arXiv subjects

Yufan Zhuang

Publications and source records attributed to Yufan Zhuang.

At least 19 recordsLinked to original sources

arXiv2026

RRSI: Regularized Recursive Self-Improvement of Agent Harnesses

An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iteratively proposing and selecting component-wise edits of an agent harness, practically establishing a form of recursive self-improvement (RSI) at the agent-system level. However, such recursive evolution may overfit by memorizing the training tasks, showing large in-distribution gains that shrink or even vanish on out-of-distribution benchmarks. We introduce Regularized Recursive Self-Improvement of Agent Harnesses (RRSI), which incorporates the principles of regularizations into harness self-improvement by constraining the evolution candidate proposal and selection. The proposer operates with a temporally annealed budget, limiting how many edits a candidate can bundle, and it encourages unexplored trajectories based on evolution history. The selector is equipped with a critic and a pruner: the critic screens benchmark-specific proposals, while the pruner, removes changes that are too small, too expensive, or no longer useful. Together these constraints favor reusable agent mechanisms over benchmark-specific ones or even noises. Across eight benchmarks spanning coding, agentic workspace and engineering design tasks, RRSI gains up to 14.1 points on the split it evolves against and up to 4.7 points on the five out-of-distribution benchmarks, while producing a harness that runs on 30% fewer policy tokens than the unregularized evolution. Code is available at https://github.com/google-research/rrsi and project page is https://regularized-rsi.com/.

cs.LG↗

arXiv2026

Register Tokens for Bounded-State Reasoning in Diffusion Language Models

Masked diffusion language models (dLLMs) generate text by iteratively denoising masked tokens with bidirectional attention. Extending reasoning across generation chunks normally requires keeping earlier generated text in context. We ask whether a dLLM can instead continue reasoning after that text is cleared, using only a fixed-size carried state. We implement this state as a small number of register tokens: dedicated fixed-position tokens whose continuous hidden states are trained to carry reasoning progress across generation chunks. We post-train dLLMs to decode a chunk of text, clear it while preserving the register values, and continue decoding from the prompt and carried state. In our main comparisons on LLaDA and Dream, registers outperform discrete-text carry on every benchmark, with gains of up to 8.5 points on math and 19.5 points on code. Registers are especially effective for bounded code generation, where correct programs usually span several chunks. Finally, registers can be further refined with reinforcement learning on long-horizon reasoning tasks.

cs.CL↗

arXiv2026

EnvHarness: Awakening Static Worlds for Agent Learning

LLM agents learn by interacting with environments, yet these environments are hand-built and static: blind to an agent's weaknesses, and quickly left behind as it improves. While recent environment generation methods attempt to address this, they require domain-specific pipelines, rely on expensive or unreliable verifiers, and still produce static environments. To alleviate the engineering burden of rebuilding environments from scratch, we propose Environment Harness (EnvHarness), a programmable layer of plug-in components that wraps a static environment to reshape its behavior without modifying the underlying logic. Operating through standard interfaces, EnvHarness applies across diverse domains while ensuring every reshaped environment retains its original verifier. To automate this process, we introduce EnvRigger, which treats the target policy as a black box, observing its execution trajectories to synthesize EnvHarness components targeting diagnosed flaws, and validating them via fresh rollouts. Across five benchmarks in four domains, EnvHarness outperforms both original environments and domain-specific environment generation pipelines, achieving up to a 9.0-point improvement on held-out instances with 9.8% fewer execution steps. Furthermore, EnvHarness provides a superior optimization signal for reinforcement learning, enabling continuous, targeted co-evolution of the policy and its environment.

