned Bellman-optimal policy, where "greediness" is regularized by a measure of residual uncertainty. This new lens of online optimization allows both a better understanding of Thompson Sampling dynamics, as well as a principled manner for policy improvement that mimics the Bellman-optimal benchmark.

cs.LG↗

SynthTools: A Framework for Scaling Synthetic Tools for Agent Development

For agentic systems to use external tools to solve complex, long-horizon tasks, we need a large set of diverse and controllable tool-use environments. We introduce SynthTools, a fully LLM-based pipeline spanning the entire lifecycle: environment generation, simulation, validation and task construction. By operating end-to-end through LLMs, our framework complements other tool-use environments bottlenecked by the complexity of real APIs, and ensures scalability and controllability by design. The framework consists of three components: top-down environment generation, which hierarchically constructs diverse, domain-grounded tool environments; environment simulation and validation, which ensures tools can be reliably emulated and filters out those that cannot; and bottom-up task and trajectory generation, which produces solvable and verifiable tasks together with multi-step trajectories, exposing control over difficulty, length, trajectory composition, and domain focus to guarantee flexibility. As a concrete instantiation, we release the dataset comprising $73{,}883$ validated tools across $6{,}800$ environments and $100$ fields, $79{,}925$ verifiable tasks as well as the pipeline to generate trajectories at scale. Training Qwen3 models of various sizes on a corpus of trajectories generated from these tasks yields gains across multiple tool-use benchmarks, including real APIs, indicating tool-use capabilities trained on synthetic data may transfer to some real environments. Together, these results suggest that SynthTools can serve as a useful infrastructure for large-scale training of tool-use agents.

cs.AI↗