arXiv subjects

Yiwen Qiu

Publications and source records attributed to Yiwen Qiu.

12 recordsLinked to original sources

arXiv2026

When Metrics Reward the Worst Translations: Internalizing Cultural Reasoning for Social Media Translation Evaluation

Automatic translation quality metrics trained on general-domain corpora systematically fail on social media content, where communicative intent is encoded in culturally loaded expressions (internet slang, homophonic ciphers, and platform-specific idioms) rather than surface token patterns. We conduct a systematic empirical analysis demonstrating that standard metrics including COMET, XCOMET, and BERTScore exhibit near-zero or negative correlation with human cultural judgments, and even display a severity inversion in which scores increase as translation quality deteriorates. We further show that this failure extends to large language model judges: Qwen3-235B achieves Cohen's kappa of only 0.162, revealing that the bottleneck is not reasoning capacity but cultural grounding: models lack the domain-specific cultural knowledge needed to identify which aspects of a translation require scrutiny. To address this, we propose CuRIL, a reinforcement learning framework that internalizes cultural reasoning: cultural annotations are prepended inside the model's reasoning, excluded from policy gradients via a token-level loss mask, and injected with a probability that decays to zero over training, progressively forcing autonomous cultural judgment. On a 1,444-sample human-annotated social media translation benchmark, Qwen3-8B trained with CuRIL achieves Cohen's kappa 0.370 and Exact Match accuracy of 45.22%, approaching Gemini-3.1-Pro with 30x fewer parameters and surpassing models up to 235B in scale. We further demonstrate that our judge produces reliable reward signals for downstream translation optimization, reducing the low-quality translation rate by over 20 percentage points under independent human evaluation.

cs.CL↗

arXiv2026

Agent-G$^2$: Gaussian Guidance for Agentic Reinforcement Learning

Hint-based reinforcement learning addresses reward sparsity in long-horizon agentic tasks by retaining a prefix of an expert trajectory before each rollout, letting the policy explore from a state closer to success. Its effectiveness hinges on the guidance depth: how much of the trajectory to keep. Existing methods treat this depth as a deterministic scalar. Scheduled approaches share one value across samples and ignore per-task heterogeneity; per-sample probing estimates it separately at the cost of extra rollouts. We find that useful guidance occupies a band of depths whose informativeness profile is approximately Gaussian around the band center, rather than concentrating at a single optimal point. We propose Agent-G$^2$, a Gaussian guidance framework that draws the depth per task from a Gaussian whose center and spread are estimated online from rollouts already collected for policy optimization, requiring no probe rollouts or learned depth predictor. The center combines a global baseline with per-cluster difficulty, and the spread tracks within-cluster variance. We evaluate Agent-G$^2$ on ALFWorld and WebShop on Qwen2.5-1.5B / 7B-Instruct. Agent-G$^2$ outperforms the strongest hint-based, hint-free, and Aux-RL baselines on ALFWorld by 2.3 / 3.9 / 7.4 points at under one-third the rollout cost of per-sample probing.

cs.AI↗

arXiv2026

Pass the Baton: Trajectory-Relayed On-Policy Distillation

On-policy distillation (OPD) grounds token-level supervision in the student's own trajectory, yet suffers from prefix failure: once the student commits to a wrong reasoning direction, all subsequent generation builds on this deviation, producing misdirected continuations that elicit unreliable supervision and waste compute. We identify a teacher-student continuation asymmetry on failed prefixes, where the teacher tends to redirect while the student continues along the original direction, and convert it into a label-free handoff trigger in Relay On-Policy Distillation (Relay-OPD). During training, Relay-OPD constructs relay trajectories by letting the teacher briefly take over at detected trigger points to produce a teacher leg, after which the student resumes and is optimized on the resulting trajectory. A limited relay budget concentrates intervention on critical early positions while limiting departure from the student policy. With a Qwen3-4B-Instruct-2507 teacher and Qwen3-0.6B/1.7B-Non-Thinking students on eight mathematical reasoning benchmarks, Relay-OPD achieves the best or second-best results on every benchmark, outperforming standard OPD by +5.73% and the strongest baseline FastOPD by +1.49% on average for 1.7B, with consistent gains at 0.6B. Training trajectory length is reduced by over 50%.

cs.CL↗

arXiv2026

Towards a holistic understanding of Selection Bias for Causal Effect Identification

Selection bias is pervasive in observational studies. For example, large scale biobanks data can exhibit ``healthy volunteer bias'' when respondents are healthier and of higher socio-economic status than the population they are meant to represent. Recovering causal effects from such sub-population is an important problem in causal inference, as estimating average treatment effects (ATE) from selected populations can result in a severely biased estimate of the ATE from the whole population. In this paper, we investigate the identifiability of the ATE under selection bias. We provide necessary and sufficient conditions for ATE identifiability, leveraging weak assumptions on probability classes to characterize propensity score and selection probability. Compared to previous works, our results extend existing graphical identifiability criteria and offer a more comprehensive understanding of causal effect identification with strictly weaker conditions in the presence of selection bias.