cs.AI↗

arXiv2026

OCC: Physical-Layer Assisted Congestion Control for Real-Time Communications

Real-time communications (RTC) is a core technology for emerging applications in 6G, such as cloud gaming, teleoperation, and extended reality (XR), which require consistently low latency and high bitrates. Existing RTC solutions fundamentally struggle to maintain low latency while supporting high bitrates due to their reliance on trial-and-error-based mechanisms. These mechanisms fail to probe the available bandwidth (ABW) promptly and accurately, leading to a trade-off between latency reliability and bandwidth utilization. The tension becomes extremely more critical as the cellular bandwidth and application's demand fluctuate with a larger range in cellular networks nowadays. To address this trade-off, we propose OCC, a novel approach that utilizes physical-layer information to explicitly obtain the ABW in real time, enabling rapid adaptation to dynamic wireless network conditions. However, the unique characteristics of RTC, including traffic bursts, application (APP) limits, and encoder lag, make the physical-layer informed control non-trivial. OCC effectively addresses these issues through three innovative strategies: frame-aware bandwidth measurement, APP-limit-aware bandwidth estimation, and encoder-friendly rate control. Extensive over-the-air experiments on an open-source cellular testbed demonstrate that OCC significantly enhances the performance of mobile RTC, reducing tail network latency by $13\%$ to $68\%$ and improving video frame bitrate by $1.2\times$ to $3.5\times$.

cs.NI↗

arXiv2026

Test-time Recursive Thinking: Self-Improvement without External Feedback

Modern Large Language Models (LLMs) have shown rapid improvements in reasoning capabilities, driven largely by reinforcement learning (RL) with verifiable rewards. Here, we ask whether these LLMs can self-improve without the need for additional training. We identify two core challenges for such systems: (i) efficiently generating diverse, high-quality candidate solutions, and (ii) reliably selecting correct answers in the absence of ground-truth supervision. To address these challenges, we propose Test-time Recursive Thinking (TRT), an iterative self-improvement framework that conditions generation on rollout-specific strategies, accumulated knowledge, and self-generated verification signals. Using TRT, open-source models reach 100% accuracy on AIME-25/24, and on LiveCodeBench's most difficult problems, closed-source models improve by 10.4-14.8 percentage points without external feedback.

cs.CL↗

arXiv2026

Mon3tr: Monocular 3D Telepresence with Pre-built Gaussian Avatars as Amortization

Immersive telepresence aims to transform human interaction in AR/VR applications by enabling lifelike full-body holographic representations for enhanced remote collaboration. However, existing systems rely on hardware-intensive multi-camera setups and demand high bandwidth for volumetric streaming, limiting their real-time performance on mobile devices. To overcome these challenges, we propose Mon3tr, a novel Monocular 3D telepresence framework that integrates 3D Gaussian splatting (3DGS) based parametric human modeling into telepresence for the first time. Mon3tr adopts an amortized computation strategy, dividing the process into a one-time offline multi-view reconstruction phase to build a user-specific avatar and a monocular online inference phase during live telepresence sessions. A single monocular RGB camera is used to capture body motions and facial expressions in real time to drive the 3DGS-based parametric human model, significantly reducing system complexity and cost. The extracted motion and appearance features are transmitted at < 0.2 Mbps over WebRTC's data channel, allowing robust adaptation to network fluctuations. On the receiver side, e.g., Meta Quest 3, we develop a lightweight 3DGS attribute deformation network to dynamically generate corrective 3DGS attribute adjustments on the pre-built avatar, synthesizing photorealistic motion and appearance at ~ 60 FPS. Extensive experiments demonstrate the state-of-the-art performance of our method, achieving a PSNR of > 28 dB for novel poses, an end-to-end latency of ~ 80 ms, and > 1000x bandwidth reduction compared to point-cloud streaming, while supporting real-time operation from monocular inputs across diverse scenarios. Our demos can be found at https://mon3tr3d.github.io.

