arXiv subjects

He Bai

Publications and source records attributed to He Bai.

At least 19 recordsLinked to original sources

arXiv2026

Comparing multilevel and fixed effect approaches in the generalized linear model setting

We extend prior work comparing linear multilevel models (MLM) and fixed effect (FE) models to the generalized linear model (GLM) setting, where the coefficient on a treatment variable is of primary interest. This leads to three insights. (i) First, as in the linear setting, MLM can be thought of as a regularized form of FE (RegFE). This explains why group-level confounding can greatly bias MLM's treatment coefficient estimates. However, unlike the linear setting, there is not an exact equivalence between MLM, fit through (approximate) maximum likelihood estimation, and RegFE in GLMs. (ii) Second, we study a generalization of "bias-corrected MLM" (bcMLM) to the GLM setting, and a corresponding "bias-corrected RegFE" (bcRegFE). None of FE, bcMLM, or bcRegFE entirely solve MLM's bias problem in GLMs, but bcMLM and bcRegFE tend to show less bias than does FE. (iii) Third, as in the linear setting, MLM's default standard errors can misspecify the true intragroup dependence structure in the GLM setting, which can yield downwardly biased standard errors. A cluster bootstrap is a more agnostic alternative. We also consider a cluster-robust standard error for (bc)RegFE. Ultimately, for non-linear GLMs, we recommend bcMLM for estimating the treatment coefficient, and a cluster bootstrap for standard errors and confidence intervals. If a bootstrap is not computationally feasible, then we recommend bcRegFE with cluster-robust standard errors, or FE with cluster-robust standard errors when group sizes are larger. We apply our recommendations to estimate the effects of taking a mathematics course in twelfth grade for students in the Los Angeles Unified School District. Our mlmfe package easily implements our recommendations in the R programming language.

stat.ME↗

arXiv2026

Targeted Function Balancing: Flexible balancing weights that leverage strategic imbalance to estimate causal effects

When estimating the average treatment effect of a binary intervention with weights, investigators are often burdened with how best to leave imbalance in the distributions of observed covariates. This requires that investigators proceed cautiously, as imbalance in key (nonlinear transformations of) covariates potentially introduces bias into the ultimate estimated effect. This paper introduces Targeted Function Balancing (TFB), a flexible covariate balancing weights approach that simplifies this problem for the researcher. TFB first regresses an outcome on covariates, and then selects weights that balance functions (of the covariates) that are probabilistically near the resulting regression function. This yields balance in the regression function's predicted values and the covariates, with the regression function's estimated variance determining how much balance in the covariates is sufficient. Notably, TFB demonstrates that intentionally leaving imbalance in some covariates can increase efficiency without introducing bias, challenging traditions that warn against imbalance in any variable. Though, TFB still does prioritize balance in the covariates that are most influential on the outcome. Additionally, TFB is entirely defined by a regression function and its estimated variance, turning the problem of how best to balance the covariates into how best to model the outcome. Kernel regularized least squares, the LASSO, and Bayesian Additive Regression Trees are considered as regression estimators. We apply TFB to estimate the effects of taking mathematics courses in twelfth grade in the Los Angeles Unified School District. The R Package tfb implements TFB.

stat.ME↗

arXiv2026

Data-Driven Optimization of Tactile Sensor Configurations for Efficient Dexterous Manipulation

Tactile sensing is critical for learning-based dexterous manipulation, yet principled guidelines for sensor placement remain largely absent. While dense sensor arrays provide rich contact feedback, they impose significant hardware costs and can even degrade policy performance by introducing redundant or conflicting inputs. This paper presents the first systematic framework for quantifying the contribution of individual tactile sensors to deep reinforcement learning (DRL) policy performance. We propose a two-stage approach: a coarse empirical pruning phase that reduces the sensor count on the Shadow Hand from 92 to 21 while retaining 93\% task performance, followed by a fine-grained active learning phase that combines Gaussian Process Regression (GPR) with Lasso regression to rank the functional importance of each remaining sensor. Our analysis reveals that sensors on the thumb, ring finger, and little finger dominate manipulation performance, while middle-finger sensors exhibit negative contributions -- actively degrading policy learning. Ablation studies across three manipulation tasks (block, egg, and pen) confirm that a 14-sensor configuration preserves over 90\% of the full-array performance. Zero-shot transfer experiments on two novel objects and cross-platform validation on the Allegro and Leap Hand further demonstrate that the identified importance rankings generalize across tasks and robot morphologies. These findings establish quantitative deployment guidelines that enable practitioners to select cost-effective sensor configurations with predictable performance trade-offs.