stat.ME↗

arXiv2026

Pause or Fabricate? Training Language Models for Grounded Reasoning

Large language models have achieved remarkable progress on complex reasoning tasks. However, they often implicitly fabricate information when inputs are incomplete, producing confident but unreliable conclusions -- a failure mode we term ungrounded reasoning. We argue that this issue arises not from insufficient reasoning capability, but from the lack of inferential boundary awareness -- the ability to recognize when the necessary premises for valid inference are missing. To address this issue, we propose Grounded Reasoning via Interactive Reinforcement Learning (GRIL), a multi-turn reinforcement learning framework for grounded reasoning under incomplete information. GRIL decomposes the reasoning process into two stages: clarify and pause, which identifies whether the available information is sufficient, and grounded reasoning, which performs task solving once the necessary premises are established. We design stage-specific rewards to penalize hallucinations, enabling models to detect gaps, stop proactively, and resume reasoning after clarification. Experiments on GSM8K-Insufficient and MetaMATH-Insufficient show that GRIL significantly improves premise detection (up to 45%), leading to a 30% increase in task success while reducing average response length by over 20%. Additional analyses confirm robustness to noisy user responses and generalization to out-of-distribution tasks.

cs.CL↗

arXiv2025

Latent Variable Causal Discovery under Selection Bias

Addressing selection bias in latent variable causal discovery is important yet underexplored, largely due to a lack of suitable statistical tools: While various tools beyond basic conditional independencies have been developed to handle latent variables, none have been adapted for selection bias. We make an attempt by studying rank constraints, which, as a generalization to conditional independence constraints, exploits the ranks of covariance submatrices in linear Gaussian models. We show that although selection can significantly complicate the joint distribution, interestingly, the ranks in the biased covariance matrices still preserve meaningful information about both causal structures and selection mechanisms. We provide a graph-theoretic characterization of such rank constraints. Using this tool, we demonstrate that the one-factor model, a classical latent variable model, can be identified under selection bias. Simulations and real-world experiments confirm the effectiveness of using our rank constraints.

cs.LG↗

arXiv2025

LAPO: Internalizing Reasoning Efficiency via Length-Adaptive Policy Optimization

Large reasoning models have achieved remarkable performance through extended chain-of-thought sequences, yet this computational freedom leads to excessive token generation even for simple problems. We present Length-Adaptive Policy Optimization (LAPO), a novel framework that transforms reasoning length control from an external constraint into an intrinsic model capability. Unlike existing approaches that impose rigid limits or rely on post-hoc interventions, LAPO enables models to internalize an understanding of appropriate reasoning depth through a two-stage reinforcement learning process. In the first stage, models learn natural reasoning patterns by discovering the statistical distribution of successful solution lengths. The second stage leverages these patterns as meta-cognitive guidance, embedding them directly within the model's reasoning context to ensure inference-time flexibility. Experiments on mathematical reasoning benchmarks demonstrate that LAPO reduces token usage by up to 40.9% while improving accuracy by 2.3%. Our analysis reveals that models trained with LAPO develop emergent abilities to allocate computational resources based on problem complexity, achieving efficient reasoning without sacrificing quality.

cs.AI↗

arXiv2024

Identifying Selections for Unsupervised Subtask Discovery

When solving long-horizon tasks, it is intriguing to decompose the high-level task into subtasks. Decomposing experiences into reusable subtasks can improve data efficiency, accelerate policy generalization, and in general provide promising solutions to multi-task reinforcement learning and imitation learning problems. However, the concept of subtasks is not sufficiently understood and modeled yet, and existing works often overlook the true structure of the data generation process: subtasks are the results of a $\textit{selection}$ mechanism on actions, rather than possible underlying confounders or intermediates. Specifically, we provide a theory to identify, and experiments to verify the existence of selection variables in such data. These selections serve as subgoals that indicate subtasks and guide policy. In light of this idea, we develop a sequential non-negative matrix factorization (seq- NMF) method to learn these subgoals and extract meaningful behavior patterns as subtasks. Our empirical results on a challenging Kitchen environment demonstrate that the learned subtasks effectively enhance the generalization to new tasks in multi-task imitation learning scenarios. The codes are provided at https://anonymous.4open.science/r/Identifying\_Selections\_for\_Unsupervised\_Subtask\_Discovery/README.md.