cs.CV↗

arXiv2025

Text Generation Beyond Discrete Token Sampling

In standard autoregressive generation, an LLM predicts the next-token distribution, samples a discrete token, and then discards the distribution, passing only the sampled token as new input. To preserve this distribution's rich information, we propose Mixture of Inputs (MoI), a training-free method for autoregressive generation. After generating a token following the standard paradigm, we construct a new input that blends the generated discrete token with the previously discarded token distribution. Specifically, we employ a Bayesian estimation method that treats the token distribution as the prior, the sampled token as the observation, and replaces the conventional one-hot vector with the continuous posterior expectation as the new model input. MoI allows the model to maintain a richer internal representation throughout the generation process, resulting in improved text quality and reasoning capabilities. On mathematical reasoning, code generation, and PhD-level QA tasks, MoI consistently improves performance across multiple models including QwQ-32B, Nemotron-Super-49B, Gemma-3-27B, and DAPO-Qwen-32B, with no additional training and negligible computational overhead.

cs.CL↗

arXiv2025

Self-Taught Agentic Long Context Understanding

Answering complex, long-context questions remains a major challenge for large language models (LLMs) as it requires effective question clarifications and context retrieval. We propose Agentic Long-Context Understanding (AgenticLU), a framework designed to enhance an LLM's understanding of such queries by integrating targeted self-clarification with contextual grounding within an agentic workflow. At the core of AgenticLU is Chain-of-Clarifications (CoC), where models refine their understanding through self-generated clarification questions and corresponding contextual groundings. By scaling inference as a tree search where each node represents a CoC step, we achieve 97.8% answer recall on NarrativeQA with a search depth of up to three and a branching factor of eight. To amortize the high cost of this search process to training, we leverage the preference pairs for each step obtained by the CoC workflow and perform two-stage model finetuning: (1) supervised finetuning to learn effective decomposition strategies, and (2) direct preference optimization to enhance reasoning quality. This enables AgenticLU models to generate clarifications and retrieve relevant context effectively and efficiently in a single inference pass. Extensive experiments across seven long-context tasks demonstrate that AgenticLU significantly outperforms state-of-the-art prompting methods and specialized long-context LLMs, achieving robust multi-hop reasoning while sustaining consistent performance as context length grows.

cs.CL↗

arXiv2025

Vector-ICL: In-context Learning with Continuous Vector Representations

Large language models (LLMs) have shown remarkable in-context learning (ICL) capabilities on textual data. We explore whether these capabilities can be extended to continuous vectors from diverse domains, obtained from black-box pretrained encoders. By aligning input data with an LLM's embedding space through lightweight projectors, we observe that LLMs can effectively process and learn from these projected vectors, which we term Vector-ICL. In particular, we find that pretraining projectors with general language modeling objectives enables Vector-ICL, while task-specific finetuning further enhances performance. In our experiments across various tasks and modalities, including text reconstruction, numerical function regression, text classification, summarization, molecule captioning, time-series classification, graph classification, and fMRI decoding, Vector-ICL often surpasses both few-shot ICL and domain-specific model or tuning. We further conduct analyses and case studies, indicating the potential of LLMs to process vector representations beyond traditional token-based paradigms.

cs.CL↗

arXiv2024

Data Contamination Can Cross Language Barriers

The opacity in developing large language models (LLMs) is raising growing concerns about the potential contamination of public benchmarks in the pre-training data. Existing contamination detection methods are typically based on the text overlap between training and evaluation data, which can be too superficial to reflect deeper forms of contamination. In this paper, we first present a cross-lingual form of contamination that inflates LLMs' performance while evading current detection methods, deliberately injected by overfitting LLMs on the translated versions of benchmark test sets. Then, we propose generalization-based approaches to unmask such deeply concealed contamination. Specifically, we examine the LLM's performance change after modifying the original benchmark by replacing the false answer choices with correct ones from other questions. Contaminated models can hardly generalize to such easier situations, where the false choices can be \emph{not even wrong}, as all choices are correct in their memorization. Experimental results demonstrate that cross-lingual contamination can easily fool existing detection methods, but not ours. In addition, we discuss the potential utilization of cross-lingual contamination in interpreting LLMs' working mechanisms and in post-training LLMs for enhanced multilingual capabilities. The code and dataset we use can be obtained from \url{https://github.com/ShangDataLab/Deep-Contam}.