cs.RO↗

arXiv2026

TRACER: Verifiable Generative Provenance for Multimodal Tool-Using Agents

Multimodal large language models increasingly solve vision-centric tasks by calling external tools for visual inspection, OCR, retrieval, calculation, and multi-step reasoning. Current tool-using agents usually expose the executed tool trajectory and the final answer, but they rarely specify which tool observation supports each generated claim. We call this missing claim-level dependency structure the provenance gap. The gap makes tool use hard to verify and hard to optimize, because useful evidence, redundant exploration, and unsupported reasoning are mixed in the same trajectory. We introduce TRACER, a framework for verifiable generative provenance in multimodal tool-using agents. Instead of adding citations after generation, TRACER generates each answer sentence together with a structured provenance record that identifies the supporting tool turn, evidence unit, and semantic support relation. Its relation space contains Quotation, Compression, and Inference, covering direct reuse, faithful condensation, and grounded derivation. TRACER verifies each record through schema checking, tool-turn alignment, source authenticity, and relation rationality, and then converts verified provenance into traceability constraints and provenance-derived local credit for reinforcement learning. We further construct TRACE-Bench, a benchmark for sentence-level provenance reconstruction from coarse multimodal tool trajectories. On TRACE-Bench, simply adding tools often introduces noise. With Qwen3-VL-8B, TRACER reaches 78.23% answer accuracy and 95.72% summary accuracy, outperforming the strongest closed-source tool-augmented baseline by 23.80 percentage points. Compared with tool-only supervised fine-tuning, it also reduces total test-set tool calls from 4949 to 3486. These results show that reliable multimodal tool reasoning depends on provenance-aware use of observations, not on more tool calls alone.

cs.CL↗

arXiv2026

From Past To Path: Masked History Learning for Next-Item Prediction in Generative Recommendation

Generative recommendation, which directly generates item identifiers, has emerged as a promising paradigm for recommendation systems. However, its potential is fundamentally constrained by the reliance on purely autoregressive training. This approach focuses solely on predicting the next item while ignoring the rich internal structure of a user's interaction history, thus failing to grasp the underlying intent. To address this limitation, we propose Masked History Learning (MHL), a novel training framework that shifts the objective from simple next-step prediction to deep comprehension of history. MHL augments the standard autoregressive objective with an auxiliary task of reconstructing masked historical items, compelling the model to understand ``why'' an item path is formed from the user's past behaviors, rather than just ``what'' item comes next. We introduce two key contributions to enhance this framework: (1) an entropy-guided masking policy that intelligently targets the most informative historical items for reconstruction, and (2) a curriculum learning scheduler that progressively transitions from history reconstruction to future prediction. Experiments on three public datasets show that our method significantly outperforms state-of-the-art generative models, highlighting that a comprehensive understanding of the past is crucial for accurately predicting a user's future path.

cs.IR↗

arXiv2026

Revisiting ASR Error Correction with Specialized Models

Language models play a central role in automatic speech recognition (ASR), yet most methods rely on text-only models unaware of ASR error patterns. Recently, large language models (LLMs) have been applied to ASR correction, but introduce latency and hallucination concerns. We revisit ASR error correction with compact seq2seq models, trained on ASR errors from real and synthetic audio. To scale training, we construct synthetic corpora via cascaded TTS and ASR, finding that matching the diversity of realistic error distributions is key. We propose correction-first decoding, where the correction model generates candidates rescored using ASR acoustic scores. With 15x fewer parameters than LLMs, our model achieves 1.5/3.3% WER on LibriSpeech test-clean/other, outperforms LLMs, generalizes across ASR architectures (CTC, Seq2seq, Transducer) and diverse domains, and provides precise corrections in the low-error regime where LLMs struggle.

cs.LG↗

arXiv2026

GenMRP: A Generative Multi-Route Planning Framework for Efficient and Personalized Real-Time Industrial Navigation