cs.LG↗

arXiv2024

Detecting and Identifying Selection Structure in Sequential Data

We argue that the selective inclusion of data points based on latent objectives is common in practical situations, such as music sequences. Since this selection process often distorts statistical analysis, previous work primarily views it as a bias to be corrected and proposes various methods to mitigate its effect. However, while controlling this bias is crucial, selection also offers an opportunity to provide a deeper insight into the hidden generation process, as it is a fundamental mechanism underlying what we observe. In particular, overlooking selection in sequential data can lead to an incomplete or overcomplicated inductive bias in modeling, such as assuming a universal autoregressive structure for all dependencies. Therefore, rather than merely viewing it as a bias, we explore the causal structure of selection in sequential data to delve deeper into the complete causal process. Specifically, we show that selection structure is identifiable without any parametric assumptions or interventional experiments. Moreover, even in cases where selection variables coexist with latent confounders, we still establish the nonparametric identifiability under appropriate structural conditions. Meanwhile, we also propose a provably correct algorithm to detect and identify selection structures as well as other types of dependencies. The framework has been validated empirically on both synthetic data and real-world music.

cs.LG↗

arXiv2023

Data-Driven Predictive Control Using Closed-Loop Data: An Instrumental Variable Approach

Current data-driven predictive control (DDPC) methods heavily rely on data collected in open-loop operation with elaborate design of inputs. However, due to safety or economic concerns, systems may have to be under feedback control, where only closed-loop data are available. In this context, it remains challenging to implement DDPC using closed-loop data. In this paper, we propose a new DDPC method using closed-loop data by means of instrumental variables (IVs). By drawing from closed-loop subspace identification, the use of two forms of IVs is suggested to address the closed-loop issues caused by feedback control and the correlation between inputs and noise. Furthermore, a new DDPC formulation with a novel IV-inspired regularizer is proposed, where a balance between control cost minimization and weighted least-squares data fitting can be made for improvement of control performance. Numerical examples and application to a simulated industrial furnace showcase the improved performance of the proposed DDPC based on closed-loop data.

math.OC↗

arXiv2022

Out-of-Dynamics Imitation Learning from Multimodal Demonstrations

Existing imitation learning works mainly assume that the demonstrator who collects demonstrations shares the same dynamics as the imitator. However, the assumption limits the usage of imitation learning, especially when collecting demonstrations for the imitator is difficult. In this paper, we study out-of-dynamics imitation learning (OOD-IL), which relaxes the assumption to that the demonstrator and the imitator have the same state spaces but could have different action spaces and dynamics. OOD-IL enables imitation learning to utilize demonstrations from a wide range of demonstrators but introduces a new challenge: some demonstrations cannot be achieved by the imitator due to the different dynamics. Prior works try to filter out such demonstrations by feasibility measurements, but ignore the fact that the demonstrations exhibit a multimodal distribution since the different demonstrators may take different policies in different dynamics. We develop a better transferability measurement to tackle this newly-emerged challenge. We firstly design a novel sequence-based contrastive clustering algorithm to cluster demonstrations from the same mode to avoid the mutual interference of demonstrations from different modes, and then learn the transferability of each demonstration with an adversarial-learning based algorithm in each cluster. Experiment results on several MuJoCo environments, a driving environment, and a simulated robot environment show that the proposed transferability measurement more accurately finds and down-weights non-transferable demonstrations and outperforms prior works on the final imitation learning performance. We show the videos of our experiment results on our website.

cs.RO↗

arXiv2022

When to Trust Your Simulator: Dynamics-Aware Hybrid Offline-and-Online Reinforcement Learning

Learning effective reinforcement learning (RL) policies to solve real-world complex tasks can be quite challenging without a high-fidelity simulation environment. In most cases, we are only given imperfect simulators with simplified dynamics, which inevitably lead to severe sim-to-real gaps in RL policy learning. The recently emerged field of offline RL provides another possibility to learn policies directly from pre-collected historical data. However, to achieve reasonable performance, existing offline RL algorithms need impractically large offline data with sufficient state-action space coverage for training. This brings up a new question: is it possible to combine learning from limited real data in offline RL and unrestricted exploration through imperfect simulators in online RL to address the drawbacks of both approaches? In this study, we propose the Dynamics-Aware Hybrid Offline-and-Online Reinforcement Learning (H2O) framework to provide an affirmative answer to this question. H2O introduces a dynamics-aware policy evaluation scheme, which adaptively penalizes the Q function learning on simulated state-action pairs with large dynamics gaps, while also simultaneously allowing learning from a fixed real-world dataset. Through extensive simulation and real-world tasks, as well as theoretical analysis, we demonstrate the superior performance of H2O against other cross-domain online and offline RL algorithms. H2O provides a brand new hybrid offline-and-online RL paradigm, which can potentially shed light on future RL algorithm design for solving practical real-world tasks.

cs.LG↗