cs.CL↗

arXiv2024

Learning a Decision Tree Algorithm with Transformers

Decision trees are renowned for their ability to achieve high predictive performance while remaining interpretable, especially on tabular data. Traditionally, they are constructed through recursive algorithms, where they partition the data at every node in a tree. However, identifying a good partition is challenging, as decision trees optimized for local segments may not yield global generalization. To address this, we introduce MetaTree, a transformer-based model trained via meta-learning to directly produce strong decision trees. Specifically, we fit both greedy decision trees and globally optimized decision trees on a large number of datasets, and train MetaTree to produce only the trees that achieve strong generalization performance. This training enables MetaTree to emulate these algorithms and intelligently adapt its strategy according to the context, thereby achieving superior generalization performance.

cs.LG↗

arXiv2024

EdgeLoc: A Communication-Adaptive Parallel System for Real-Time Localization in Infrastructure-Assisted Autonomous Driving

This paper presents EdgeLoc, an infrastructure-assisted, real-time localization system for autonomous driving that addresses the incompatibility between traditional localization methods and deep learning approaches. The system is built on top of the Robot Operating System (ROS) and combines the real-time performance of traditional methods with the high accuracy of deep learning approaches. The system leverages edge computing capabilities of roadside units (RSUs) for precise localization to enhance on-vehicle localization that is based on the real-time visual odometry. EdgeLoc is a parallel processing system, utilizing a proposed uncertainty-aware pose fusion solution. It achieves communication adaptivity through online learning and addresses fluctuations via window-based detection. Moreover, it achieves optimal latency and maximum improvement by utilizing auto-splitting vehicle-infrastructure collaborative inference, as well as online distribution learning for decision-making. Even with the most basic end-to-end deep neural network for localization estimation, EdgeLoc realizes a 67.75\% reduction in the localization error for real-time local visual odometry, a 29.95\% reduction for non-real-time collaborative inference, and a 30.26\% reduction compared to Kalman filtering. Finally, accuracy-to-latency conversion was experimentally validated, and an overall experiment was conducted on a practical cellular network. The system is open sourced at https://github.com/LoganCome/EdgeAssistedLocalization.

cs.DC↗

arXiv2024

Rate-Splitting with Hybrid Messages: DoF Analysis of the Two-User MIMO Broadcast Channel with Imperfect CSIT

Most of the existing research on degrees-of-freedom (DoF) with imperfect channel state information at the transmitter (CSIT) assume the messages are private, which may not reflect reality as the two receivers can request the same content. To overcome this limitation, we therefore consider the hybrid unicast and multicast messages. In particular, we characterize the optimal DoF region for the two-user multiple-input multiple-output (MIMO) broadcast channel (BC) with imperfect CSIT and hybrid messages. For the converse, we establish a three-step procedure to exploit the utmost possible relaxation. For the achievability, since the DoF region is with specific three-dimensional structure regarding antenna configurations and CSIT qualities, we verify the existence or non-existence of corner point candidates via the feature of antenna configurations and CSIT qualities categorization and provide a hybrid message-aware rate-splitting scheme. Besides, we show that to achieve the strictly positive corner points, it is unnecessary to split the unicast messages into private and common parts. This implies adding a multicast message may mitigate the rate-splitting complexity.

cs.IT↗

arXiv2023

WavSpA: Wavelet Space Attention for Boosting Transformers' Long Sequence Learning Ability

Transformer and its variants are fundamental neural architectures in deep learning. Recent works show that learning attention in the Fourier space can improve the long sequence learning capability of Transformers. We argue that wavelet transform shall be a better choice because it captures both position and frequency information with linear time complexity. Therefore, in this paper, we systematically study the synergy between wavelet transform and Transformers. We propose Wavelet Space Attention (WavSpA) that facilitates attention learning in a learnable wavelet coefficient space which replaces the attention in Transformers by (1) applying forward wavelet transform to project the input sequences to multi-resolution bases, (2) conducting attention learning in the wavelet coefficient space, and (3) reconstructing the representation in input space via backward wavelet transform. Extensive experiments on the Long Range Arena demonstrate that learning attention in the wavelet space using either fixed or adaptive wavelets can consistently improve Transformer's performance and also significantly outperform learning in Fourier space. We further show our method can enhance Transformer's reasoning extrapolation capability over distance on the LEGO chain-of-reasoning task.