Existing industrial-scale navigation applications contend with massive road networks, typically employing two main categories of approaches for route planning. The first relies on precomputed road costs for optimal routing and heuristic algorithms for generating alternatives, while the second, generative methods, has recently gained significant attention. However, the former struggles with personalization and route diversity, while the latter fails to meet the efficiency requirements of large-scale real-time scenarios. To address these limitations, we propose GenMRP, a generative framework for multi-route planning. To ensure generation efficiency, GenMRP first introduces a skeleton-to-capillary approach that dynamically constructs a relevant sub-network significantly smaller than the full road network. Within this sub-network, routes are generated iteratively. The first iteration identifies the optimal route, while the subsequent ones generate alternatives that balance quality and diversity using the newly proposed correctional boosting approach. Each iteration incorporates road features, user historical sequences, and previously generated routes into a Link Cost Model to update road costs, followed by route generation using the Dijkstra algorithm. Extensive experiments show that GenMRP achieves state-of-the-art performance with high efficiency in both offline and online environments. To facilitate further research, we have publicly released the training and evaluation dataset. GenMRP has been fully deployed in a real-world navigation app, demonstrating its effectiveness and benefits.

cs.RO↗

arXiv2025

Structured Cooperative Multi-Agent Reinforcement Learning: a Bayesian Network Perspective

The empirical success of multi-agent reinforcement learning (MARL) has motivated the search for more efficient and scalable algorithms for large scale multi-agent systems. However, existing state-of-the-art algorithms do not fully exploit inter-agent coupling information to develop MARL algorithms. In this paper, we propose a systematic approach to leverage structures in the inter-agent couplings for efficient model-free reinforcement learning. We model the cooperative MARL problem via a Bayesian network and characterize the subset of agents, termed as the value dependency set, whose information is required by each agent to estimate its local action value function exactly. Moreover, we propose a partially decentralized training decentralized execution (P-DTDE) paradigm based on the value dependency set. We theoretically establish that the total variance of our P-DTDE policy gradient estimator is less than the centralized training decentralized execution (CTDE) policy gradient estimator. We derive a multi-agent policy gradient theorem based on the P-DTDE scheme and develop a scalable actor-critic algorithm. We demonstrate the efficiency and scalability of the proposed algorithm on multi-warehouse resource allocation and multi-zone temperature control examples. For dense value dependency sets, we propose an approximation scheme based on truncation of the Bayesian network and empirically show that it achieves a faster convergence than the exact value dependence set for applications with a large number of agents.

cs.MA↗

arXiv2025

Approximate constrained stochastic optimal control via parameterized input inference

Approximate methods to solve stochastic optimal control (SOC) problems have received significant interest from researchers in the past decade. Probabilistic inference approaches to SOC have been developed to solve nonlinear quadratic Gaussian problems. In this work, we propose an Expectation-Maximization (EM) based inference procedure to generate state-feedback controls for constrained SOC problems. We consider the inequality constraints for the state and controls and also the structural constraints for the controls. We employ barrier functions to address state and control constraints. We show that the expectation step leads to smoothing of the state-control pair while the the maximization step on the non-zero subsets of the control parameters allows inference of structured stochastic optimal controllers. We demonstrate the effectiveness of the algorithm on unicycle obstacle avoidance, four-unicycle formation control, and quadcopter navigation in windy environment examples. In these examples, we perform an empirical study on the parametric effect of barrier functions on the state constraint satisfaction. We also present a comparative study of smoothing algorithms on the performance of the proposed approach.

eess.SY↗

arXiv2025

Exploiting inter-agent coupling information for efficient reinforcement learning of cooperative LQR

Developing scalable and efficient reinforcement learning algorithms for cooperative multi-agent control has received significant attention over the past years. Existing literature has proposed inexact decompositions of local Q-functions based on empirical information structures between the agents. In this paper, we exploit inter-agent coupling information and propose a systematic approach to exactly decompose the local Q-function of each agent. We develop an approximate least square policy iteration algorithm based on the proposed decomposition and identify two architectures to learn the local Q-function for each agent. We establish that the worst-case sample complexity of the decomposition is equal to the centralized case and derive necessary and sufficient graphical conditions on the inter-agent couplings to achieve better sample efficiency. We demonstrate the improved sample efficiency and computational efficiency on numerical examples.

eess.SY↗

arXiv2025

Exploring Prediction Targets in Masked Pre-Training for Speech Foundation Models

Speech foundation models, such as HuBERT and its variants, are pre-trained on large amounts of unlabeled speech data and then used for a range of downstream tasks. These models use a masked prediction objective, where the model learns to predict information about masked input segments from the unmasked context. The choice of prediction targets in this framework impacts their performance on downstream tasks. For instance, models pre-trained with targets that capture prosody learn representations suited for speaker-related tasks, while those pre-trained with targets that capture phonetics learn representations suited for content-related tasks. Moreover, prediction targets can differ in the level of detail they capture. Models pre-trained with targets that encode fine-grained acoustic features perform better on tasks like denoising, while those pre-trained with targets focused on higher-level abstractions are more effective for content-related tasks. Despite the importance of prediction targets, the design choices that affect them have not been thoroughly studied. This work explores the design choices and their impact on downstream task performance. Our results indicate that the commonly used design choices for HuBERT can be suboptimal. We propose approaches to create more informative prediction targets and demonstrate their effectiveness through improvements across various downstream tasks.

eess.AS↗

arXiv2024

Confined Magnetization at the Sublattice-Matched Ruthenium Oxide Heterointerface

Creating a heterostructure by combining two magnetically and structurally distinct ruthenium oxides is a crucial approach for investigating their emergent magnetic states and interactions. Previously, research has predominantly concentrated on the intrinsic properties of the ferromagnet SrRuO3 and recently discovered altermagnet RuO2 solely. Here, we engineered an ultrasharp sublattice-matched heterointerface using pseudo-cubic SrRuO3 and rutile RuO2, conducting an in-depth analysis of their spin interactions. Structurally, to accommodate the lattice symmetry mismatch, the inverted RuO2 layer undergoes an in-plane rotation of 18 degrees during epitaxial growth on SrRuO3 layer, resulting in an interesting and rotational interface with perfect crystallinity and negligible chemical intermixing. Performance-wise, the interfacial layer of 6 nm in RuO2 adjacent to SrRuO3 exhibits a nonzero magnetic moment, contributing to an enhanced anomalous Hall effect (AHE) at low temperatures. Furthermore, our observations indicate that, in contrast to SrRuO3 single layers, the AHE of [(RuO2)15/(SrRuO3)n] heterostructures shows nonlinear behavior and reaches its maximum when the SrRuO3 thickness reaches tens of nm. These results suggest that the interfacial magnetic interaction surpasses that of all-perovskite oxides (~5-unit cells). This study underscores the significance and potential applications of magnetic interactions based on the crystallographic asymmetric interfaces in the design of spintronic devices.

cond-mat.mtrl-sci↗

arXiv2024

Divide-or-Conquer? Which Part Should You Distill Your LLM?

Recent methods have demonstrated that Large Language Models (LLMs) can solve reasoning tasks better when they are encouraged to solve subtasks of the main task first. In this paper we devise a similar strategy that breaks down reasoning tasks into a problem decomposition phase and a problem solving phase and show that the strategy is able to outperform a single stage solution. Further, we hypothesize that the decomposition should be easier to distill into a smaller model compared to the problem solving because the latter requires large amounts of domain knowledge while the former only requires learning general problem solving strategies. We propose methods to distill these two capabilities and evaluate their impact on reasoning outcomes and inference cost. We find that we can distill the problem decomposition phase and at the same time achieve good generalization across tasks, datasets, and models. However, it is harder to distill the problem solving capability without losing performance and the resulting distilled model struggles with generalization. These results indicate that by using smaller, distilled problem decomposition models in combination with problem solving LLMs we can achieve reasoning with cost-efficient inference and local adaptation.

cs.CL↗

arXiv2024

How Far Are We from Intelligent Visual Deductive Reasoning?

Vision-Language Models (VLMs) have recently demonstrated incredible strides on diverse vision language tasks. We dig into vision-based deductive reasoning, a more sophisticated but less explored realm, and find previously unexposed blindspots in the current SOTA VLMs. Specifically, we leverage Raven's Progressive Matrices (RPMs), to assess VLMs' abilities to perform multi-hop relational and deductive reasoning relying solely on visual clues. We perform comprehensive evaluations of several popular VLMs employing standard strategies such as in-context learning, self-consistency, and Chain-of-thoughts (CoT) on three diverse datasets, including the Mensa IQ test, IntelligenceTest, and RAVEN. The results reveal that despite the impressive capabilities of LLMs in text-based reasoning, we are still far from achieving comparable proficiency in visual deductive reasoning. We found that certain standard strategies that are effective when applied to LLMs do not seamlessly translate to the challenges presented by visual reasoning tasks. A detailed analysis reveals that VLMs struggle to solve these tasks mainly because they are unable to perceive and comprehend multiple, confounding abstract patterns in RPM examples.

cs.AI↗

arXiv2024

Reinforcement Learning-based Control of Nonlinear Systems using Carleman Approximation: Structured and Unstructured Designs