cs.CL↗

arXiv2022

Reconfigurable Intelligent Surface Assisted OFDM Relaying: Subcarrier Matching with Balanced SNR

This paper considers a reconfigurable intelligent surface (RIS) aided orthogonal frequency division multiplexing (OFDM) relaying system, and investigates the joint design of RIS passive beamforming and subcarrier matching under two cases, where Case-I ignores the source-RIS-destination signal, while Case-II explores this signal for rate enhancement. We formulate a mixed-integer nonlinear programming (MINIP) problem to maximize the sum achievable rate of all subcarriers by jointly optimizing the passive beamforming and subcarrier matching. To solve this problem, we first develop a branch-and-bound (BnB)-based alternating optimization algorithm for attaining a near-optimal solution. Then, a low-complexity difference-of-convex penalty-based algorithm and learning-to-optimize approach are also proposed. Finally, simulation results demonstrate that the RIS-assisted OFDM relaying system achieves a substantial achievable rate gain as compared to that without RIS since RIS recasts the subcarrier matching and balances the signal-to-noise ratio (SNR) among different subcarrier pairs.

cs.IT↗

arXiv2022

Secure Rate-Splitting for MIMO Broadcast Channel with Imperfect CSIT and a Jammer

In this paper, we investigate the secure rate-splitting for the two-user multiple-input multiple-output (MIMO) broadcast channel with imperfect channel state information at the transmitter (CSIT) and a multiple-antenna jammer, where each receiver has an equal number of antennas and the jammer has perfect channel state information (CSI). Specifically, we design a secure rate-splitting multiple-access strategy, where the security of split private and common messages is ensured by precoder design with joint nulling and aligning the leakage information, regarding different antenna configurations. Moreover, we show that the sum-secure degrees-of-freedom (SDoF) achieved by secure rate-splitting is optimal and outperforms that by conventional zero-forcing. Therefore, we reveal the sum-SDoF of the two-user MIMO broadcast channel with imperfect CSIT and a jammer, and validate the superiority of rate-splitting for the security purpose in this scenario with emphasis of MIMO.

cs.IT↗

arXiv2022

The Generalized Degrees-of-Freedom of the Asymmetric Interference Channel with Delayed CSIT

In this paper, we investigate the generalized degrees-of-freedom (GDoF) of the asymmetric interference channel with delayed channel state information at the transmitter (CSIT), where each transmitter has two antennas, each receiver has one antenna, and the strength for each interfering link can vary. The optimal sum-GDoF is characterized by matched converse and achievability proof. Through our results, we also reveal that in our antenna setting, the symmetric GDoF lower bound in [Mohanty et. al, TIT 2019] can be elevated, and the symmetric GDoF upper bound in [Mohanty et. al, TIT 2019] is tight in fact.

cs.IT↗

arXiv2022

Data-Driven AI Model Signal-Awareness Enhancement and Introspection

AI modeling for source code understanding tasks has been making significant progress, and is being adopted in production development pipelines. However, reliability concerns, especially whether the models are actually learning task-related aspects of source code, are being raised. While recent model-probing approaches have observed a lack of signal awareness in many AI-for-code models, i.e. models not capturing task-relevant signals, they do not offer solutions to rectify this problem. In this paper, we explore data-driven approaches to enhance models' signal-awareness: 1) we combine the SE concept of code complexity with the AI technique of curriculum learning; 2) we incorporate SE assistance into AI models by customizing Delta Debugging to generate simplified signal-preserving programs, augmenting them to the training dataset. With our techniques, we achieve up to 4.8x improvement in model signal awareness. Using the notion of code complexity, we further present a novel model learning introspection approach from the perspective of the dataset.

cs.SE↗