We develop data-driven reinforcement learning (RL) control designs for input-affine nonlinear systems. We use Carleman linearization to express the state-space representation of the nonlinear dynamical model in the Carleman space, and develop a real-time algorithm that can learn nonlinear state-feedback controllers using state and input measurements in the infinite-dimensional Carleman space. Thereafter, we study the practicality of having a finite-order truncation of the control signal, followed by its closed-loop stability analysis. Finally, we develop two additional designs that can learn structured as well as sparse representations of the RL-based nonlinear controller, and provide theoretical conditions for ensuring their closed-loop stability. We present numerical examples to show how our proposed method generates closed-loop responses that are close to the optimal performance of the nonlinear plant. We also compare our designs to other data-driven nonlinear RL control methods such as those based on neural networks, and illustrate their relative advantages and drawbacks.

eess.SY↗

arXiv2024

KGLens: Towards Efficient and Effective Knowledge Probing of Large Language Models with Knowledge Graphs

Large Language Models (LLMs) might hallucinate facts, while curated Knowledge Graph (KGs) are typically factually reliable especially with domain-specific knowledge. Measuring the alignment between KGs and LLMs can effectively probe the factualness and identify the knowledge blind spots of LLMs. However, verifying the LLMs over extensive KGs can be expensive. In this paper, we present KGLens, a Thompson-sampling-inspired framework aimed at effectively and efficiently measuring the alignment between KGs and LLMs. KGLens features a graph-guided question generator for converting KGs into natural language, along with a carefully designed importance sampling strategy based on parameterized KG structure to expedite KG traversal. Our simulation experiment compares the brute force method with KGLens under six different sampling methods, demonstrating that our approach achieves superior probing efficiency. Leveraging KGLens, we conducted in-depth analyses of the factual accuracy of ten LLMs across three large domain-specific KGs from Wikidata, composing over 19K edges, 700 relations, and 21K entities. Human evaluation results indicate that KGLens can assess LLMs with a level of accuracy nearly equivalent to that of human annotators, achieving 95.7% of the accuracy rate.

cs.AI↗

arXiv2024

Real-time Dexterous Telemanipulation with an End-Effect-Oriented Learning-based Approach

Dexterous telemanipulation is crucial in advancing human-robot systems, especially in tasks requiring precise and safe manipulation. However, it faces significant challenges due to the physical differences between human and robotic hands, the dynamic interaction with objects, and the indirect control and perception of the remote environment. Current approaches predominantly focus on mapping the human hand onto robotic counterparts to replicate motions, which exhibits a critical oversight: it often neglects the physical interaction with objects and relegates the interaction burden to the human to adapt and make laborious adjustments in response to the indirect and counter-intuitive observation of the remote environment. This work develops an End-Effects-Oriented Learning-based Dexterous Telemanipulation (EFOLD) framework to address telemanipulation tasks. EFOLD models telemanipulation as a Markov Game, introducing multiple end-effect features to interpret the human operator's commands during interaction with objects. These features are used by a Deep Reinforcement Learning policy to control the robot and reproduce such end effects. EFOLD was evaluated with real human subjects and two end-effect extraction methods for controlling a virtual Shadow Robot Hand in telemanipulation tasks. EFOLD achieved real-time control capability with low command following latency (delay<0.11s) and highly accurate tracking (MSE<0.084 rad).

cs.RO↗

arXiv2024

Asynchronous Distributed Reinforcement Learning for LQR Control via Zeroth-Order Block Coordinate Descent

Recently introduced distributed zeroth-order optimization (ZOO) algorithms have shown their utility in distributed reinforcement learning (RL). Unfortunately, in the gradient estimation process, almost all of them require random samples with the same dimension as the global variable and/or require evaluation of the global cost function, which may induce high estimation variance for large-scale networks. In this paper, we propose a novel distributed zeroth-order algorithm by leveraging the network structure inherent in the optimization objective, which allows each agent to estimate its local gradient by local cost evaluation independently, without use of any consensus protocol. The proposed algorithm exhibits an asynchronous update scheme, and is designed for stochastic non-convex optimization with a possibly non-convex feasible domain based on the block coordinate descent method. The algorithm is later employed as a distributed model-free RL algorithm for distributed linear quadratic regulator design, where a learning graph is designed to describe the required interaction relationship among agents in distributed learning. We provide an empirical validation of the proposed algorithm to benchmark its performance on convergence rate and variance against a centralized ZOO algorithm.

eess.SY↗